diff --git a/cts/cts-scheduler.in b/cts/cts-scheduler.in
index 8fa16fb69a..f2957bad66 100644
--- a/cts/cts-scheduler.in
+++ b/cts/cts-scheduler.in
@@ -1,1459 +1,1461 @@
#!@PYTHON@
""" Regression tests for Pacemaker's scheduler
"""
# Pacemaker targets compatibility with Python 2.7 and 3.2+
from __future__ import print_function, unicode_literals, absolute_import, division
-__copyright__ = "Copyright 2004-2019 the Pacemaker project contributors"
+__copyright__ = "Copyright 2004-2020 the Pacemaker project contributors"
__license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY"
import io
import os
import re
import sys
import stat
import shlex
import shutil
import argparse
import subprocess
import platform
DESC = """Regression tests for Pacemaker's scheduler"""
# Each entry in TESTS is a group of tests, where each test consists of a
# test base name, test description, and additional test arguments.
# Test groups will be separated by newlines in output.
TESTS = [
[
[ "simple1", "Offline" ],
[ "simple2", "Start" ],
[ "simple3", "Start 2" ],
[ "simple4", "Start Failed" ],
[ "simple6", "Stop Start" ],
[ "simple7", "Shutdown" ],
#[ "simple8", "Stonith" ],
#[ "simple9", "Lower version" ],
#[ "simple10", "Higher version" ],
[ "simple11", "Priority (ne)" ],
[ "simple12", "Priority (eq)" ],
[ "simple8", "Stickiness" ],
],
[
[ "group1", "Group" ],
[ "group2", "Group + Native" ],
[ "group3", "Group + Group" ],
[ "group4", "Group + Native (nothing)" ],
[ "group5", "Group + Native (move)" ],
[ "group6", "Group + Group (move)" ],
[ "group7", "Group colocation" ],
[ "group13", "Group colocation (cant run)" ],
[ "group8", "Group anti-colocation" ],
[ "group9", "Group recovery" ],
[ "group10", "Group partial recovery" ],
[ "group11", "Group target_role" ],
[ "group14", "Group stop (graph terminated)" ],
[ "group15", "Negative group colocation" ],
[ "bug-1573", "Partial stop of a group with two children" ],
[ "bug-1718", "Mandatory group ordering - Stop group_FUN" ],
[ "bug-lf-2613", "Move group on failure" ],
[ "bug-lf-2619", "Move group on clone failure" ],
[ "group-fail", "Ensure stop order is preserved for partially active groups" ],
[ "group-unmanaged", "No need to restart r115 because r114 is unmanaged" ],
[ "group-unmanaged-stopped", "Make sure r115 is stopped when r114 fails" ],
[ "group-dependents", "Account for the location preferences of things colocated with a group" ],
[ "group-stop-ordering", "Ensure blocked group member stop does not force other member stops" ],
],
[
[ "rsc_dep1", "Must not" ],
[ "rsc_dep3", "Must" ],
[ "rsc_dep5", "Must not 3" ],
[ "rsc_dep7", "Must 3" ],
[ "rsc_dep10", "Must (but cant)" ],
[ "rsc_dep2", "Must (running)" ],
[ "rsc_dep8", "Must (running : alt)" ],
[ "rsc_dep4", "Must (running + move)" ],
[ "asymmetric", "Asymmetric - require explicit location constraints" ],
],
[
[ "orphan-0", "Orphan ignore" ],
[ "orphan-1", "Orphan stop" ],
[ "orphan-2", "Orphan stop, remove failcount" ],
],
[
[ "params-0", "Params: No change" ],
[ "params-1", "Params: Changed" ],
[ "params-2", "Params: Resource definition" ],
[ "params-4", "Params: Reload" ],
[ "params-5", "Params: Restart based on probe digest" ],
[ "novell-251689", "Resource definition change + target_role=stopped" ],
[ "bug-lf-2106", "Restart all anonymous clone instances after config change" ],
[ "params-6", "Params: Detect reload in previously migrated resource" ],
[ "nvpair-id-ref", "Support id-ref in nvpair with optional name" ],
[ "not-reschedule-unneeded-monitor",
"Do not reschedule unneeded monitors while resource definitions have changed" ],
[ "reload-becomes-restart", "Cancel reload if restart becomes required" ],
],
[
[ "target-0", "Target Role : baseline" ],
[ "target-1", "Target Role : master" ],
[ "target-2", "Target Role : invalid" ],
],
[
[ "base-score", "Set a node's default score for all nodes" ],
],
[
[ "date-1", "Dates", [ "-t", "2005-020" ] ],
[ "date-2", "Date Spec - Pass", [ "-t", "2005-020T12:30" ] ],
[ "date-3", "Date Spec - Fail", [ "-t", "2005-020T11:30" ] ],
[ "origin", "Timing of recurring operations", [ "-t", "2014-05-07 00:28:00" ] ],
[ "probe-0", "Probe (anon clone)" ],
[ "probe-1", "Pending Probe" ],
[ "probe-2", "Correctly re-probe cloned groups" ],
[ "probe-3", "Probe (pending node)" ],
[ "probe-4", "Probe (pending node + stopped resource)" ],
[ "standby", "Standby" ],
[ "comments", "Comments" ],
],
[
[ "one-or-more-0", "Everything starts" ],
[ "one-or-more-1", "Nothing starts because of A" ],
[ "one-or-more-2", "D can start because of C" ],
[ "one-or-more-3", "D cannot start because of B and C" ],
[ "one-or-more-4", "D cannot start because of target-role" ],
[ "one-or-more-5", "Start A and F even though C and D are stopped" ],
[ "one-or-more-6", "Leave A running even though B is stopped" ],
[ "one-or-more-7", "Leave A running even though C is stopped" ],
[ "bug-5140-require-all-false", "Allow basegrp:0 to stop" ],
[ "clone-require-all-1", "clone B starts node 3 and 4" ],
[ "clone-require-all-2", "clone B remains stopped everywhere" ],
[ "clone-require-all-3", "clone B stops everywhere because A stops everywhere" ],
[ "clone-require-all-4", "clone B remains on node 3 and 4 with only one instance of A remaining" ],
[ "clone-require-all-5", "clone B starts on node 1 3 and 4" ],
[ "clone-require-all-6", "clone B remains active after shutting down instances of A" ],
[ "clone-require-all-7",
"clone A and B both start at the same time. all instances of A start before B" ],
[ "clone-require-all-no-interleave-1", "C starts everywhere after A and B" ],
[ "clone-require-all-no-interleave-2",
"C starts on nodes 1, 2, and 4 with only one active instance of B" ],
[ "clone-require-all-no-interleave-3",
"C remains active when instance of B is stopped on one node and started on another" ],
[ "one-or-more-unrunnable-instances", "Avoid dependencies on instances that won't ever be started" ],
],
[
[ "location-date-rules-1", "Use location constraints with ineffective date-based rules" ],
[ "location-date-rules-2", "Use location constraints with effective date-based rules" ],
[ "nvpair-date-rules-1", "Use nvpair blocks with a variety of date-based rules" ],
],
[
[ "order1", "Order start 1" ],
[ "order2", "Order start 2" ],
[ "order3", "Order stop" ],
[ "order4", "Order (multiple)" ],
[ "order5", "Order (move)" ],
[ "order6", "Order (move w/ restart)" ],
[ "order7", "Order (mandatory)" ],
[ "order-optional", "Order (score=0)" ],
[ "order-required", "Order (score=INFINITY)" ],
[ "bug-lf-2171", "Prevent group start when clone is stopped" ],
[ "order-clone", "Clone ordering should be able to prevent startup of dependent clones" ],
[ "order-sets", "Ordering for resource sets" ],
[ "order-serialize", "Serialize resources without inhibiting migration" ],
[ "order-serialize-set", "Serialize a set of resources without inhibiting migration" ],
[ "clone-order-primitive", "Order clone start after a primitive" ],
[ "clone-order-16instances", "Verify ordering of 16 cloned resources" ],
[ "order-optional-keyword", "Order (optional keyword)" ],
[ "order-mandatory", "Order (mandatory keyword)" ],
[ "bug-lf-2493",
"Don't imply colocation requirements when applying ordering constraints with clones" ],
[ "ordered-set-basic-startup", "Constraint set with default order settings" ],
[ "ordered-set-natural", "Allow natural set ordering" ],
[ "order-wrong-kind", "Order (error)" ],
],
[
[ "coloc-loop", "Colocation - loop" ],
[ "coloc-many-one", "Colocation - many-to-one" ],
[ "coloc-list", "Colocation - many-to-one with list" ],
[ "coloc-group", "Colocation - groups" ],
[ "coloc-slave-anti", "Anti-colocation with slave shouldn't prevent master colocation" ],
[ "coloc-attr", "Colocation based on node attributes" ],
[ "coloc-negative-group", "Negative colocation with a group" ],
[ "coloc-intra-set", "Intra-set colocation" ],
[ "bug-lf-2435", "Colocation sets with a negative score" ],
[ "coloc-clone-stays-active",
"Ensure clones don't get stopped/demoted because a dependent must stop" ],
[ "coloc_fp_logic", "Verify floating point calculations in colocation are working" ],
[ "colo_master_w_native",
"cl#5070 - Verify promotion order is affected when colocating master to native rsc" ],
[ "colo_slave_w_native",
"cl#5070 - Verify promotion order is affected when colocating slave to native rsc" ],
[ "anti-colocation-order",
"cl#5187 - Prevent resources in an anti-colocation from even temporarily running on a same node" ],
[ "anti-colocation-master", "Organize order of actions for master resources in anti-colocations" ],
[ "anti-colocation-slave", "Organize order of actions for slave resources in anti-colocations" ],
[ "enforce-colo1", "Always enforce B with A INFINITY" ],
[ "complex_enforce_colo", "Always enforce B with A INFINITY. (make sure heat-engine stops)" ],
],
[
[ "rsc-sets-seq-true", "Resource Sets - sequential=false" ],
[ "rsc-sets-seq-false", "Resource Sets - sequential=true" ],
[ "rsc-sets-clone", "Resource Sets - Clone" ],
[ "rsc-sets-master", "Resource Sets - Master" ],
[ "rsc-sets-clone-1", "Resource Sets - Clone (lf#2404)" ],
],
[
[ "attrs1", "string: eq (and)" ],
[ "attrs2", "string: lt / gt (and)" ],
[ "attrs3", "string: ne (or)" ],
[ "attrs4", "string: exists" ],
[ "attrs5", "string: not_exists" ],
[ "attrs6", "is_dc: true" ],
[ "attrs7", "is_dc: false" ],
[ "attrs8", "score_attribute" ],
[ "per-node-attrs", "Per node resource parameters" ],
],
[
[ "mon-rsc-1", "Schedule Monitor - start" ],
[ "mon-rsc-2", "Schedule Monitor - move" ],
[ "mon-rsc-3", "Schedule Monitor - pending start" ],
[ "mon-rsc-4", "Schedule Monitor - move/pending start" ],
],
[
[ "rec-rsc-0", "Resource Recover - no start" ],
[ "rec-rsc-1", "Resource Recover - start" ],
[ "rec-rsc-2", "Resource Recover - monitor" ],
[ "rec-rsc-3", "Resource Recover - stop - ignore" ],
[ "rec-rsc-4", "Resource Recover - stop - block" ],
[ "rec-rsc-5", "Resource Recover - stop - fence" ],
[ "rec-rsc-6", "Resource Recover - multiple - restart" ],
[ "rec-rsc-7", "Resource Recover - multiple - stop" ],
[ "rec-rsc-8", "Resource Recover - multiple - block" ],
[ "rec-rsc-9", "Resource Recover - group/group" ],
[ "monitor-recovery", "on-fail=block + resource recovery detected by recurring monitor" ],
[ "stop-failure-no-quorum", "Stop failure without quorum" ],
[ "stop-failure-no-fencing", "Stop failure without fencing available" ],
[ "stop-failure-with-fencing", "Stop failure with fencing available" ],
[ "multiple-active-block-group", "Support of multiple-active=block for resource groups" ],
[ "multiple-monitor-one-failed",
"Consider resource failed if any of the configured monitor operations failed" ],
],
[
[ "quorum-1", "No quorum - ignore" ],
[ "quorum-2", "No quorum - freeze" ],
[ "quorum-3", "No quorum - stop" ],
[ "quorum-4", "No quorum - start anyway" ],
[ "quorum-5", "No quorum - start anyway (group)" ],
[ "quorum-6", "No quorum - start anyway (clone)" ],
[ "bug-cl-5212", "No promotion with no-quorum-policy=freeze" ],
[ "suicide-needed-inquorate", "no-quorum-policy=suicide: suicide necessary" ],
[ "suicide-not-needed-initial-quorum",
"no-quorum-policy=suicide: suicide not necessary at initial quorum" ],
[ "suicide-not-needed-never-quorate",
"no-quorum-policy=suicide: suicide not necessary if never quorate" ],
[ "suicide-not-needed-quorate", "no-quorum-policy=suicide: suicide necessary if quorate" ],
],
[
[ "rec-node-1", "Node Recover - Startup - no fence" ],
[ "rec-node-2", "Node Recover - Startup - fence" ],
[ "rec-node-3", "Node Recover - HA down - no fence" ],
[ "rec-node-4", "Node Recover - HA down - fence" ],
[ "rec-node-5", "Node Recover - CRM down - no fence" ],
[ "rec-node-6", "Node Recover - CRM down - fence" ],
[ "rec-node-7", "Node Recover - no quorum - ignore" ],
[ "rec-node-8", "Node Recover - no quorum - freeze" ],
[ "rec-node-9", "Node Recover - no quorum - stop" ],
[ "rec-node-10", "Node Recover - no quorum - stop w/fence" ],
[ "rec-node-11", "Node Recover - CRM down w/ group - fence" ],
[ "rec-node-12", "Node Recover - nothing active - fence" ],
[ "rec-node-13", "Node Recover - failed resource + shutdown - fence" ],
[ "rec-node-15", "Node Recover - unknown lrm section" ],
[ "rec-node-14", "Serialize all stonith's" ],
],
[
[ "multi1", "Multiple Active (stop/start)" ],
],
[
[ "migrate-begin", "Normal migration" ],
[ "migrate-success", "Completed migration" ],
[ "migrate-partial-1", "Completed migration, missing stop on source" ],
[ "migrate-partial-2", "Successful migrate_to only" ],
[ "migrate-partial-3", "Successful migrate_to only, target down" ],
[ "migrate-partial-4", "Migrate from the correct host after migrate_to+migrate_from" ],
[ "bug-5186-partial-migrate", "Handle partial migration when src node loses membership" ],
[ "migrate-fail-2", "Failed migrate_from" ],
[ "migrate-fail-3", "Failed migrate_from + stop on source" ],
[ "migrate-fail-4",
"Failed migrate_from + stop on target - ideally we wouldn't need to re-stop on target" ],
[ "migrate-fail-5", "Failed migrate_from + stop on source and target" ],
[ "migrate-fail-6", "Failed migrate_to" ],
[ "migrate-fail-7", "Failed migrate_to + stop on source" ],
[ "migrate-fail-8",
"Failed migrate_to + stop on target - ideally we wouldn't need to re-stop on target" ],
[ "migrate-fail-9", "Failed migrate_to + stop on source and target" ],
[ "migration-ping-pong", "Old migrate_to failure + successful migrate_from on same node" ],
[ "migrate-stop", "Migration in a stopping stack" ],
[ "migrate-start", "Migration in a starting stack" ],
[ "migrate-stop_start", "Migration in a restarting stack" ],
[ "migrate-stop-complex", "Migration in a complex stopping stack" ],
[ "migrate-start-complex", "Migration in a complex starting stack" ],
[ "migrate-stop-start-complex", "Migration in a complex moving stack" ],
[ "migrate-shutdown", "Order the post-migration 'stop' before node shutdown" ],
[ "migrate-1", "Migrate (migrate)" ],
[ "migrate-2", "Migrate (stable)" ],
[ "migrate-3", "Migrate (failed migrate_to)" ],
[ "migrate-4", "Migrate (failed migrate_from)" ],
[ "novell-252693", "Migration in a stopping stack" ],
[ "novell-252693-2", "Migration in a starting stack" ],
[ "novell-252693-3", "Non-Migration in a starting and stopping stack" ],
[ "bug-1820", "Migration in a group" ],
[ "bug-1820-1", "Non-migration in a group" ],
[ "migrate-5", "Primitive migration with a clone" ],
[ "migrate-fencing", "Migration after Fencing" ],
[ "migrate-both-vms", "Migrate two VMs that have no colocation" ],
[ "migration-behind-migrating-remote", "Migrate resource behind migrating remote connection" ],
[ "1-a-then-bm-move-b", "Advanced migrate logic. A then B. migrate B" ],
[ "2-am-then-b-move-a", "Advanced migrate logic, A then B, migrate A without stopping B" ],
[ "3-am-then-bm-both-migrate", "Advanced migrate logic. A then B. migrate both" ],
[ "4-am-then-bm-b-not-migratable", "Advanced migrate logic, A then B, B not migratable" ],
[ "5-am-then-bm-a-not-migratable", "Advanced migrate logic. A then B. move both, a not migratable" ],
[ "6-migrate-group", "Advanced migrate logic, migrate a group" ],
[ "7-migrate-group-one-unmigratable",
"Advanced migrate logic, migrate group mixed with allow-migrate true/false" ],
[ "8-am-then-bm-a-migrating-b-stopping",
"Advanced migrate logic, A then B, A migrating, B stopping" ],
[ "9-am-then-bm-b-migrating-a-stopping",
"Advanced migrate logic, A then B, B migrate, A stopping" ],
[ "10-a-then-bm-b-move-a-clone",
"Advanced migrate logic, A clone then B, migrate B while stopping A" ],
[ "11-a-then-bm-b-move-a-clone-starting",
"Advanced migrate logic, A clone then B, B moving while A is start/stopping" ],
[ "a-promote-then-b-migrate", "A promote then B start. migrate B" ],
[ "a-demote-then-b-migrate", "A demote then B stop. migrate B" ],
# @TODO: If pacemaker implements versioned attributes, uncomment this test
#[ "migrate-versioned", "Disable migration for versioned resources" ],
[ "bug-lf-2422", "Dependency on partially active group - stop ocfs:*" ],
],
[
[ "clone-anon-probe-1", "Probe the correct (anonymous) clone instance for each node" ],
[ "clone-anon-probe-2", "Avoid needless re-probing of anonymous clones" ],
[ "clone-anon-failcount", "Merge failcounts for anonymous clones" ],
[ "force-anon-clone-max", "Update clone-max properly when forcing a clone to be anonymous" ],
[ "anon-instance-pending", "Assign anonymous clone instance numbers properly when action pending" ],
[ "inc0", "Incarnation start" ],
[ "inc1", "Incarnation start order" ],
[ "inc2", "Incarnation silent restart, stop, move" ],
[ "inc3", "Inter-incarnation ordering, silent restart, stop, move" ],
[ "inc4", "Inter-incarnation ordering, silent restart, stop, move (ordered)" ],
[ "inc5", "Inter-incarnation ordering, silent restart, stop, move (restart 1)" ],
[ "inc6", "Inter-incarnation ordering, silent restart, stop, move (restart 2)" ],
[ "inc7", "Clone colocation" ],
[ "inc8", "Clone anti-colocation" ],
[ "inc9", "Non-unique clone" ],
[ "inc10", "Non-unique clone (stop)" ],
[ "inc11", "Primitive colocation with clones" ],
[ "inc12", "Clone shutdown" ],
[ "cloned-group", "Make sure only the correct number of cloned groups are started" ],
[ "cloned-group-stop", "Ensure stopping qpidd also stops glance and cinder" ],
[ "clone-no-shuffle", "Don't prioritize allocation of instances that must be moved" ],
[ "clone-max-zero", "Orphan processing with clone-max=0" ],
[ "clone-anon-dup",
"Bug LF#2087 - Correctly parse the state of anonymous clones that are active more than once per node" ],
[ "bug-lf-2160", "Don't shuffle clones due to colocation" ],
[ "bug-lf-2213", "clone-node-max enforcement for cloned groups" ],
[ "bug-lf-2153", "Clone ordering constraints" ],
[ "bug-lf-2361", "Ensure clones observe mandatory ordering constraints if the LHS is unrunnable" ],
[ "bug-lf-2317", "Avoid needless restart of primitive depending on a clone" ],
[ "clone-colocate-instance-1", "Colocation with a specific clone instance (negative example)" ],
[ "clone-colocate-instance-2", "Colocation with a specific clone instance" ],
[ "clone-order-instance", "Ordering with specific clone instances" ],
[ "bug-lf-2453", "Enforce mandatory clone ordering without colocation" ],
[ "bug-lf-2508", "Correctly reconstruct the status of anonymous cloned groups" ],
[ "bug-lf-2544", "Balanced clone placement" ],
[ "bug-lf-2445", "Redistribute clones with node-max > 1 and stickiness = 0" ],
[ "bug-lf-2574", "Avoid clone shuffle" ],
[ "bug-lf-2581", "Avoid group restart due to unrelated clone (re)start" ],
[ "bug-cl-5168", "Don't shuffle clones" ],
[ "bug-cl-5170", "Prevent clone from starting with on-fail=block" ],
[ "clone-fail-block-colocation", "Move colocated group when failed clone has on-fail=block" ],
[ "clone-interleave-1",
"Clone-3 cannot start on pcmk-1 due to interleaved ordering (no colocation)" ],
[ "clone-interleave-2", "Clone-3 must stop on pcmk-1 due to interleaved ordering (no colocation)" ],
[ "clone-interleave-3",
"Clone-3 must be recovered on pcmk-1 due to interleaved ordering (no colocation)" ],
[ "rebalance-unique-clones", "Rebalance unique clone instances with no stickiness" ],
[ "clone-requires-quorum-recovery", "Clone with requires=quorum on failed node needing recovery" ],
[ "clone-requires-quorum",
"Clone with requires=quorum with presumed-inactive instance on failed node" ],
],
[
[ "cloned_start_one", "order first clone then clone... first clone_min=2" ],
[ "cloned_start_two", "order first clone then clone... first clone_min=2" ],
[ "cloned_stop_one", "order first clone then clone... first clone_min=2" ],
[ "cloned_stop_two", "order first clone then clone... first clone_min=2" ],
[ "clone_min_interleave_start_one",
"order first clone then clone... first clone_min=2 and then has interleave=true" ],
[ "clone_min_interleave_start_two",
"order first clone then clone... first clone_min=2 and then has interleave=true" ],
[ "clone_min_interleave_stop_one",
"order first clone then clone... first clone_min=2 and then has interleave=true" ],
[ "clone_min_interleave_stop_two",
"order first clone then clone... first clone_min=2 and then has interleave=true" ],
[ "clone_min_start_one", "order first clone then primitive... first clone_min=2" ],
[ "clone_min_start_two", "order first clone then primitive... first clone_min=2" ],
[ "clone_min_stop_all", "order first clone then primitive... first clone_min=2" ],
[ "clone_min_stop_one", "order first clone then primitive... first clone_min=2" ],
[ "clone_min_stop_two", "order first clone then primitive... first clone_min=2" ],
],
[
[ "unfence-startup", "Clean unfencing" ],
[ "unfence-definition", "Unfencing when the agent changes" ],
[ "unfence-parameters", "Unfencing when the agent parameters changes" ],
[ "unfence-device", "Unfencing when a cluster has only fence devices" ],
],
[
[ "master-0", "Stopped -> Slave" ],
[ "master-1", "Stopped -> Promote" ],
[ "master-2", "Stopped -> Promote : notify" ],
[ "master-3", "Stopped -> Promote : master location" ],
[ "master-4", "Started -> Promote : master location" ],
[ "master-5", "Promoted -> Promoted" ],
[ "master-6", "Promoted -> Promoted (2)" ],
[ "master-7", "Promoted -> Fenced" ],
[ "master-8", "Promoted -> Fenced -> Moved" ],
[ "master-9", "Stopped + Promotable + No quorum" ],
[ "master-10", "Stopped -> Promotable : notify with monitor" ],
[ "master-11", "Stopped -> Promote : colocation" ],
[ "novell-239082", "Demote/Promote ordering" ],
[ "novell-239087", "Stable master placement" ],
[ "master-12", "Promotion based solely on rsc_location constraints" ],
[ "master-13", "Include preferences of colocated resources when placing master" ],
[ "master-demote", "Ordering when actions depends on demoting a slave resource" ],
[ "master-ordering", "Prevent resources from starting that need a master" ],
[ "bug-1765", "Master-Master Colocation (do not stop the slaves)" ],
[ "master-group", "Promotion of cloned groups" ],
[ "bug-lf-1852", "Don't shuffle master/slave instances unnecessarily" ],
[ "master-failed-demote", "Don't retry failed demote actions" ],
[ "master-failed-demote-2", "Don't retry failed demote actions (notify=false)" ],
[ "master-depend",
"Ensure resources that depend on the master don't get allocated until the master does" ],
[ "master-reattach", "Re-attach to a running master" ],
[ "master-allow-start", "Don't include master score if it would prevent allocation" ],
[ "master-colocation",
"Allow master instances placemaker to be influenced by colocation constraints" ],
[ "master-pseudo", "Make sure promote/demote pseudo actions are created correctly" ],
[ "master-role", "Prevent target-role from promoting more than master-max instances" ],
[ "bug-lf-2358", "Master-Master anti-colocation" ],
[ "master-promotion-constraint", "Mandatory master colocation constraints" ],
[ "unmanaged-master", "Ensure role is preserved for unmanaged resources" ],
[ "master-unmanaged-monitor", "Start the correct monitor operation for unmanaged masters" ],
[ "master-demote-2", "Demote does not clear past failure" ],
[ "master-move", "Move master based on failure of colocated group" ],
[ "master-probed-score", "Observe the promotion score of probed resources" ],
[ "colocation_constraint_stops_master",
"cl#5054 - Ensure master is demoted when stopped by colocation constraint" ],
[ "colocation_constraint_stops_slave",
"cl#5054 - Ensure slave is not demoted when stopped by colocation constraint" ],
[ "order_constraint_stops_master",
"cl#5054 - Ensure master is demoted when stopped by order constraint" ],
[ "order_constraint_stops_slave",
"cl#5054 - Ensure slave is not demoted when stopped by order constraint" ],
[ "master_monitor_restart", "cl#5072 - Ensure master monitor operation will start after promotion" ],
[ "bug-rh-880249", "Handle replacement of an m/s resource with a primitive" ],
[ "bug-5143-ms-shuffle", "Prevent master shuffling due to promotion score" ],
[ "master-demote-block", "Block promotion if demote fails with on-fail=block" ],
[ "master-dependent-ban",
"Don't stop instances from being active because a dependent is banned from that host" ],
[ "master-stop", "Stop instances due to location constraint with role=Started" ],
[ "master-partially-demoted-group", "Allow partially demoted group to finish demoting" ],
[ "bug-cl-5213", "Ensure role colocation with -INFINITY is enforced" ],
[ "bug-cl-5219", "Allow unrelated resources with a common colocation target to remain promoted" ],
[ "master-asymmetrical-order",
"Fix the behaviors of multi-state resources with asymmetrical ordering" ],
[ "master-notify", "Master promotion with notifies" ],
[ "master-score-startup", "Use permanent master scores without LRM history" ],
[ "failed-demote-recovery", "Recover resource in slave role after demote fails" ],
[ "failed-demote-recovery-master", "Recover resource in master role after demote fails" ],
],
[
[ "history-1", "Correctly parse stateful-1 resource state" ],
],
[
[ "managed-0", "Managed (reference)" ],
[ "managed-1", "Not managed - down" ],
[ "managed-2", "Not managed - up" ],
[ "bug-5028", "Shutdown should block if anything depends on an unmanaged resource" ],
[ "bug-5028-detach", "Ensure detach still works" ],
[ "bug-5028-bottom",
"Ensure shutdown still blocks if the blocked resource is at the bottom of the stack" ],
[ "unmanaged-stop-1",
"cl#5155 - Block the stop of resources if any depending resource is unmanaged" ],
[ "unmanaged-stop-2",
"cl#5155 - Block the stop of resources if the first resource in a mandatory stop order is unmanaged" ],
[ "unmanaged-stop-3",
"cl#5155 - Block the stop of resources if any depending resource in a group is unmanaged" ],
[ "unmanaged-stop-4",
"cl#5155 - Block the stop of resources if any depending resource in the middle of a group is unmanaged" ],
[ "unmanaged-block-restart",
"Block restart of resources if any dependent resource in a group is unmanaged" ],
],
[
[ "interleave-0", "Interleave (reference)" ],
[ "interleave-1", "coloc - not interleaved" ],
[ "interleave-2", "coloc - interleaved" ],
[ "interleave-3", "coloc - interleaved (2)" ],
[ "interleave-pseudo-stop", "Interleaved clone during stonith" ],
[ "interleave-stop", "Interleaved clone during stop" ],
[ "interleave-restart", "Interleaved clone during dependency restart" ],
],
[
[ "notify-0", "Notify reference" ],
[ "notify-1", "Notify simple" ],
[ "notify-2", "Notify simple, confirm" ],
[ "notify-3", "Notify move, confirm" ],
[ "novell-239079", "Notification priority" ],
#[ "notify-2", "Notify - 764" ],
[ "notifs-for-unrunnable", "Don't schedule notifications for an unrunnable action" ],
[ "route-remote-notify", "Route remote notify actions through correct cluster node" ],
[ "notify-behind-stopping-remote", "Don't schedule notifications behind stopped remote" ],
],
[
[ "594", "OSDL #594 - Unrunnable actions scheduled in transition" ],
[ "662", "OSDL #662 - Two resources start on one node when incarnation_node_max = 1" ],
[ "696", "OSDL #696 - CRM starts stonith RA without monitor" ],
[ "726", "OSDL #726 - Attempting to schedule rsc_posic041_monitor_5000 _after_ a stop" ],
[ "735", "OSDL #735 - Correctly detect that rsc_hadev1 is stopped on hadev3" ],
[ "764", "OSDL #764 - Missing monitor op for DoFencing:child_DoFencing:1" ],
[ "797", "OSDL #797 - Assert triggered: task_id_i > max_call_id" ],
[ "829", "OSDL #829" ],
[ "994",
"OSDL #994 - Stopping the last resource in a resource group causes the entire group to be restarted" ],
[ "994-2", "OSDL #994 - with a dependent resource" ],
[ "1360", "OSDL #1360 - Clone stickiness" ],
[ "1484", "OSDL #1484 - on_fail=stop" ],
[ "1494", "OSDL #1494 - Clone stability" ],
[ "unrunnable-1", "Unrunnable" ],
[ "unrunnable-2", "Unrunnable 2" ],
[ "stonith-0", "Stonith loop - 1" ],
[ "stonith-1", "Stonith loop - 2" ],
[ "stonith-2", "Stonith loop - 3" ],
[ "stonith-3", "Stonith startup" ],
[ "stonith-4", "Stonith node state" ],
[ "dc-fence-ordering", "DC needs fencing while other nodes are shutting down" ],
[ "bug-1572-1", "Recovery of groups depending on master/slave" ],
[ "bug-1572-2", "Recovery of groups depending on master/slave when the master is never re-promoted" ],
[ "bug-1685", "Depends-on-master ordering" ],
[ "bug-1822", "Don't promote partially active groups" ],
[ "bug-pm-11", "New resource added to a m/s group" ],
[ "bug-pm-12", "Recover only the failed portion of a cloned group" ],
[ "bug-n-387749", "Don't shuffle clone instances" ],
[ "bug-n-385265",
"Don't ignore the failure stickiness of group children - resource_idvscommon should stay stopped" ],
[ "bug-n-385265-2",
"Ensure groups are migrated instead of remaining partially active on the current node" ],
[ "bug-lf-1920", "Correctly handle probes that find active resources" ],
[ "bnc-515172", "Location constraint with multiple expressions" ],
[ "colocate-primitive-with-clone", "Optional colocation with a clone" ],
[ "use-after-free-merge", "Use-after-free in native_merge_weights" ],
[ "bug-lf-2551", "STONITH ordering for stop" ],
[ "bug-lf-2606", "Stonith implies demote" ],
[ "bug-lf-2474", "Ensure resource op timeout takes precedence over op_defaults" ],
[ "bug-suse-707150", "Prevent vm-01 from starting due to colocation/ordering" ],
[ "bug-5014-A-start-B-start", "Verify when A starts B starts using symmetrical=false" ],
[ "bug-5014-A-stop-B-started",
"Verify when A stops B does not stop if it has already started using symmetric=false" ],
[ "bug-5014-A-stopped-B-stopped",
"Verify when A is stopped and B has not started, B does not start before A using symmetric=false" ],
[ "bug-5014-CthenAthenB-C-stopped",
"Verify when C then A is symmetrical=true, A then B is symmetric=false, and C is stopped that nothing starts" ],
[ "bug-5014-CLONE-A-start-B-start",
"Verify when A starts B starts using clone resources with symmetric=false" ],
[ "bug-5014-CLONE-A-stop-B-started",
"Verify when A stops B does not stop if it has already started using clone resources with symmetric=false" ],
[ "bug-5014-GROUP-A-start-B-start",
"Verify when A starts B starts when using group resources with symmetric=false" ],
[ "bug-5014-GROUP-A-stopped-B-started",
"Verify when A stops B does not stop if it has already started using group resources with symmetric=false" ],
[ "bug-5014-GROUP-A-stopped-B-stopped",
"Verify when A is stopped and B has not started, B does not start before A using group resources with symmetric=false" ],
[ "bug-5014-ordered-set-symmetrical-false",
"Verify ordered sets work with symmetrical=false" ],
[ "bug-5014-ordered-set-symmetrical-true",
"Verify ordered sets work with symmetrical=true" ],
[ "bug-5007-masterslave_colocation",
"Verify use of colocation scores other than INFINITY and -INFINITY work on multi-state resources" ],
[ "bug-5038", "Prevent restart of anonymous clones when clone-max decreases" ],
[ "bug-5025-1", "Automatically clean up failcount after resource config change with reload" ],
[ "bug-5025-2", "Make sure clear failcount action isn't set when config does not change" ],
[ "bug-5025-3", "Automatically clean up failcount after resource config change with restart" ],
[ "bug-5025-4", "Clear failcount when last failure is a start op and rsc attributes changed" ],
[ "failcount", "Ensure failcounts are correctly expired" ],
[ "failcount-block", "Ensure failcounts are not expired when on-fail=block is present" ],
[ "per-op-failcount", "Ensure per-operation failcount is handled and not passed to fence agent" ],
[ "on-fail-ignore", "Ensure on-fail=ignore works even beyond migration-threshold" ],
[ "monitor-onfail-restart", "bug-5058 - Monitor failure with on-fail set to restart" ],
[ "monitor-onfail-stop", "bug-5058 - Monitor failure wiht on-fail set to stop" ],
[ "bug-5059", "No need to restart p_stateful1:*" ],
[ "bug-5069-op-enabled", "Test on-fail=ignore with failure when monitor is enabled" ],
[ "bug-5069-op-disabled", "Test on-fail-ignore with failure when monitor is disabled" ],
[ "obsolete-lrm-resource", "cl#5115 - Do not use obsolete lrm_resource sections" ],
[ "expire-non-blocked-failure",
"Ignore failure-timeout only if the failed operation has on-fail=block" ],
[ "asymmetrical-order-move", "Respect asymmetrical ordering when trying to move resources" ],
[ "asymmetrical-order-restart", "Respect asymmetrical ordering when restarting dependent resource" ],
[ "start-then-stop-with-unfence", "Avoid graph loop with start-then-stop constraint plus unfencing" ],
[ "order-expired-failure", "Order failcount cleanup after remote fencing" ],
[ "ignore_stonith_rsc_order1",
"cl#5056- Ignore order constraint between stonith and non-stonith rsc" ],
[ "ignore_stonith_rsc_order2",
"cl#5056- Ignore order constraint with group rsc containing mixed stonith and non-stonith" ],
[ "ignore_stonith_rsc_order3", "cl#5056- Ignore order constraint, stonith clone and mixed group" ],
[ "ignore_stonith_rsc_order4",
"cl#5056- Ignore order constraint, stonith clone and clone with nested mixed group" ],
[ "honor_stonith_rsc_order1",
"cl#5056- Honor order constraint, stonith clone and pure stonith group(single rsc)" ],
[ "honor_stonith_rsc_order2",
"cl#5056- Honor order constraint, stonith clone and pure stonith group(multiple rsc)" ],
[ "honor_stonith_rsc_order3",
"cl#5056- Honor order constraint, stonith clones with nested pure stonith group" ],
[ "honor_stonith_rsc_order4",
"cl#5056- Honor order constraint, between two native stonith rscs" ],
[ "multiply-active-stonith", "Multiply active stonith" ],
[ "probe-timeout", "cl#5099 - Default probe timeout" ],
[ "order-first-probes",
"cl#5301 - respect order constraints when relevant resources are being probed" ],
[ "concurrent-fencing", "Allow performing fencing operations in parallel" ],
],
[
[ "systemhealth1", "System Health () #1" ],
[ "systemhealth2", "System Health () #2" ],
[ "systemhealth3", "System Health () #3" ],
[ "systemhealthn1", "System Health (None) #1" ],
[ "systemhealthn2", "System Health (None) #2" ],
[ "systemhealthn3", "System Health (None) #3" ],
[ "systemhealthm1", "System Health (Migrate On Red) #1" ],
[ "systemhealthm2", "System Health (Migrate On Red) #2" ],
[ "systemhealthm3", "System Health (Migrate On Red) #3" ],
[ "systemhealtho1", "System Health (Only Green) #1" ],
[ "systemhealtho2", "System Health (Only Green) #2" ],
[ "systemhealtho3", "System Health (Only Green) #3" ],
[ "systemhealthp1", "System Health (Progessive) #1" ],
[ "systemhealthp2", "System Health (Progessive) #2" ],
[ "systemhealthp3", "System Health (Progessive) #3" ],
],
[
[ "utilization", "Placement Strategy - utilization" ],
[ "minimal", "Placement Strategy - minimal" ],
[ "balanced", "Placement Strategy - balanced" ],
],
[
[ "placement-stickiness", "Optimized Placement Strategy - stickiness" ],
[ "placement-priority", "Optimized Placement Strategy - priority" ],
[ "placement-location", "Optimized Placement Strategy - location" ],
[ "placement-capacity", "Optimized Placement Strategy - capacity" ],
],
[
[ "utilization-order1", "Utilization Order - Simple" ],
[ "utilization-order2", "Utilization Order - Complex" ],
[ "utilization-order3", "Utilization Order - Migrate" ],
[ "utilization-order4", "Utilization Order - Live Migration (bnc#695440)" ],
[ "utilization-shuffle",
"Don't displace prmExPostgreSQLDB2 on act2, Start prmExPostgreSQLDB1 on act3" ],
[ "load-stopped-loop", "Avoid transition loop due to load_stopped (cl#5044)" ],
[ "load-stopped-loop-2",
"cl#5235 - Prevent graph loops that can be introduced by load_stopped -> migrate_to ordering" ],
],
[
[ "colocated-utilization-primitive-1", "Colocated Utilization - Primitive" ],
[ "colocated-utilization-primitive-2", "Colocated Utilization - Choose the most capable node" ],
[ "colocated-utilization-group", "Colocated Utilization - Group" ],
[ "colocated-utilization-clone", "Colocated Utilization - Clone" ],
[ "utilization-check-allowed-nodes",
"Only check the capacities of the nodes that can run the resource" ],
],
[
[ "reprobe-target_rc", "Ensure correct target_rc for reprobe of inactive resources" ],
[ "node-maintenance-1", "cl#5128 - Node maintenance" ],
[ "node-maintenance-2", "cl#5128 - Node maintenance (coming out of maintenance mode)" ],
[ "shutdown-maintenance-node", "Do not fence a maintenance node if it shuts down cleanly" ],
[ "rsc-maintenance", "Per-resource maintenance" ],
],
[
[ "not-installed-agent", "The resource agent is missing" ],
[ "not-installed-tools", "Something the resource agent needs is missing" ],
],
[
[ "stopped-monitor-00", "Stopped Monitor - initial start" ],
[ "stopped-monitor-01", "Stopped Monitor - failed started" ],
[ "stopped-monitor-02", "Stopped Monitor - started multi-up" ],
[ "stopped-monitor-03", "Stopped Monitor - stop started" ],
[ "stopped-monitor-04", "Stopped Monitor - failed stop" ],
[ "stopped-monitor-05", "Stopped Monitor - start unmanaged" ],
[ "stopped-monitor-06", "Stopped Monitor - unmanaged multi-up" ],
[ "stopped-monitor-07", "Stopped Monitor - start unmanaged multi-up" ],
[ "stopped-monitor-08", "Stopped Monitor - migrate" ],
[ "stopped-monitor-09", "Stopped Monitor - unmanage started" ],
[ "stopped-monitor-10", "Stopped Monitor - unmanaged started multi-up" ],
[ "stopped-monitor-11", "Stopped Monitor - stop unmanaged started" ],
[ "stopped-monitor-12", "Stopped Monitor - unmanaged started multi-up (target-role=Stopped)" ],
[ "stopped-monitor-20", "Stopped Monitor - initial stop" ],
[ "stopped-monitor-21", "Stopped Monitor - stopped single-up" ],
[ "stopped-monitor-22", "Stopped Monitor - stopped multi-up" ],
[ "stopped-monitor-23", "Stopped Monitor - start stopped" ],
[ "stopped-monitor-24", "Stopped Monitor - unmanage stopped" ],
[ "stopped-monitor-25", "Stopped Monitor - unmanaged stopped multi-up" ],
[ "stopped-monitor-26", "Stopped Monitor - start unmanaged stopped" ],
[ "stopped-monitor-27", "Stopped Monitor - unmanaged stopped multi-up (target-role=Started)" ],
[ "stopped-monitor-30", "Stopped Monitor - new node started" ],
[ "stopped-monitor-31", "Stopped Monitor - new node stopped" ],
],
[
# This is a combo test to check:
# - probe timeout defaults to the minimum-interval monitor's
# - duplicate recurring operations are ignored
# - if timeout spec is bad, the default timeout is used
# - failure is blocked with on-fail=block even if ISO8601 interval is specified
# - started/stopped role monitors are started/stopped on right nodes
[ "intervals", "Recurring monitor interval handling" ],
],
[
[ "ticket-primitive-1", "Ticket - Primitive (loss-policy=stop, initial)" ],
[ "ticket-primitive-2", "Ticket - Primitive (loss-policy=stop, granted)" ],
[ "ticket-primitive-3", "Ticket - Primitive (loss-policy-stop, revoked)" ],
[ "ticket-primitive-4", "Ticket - Primitive (loss-policy=demote, initial)" ],
[ "ticket-primitive-5", "Ticket - Primitive (loss-policy=demote, granted)" ],
[ "ticket-primitive-6", "Ticket - Primitive (loss-policy=demote, revoked)" ],
[ "ticket-primitive-7", "Ticket - Primitive (loss-policy=fence, initial)" ],
[ "ticket-primitive-8", "Ticket - Primitive (loss-policy=fence, granted)" ],
[ "ticket-primitive-9", "Ticket - Primitive (loss-policy=fence, revoked)" ],
[ "ticket-primitive-10", "Ticket - Primitive (loss-policy=freeze, initial)" ],
[ "ticket-primitive-11", "Ticket - Primitive (loss-policy=freeze, granted)" ],
[ "ticket-primitive-12", "Ticket - Primitive (loss-policy=freeze, revoked)" ],
[ "ticket-primitive-13", "Ticket - Primitive (loss-policy=stop, standby, granted)" ],
[ "ticket-primitive-14", "Ticket - Primitive (loss-policy=stop, granted, standby)" ],
[ "ticket-primitive-15", "Ticket - Primitive (loss-policy=stop, standby, revoked)" ],
[ "ticket-primitive-16", "Ticket - Primitive (loss-policy=demote, standby, granted)" ],
[ "ticket-primitive-17", "Ticket - Primitive (loss-policy=demote, granted, standby)" ],
[ "ticket-primitive-18", "Ticket - Primitive (loss-policy=demote, standby, revoked)" ],
[ "ticket-primitive-19", "Ticket - Primitive (loss-policy=fence, standby, granted)" ],
[ "ticket-primitive-20", "Ticket - Primitive (loss-policy=fence, granted, standby)" ],
[ "ticket-primitive-21", "Ticket - Primitive (loss-policy=fence, standby, revoked)" ],
[ "ticket-primitive-22", "Ticket - Primitive (loss-policy=freeze, standby, granted)" ],
[ "ticket-primitive-23", "Ticket - Primitive (loss-policy=freeze, granted, standby)" ],
[ "ticket-primitive-24", "Ticket - Primitive (loss-policy=freeze, standby, revoked)" ],
],
[
[ "ticket-group-1", "Ticket - Group (loss-policy=stop, initial)" ],
[ "ticket-group-2", "Ticket - Group (loss-policy=stop, granted)" ],
[ "ticket-group-3", "Ticket - Group (loss-policy-stop, revoked)" ],
[ "ticket-group-4", "Ticket - Group (loss-policy=demote, initial)" ],
[ "ticket-group-5", "Ticket - Group (loss-policy=demote, granted)" ],
[ "ticket-group-6", "Ticket - Group (loss-policy=demote, revoked)" ],
[ "ticket-group-7", "Ticket - Group (loss-policy=fence, initial)" ],
[ "ticket-group-8", "Ticket - Group (loss-policy=fence, granted)" ],
[ "ticket-group-9", "Ticket - Group (loss-policy=fence, revoked)" ],
[ "ticket-group-10", "Ticket - Group (loss-policy=freeze, initial)" ],
[ "ticket-group-11", "Ticket - Group (loss-policy=freeze, granted)" ],
[ "ticket-group-12", "Ticket - Group (loss-policy=freeze, revoked)" ],
[ "ticket-group-13", "Ticket - Group (loss-policy=stop, standby, granted)" ],
[ "ticket-group-14", "Ticket - Group (loss-policy=stop, granted, standby)" ],
[ "ticket-group-15", "Ticket - Group (loss-policy=stop, standby, revoked)" ],
[ "ticket-group-16", "Ticket - Group (loss-policy=demote, standby, granted)" ],
[ "ticket-group-17", "Ticket - Group (loss-policy=demote, granted, standby)" ],
[ "ticket-group-18", "Ticket - Group (loss-policy=demote, standby, revoked)" ],
[ "ticket-group-19", "Ticket - Group (loss-policy=fence, standby, granted)" ],
[ "ticket-group-20", "Ticket - Group (loss-policy=fence, granted, standby)" ],
[ "ticket-group-21", "Ticket - Group (loss-policy=fence, standby, revoked)" ],
[ "ticket-group-22", "Ticket - Group (loss-policy=freeze, standby, granted)" ],
[ "ticket-group-23", "Ticket - Group (loss-policy=freeze, granted, standby)" ],
[ "ticket-group-24", "Ticket - Group (loss-policy=freeze, standby, revoked)" ],
],
[
[ "ticket-clone-1", "Ticket - Clone (loss-policy=stop, initial)" ],
[ "ticket-clone-2", "Ticket - Clone (loss-policy=stop, granted)" ],
[ "ticket-clone-3", "Ticket - Clone (loss-policy-stop, revoked)" ],
[ "ticket-clone-4", "Ticket - Clone (loss-policy=demote, initial)" ],
[ "ticket-clone-5", "Ticket - Clone (loss-policy=demote, granted)" ],
[ "ticket-clone-6", "Ticket - Clone (loss-policy=demote, revoked)" ],
[ "ticket-clone-7", "Ticket - Clone (loss-policy=fence, initial)" ],
[ "ticket-clone-8", "Ticket - Clone (loss-policy=fence, granted)" ],
[ "ticket-clone-9", "Ticket - Clone (loss-policy=fence, revoked)" ],
[ "ticket-clone-10", "Ticket - Clone (loss-policy=freeze, initial)" ],
[ "ticket-clone-11", "Ticket - Clone (loss-policy=freeze, granted)" ],
[ "ticket-clone-12", "Ticket - Clone (loss-policy=freeze, revoked)" ],
[ "ticket-clone-13", "Ticket - Clone (loss-policy=stop, standby, granted)" ],
[ "ticket-clone-14", "Ticket - Clone (loss-policy=stop, granted, standby)" ],
[ "ticket-clone-15", "Ticket - Clone (loss-policy=stop, standby, revoked)" ],
[ "ticket-clone-16", "Ticket - Clone (loss-policy=demote, standby, granted)" ],
[ "ticket-clone-17", "Ticket - Clone (loss-policy=demote, granted, standby)" ],
[ "ticket-clone-18", "Ticket - Clone (loss-policy=demote, standby, revoked)" ],
[ "ticket-clone-19", "Ticket - Clone (loss-policy=fence, standby, granted)" ],
[ "ticket-clone-20", "Ticket - Clone (loss-policy=fence, granted, standby)" ],
[ "ticket-clone-21", "Ticket - Clone (loss-policy=fence, standby, revoked)" ],
[ "ticket-clone-22", "Ticket - Clone (loss-policy=freeze, standby, granted)" ],
[ "ticket-clone-23", "Ticket - Clone (loss-policy=freeze, granted, standby)" ],
[ "ticket-clone-24", "Ticket - Clone (loss-policy=freeze, standby, revoked)" ],
],
[
[ "ticket-master-1", "Ticket - Master (loss-policy=stop, initial)" ],
[ "ticket-master-2", "Ticket - Master (loss-policy=stop, granted)" ],
[ "ticket-master-3", "Ticket - Master (loss-policy-stop, revoked)" ],
[ "ticket-master-4", "Ticket - Master (loss-policy=demote, initial)" ],
[ "ticket-master-5", "Ticket - Master (loss-policy=demote, granted)" ],
[ "ticket-master-6", "Ticket - Master (loss-policy=demote, revoked)" ],
[ "ticket-master-7", "Ticket - Master (loss-policy=fence, initial)" ],
[ "ticket-master-8", "Ticket - Master (loss-policy=fence, granted)" ],
[ "ticket-master-9", "Ticket - Master (loss-policy=fence, revoked)" ],
[ "ticket-master-10", "Ticket - Master (loss-policy=freeze, initial)" ],
[ "ticket-master-11", "Ticket - Master (loss-policy=freeze, granted)" ],
[ "ticket-master-12", "Ticket - Master (loss-policy=freeze, revoked)" ],
[ "ticket-master-13", "Ticket - Master (loss-policy=stop, standby, granted)" ],
[ "ticket-master-14", "Ticket - Master (loss-policy=stop, granted, standby)" ],
[ "ticket-master-15", "Ticket - Master (loss-policy=stop, standby, revoked)" ],
[ "ticket-master-16", "Ticket - Master (loss-policy=demote, standby, granted)" ],
[ "ticket-master-17", "Ticket - Master (loss-policy=demote, granted, standby)" ],
[ "ticket-master-18", "Ticket - Master (loss-policy=demote, standby, revoked)" ],
[ "ticket-master-19", "Ticket - Master (loss-policy=fence, standby, granted)" ],
[ "ticket-master-20", "Ticket - Master (loss-policy=fence, granted, standby)" ],
[ "ticket-master-21", "Ticket - Master (loss-policy=fence, standby, revoked)" ],
[ "ticket-master-22", "Ticket - Master (loss-policy=freeze, standby, granted)" ],
[ "ticket-master-23", "Ticket - Master (loss-policy=freeze, granted, standby)" ],
[ "ticket-master-24", "Ticket - Master (loss-policy=freeze, standby, revoked)" ],
],
[
[ "ticket-rsc-sets-1", "Ticket - Resource sets (1 ticket, initial)" ],
[ "ticket-rsc-sets-2", "Ticket - Resource sets (1 ticket, granted)" ],
[ "ticket-rsc-sets-3", "Ticket - Resource sets (1 ticket, revoked)" ],
[ "ticket-rsc-sets-4", "Ticket - Resource sets (2 tickets, initial)" ],
[ "ticket-rsc-sets-5", "Ticket - Resource sets (2 tickets, granted)" ],
[ "ticket-rsc-sets-6", "Ticket - Resource sets (2 tickets, granted)" ],
[ "ticket-rsc-sets-7", "Ticket - Resource sets (2 tickets, revoked)" ],
[ "ticket-rsc-sets-8", "Ticket - Resource sets (1 ticket, standby, granted)" ],
[ "ticket-rsc-sets-9", "Ticket - Resource sets (1 ticket, granted, standby)" ],
[ "ticket-rsc-sets-10", "Ticket - Resource sets (1 ticket, standby, revoked)" ],
[ "ticket-rsc-sets-11", "Ticket - Resource sets (2 tickets, standby, granted)" ],
[ "ticket-rsc-sets-12", "Ticket - Resource sets (2 tickets, standby, granted)" ],
[ "ticket-rsc-sets-13", "Ticket - Resource sets (2 tickets, granted, standby)" ],
[ "ticket-rsc-sets-14", "Ticket - Resource sets (2 tickets, standby, revoked)" ],
[ "cluster-specific-params", "Cluster-specific instance attributes based on rules" ],
[ "site-specific-params", "Site-specific instance attributes based on rules" ],
],
[
[ "template-1", "Template - 1" ],
[ "template-2", "Template - 2" ],
[ "template-3", "Template - 3 (merge operations)" ],
[ "template-coloc-1", "Template - Colocation 1" ],
[ "template-coloc-2", "Template - Colocation 2" ],
[ "template-coloc-3", "Template - Colocation 3" ],
[ "template-order-1", "Template - Order 1" ],
[ "template-order-2", "Template - Order 2" ],
[ "template-order-3", "Template - Order 3" ],
[ "template-ticket", "Template - Ticket" ],
[ "template-rsc-sets-1", "Template - Resource Sets 1" ],
[ "template-rsc-sets-2", "Template - Resource Sets 2" ],
[ "template-rsc-sets-3", "Template - Resource Sets 3" ],
[ "template-rsc-sets-4", "Template - Resource Sets 4" ],
[ "template-clone-primitive", "Cloned primitive from template" ],
[ "template-clone-group", "Cloned group from template" ],
[ "location-sets-templates", "Resource sets and templates - Location" ],
[ "tags-coloc-order-1", "Tags - Colocation and Order (Simple)" ],
[ "tags-coloc-order-2", "Tags - Colocation and Order (Resource Sets with Templates)" ],
[ "tags-location", "Tags - Location" ],
[ "tags-ticket", "Tags - Ticket" ],
],
[
[ "container-1", "Container - initial" ],
[ "container-2", "Container - monitor failed" ],
[ "container-3", "Container - stop failed" ],
[ "container-4", "Container - reached migration-threshold" ],
[ "container-group-1", "Container in group - initial" ],
[ "container-group-2", "Container in group - monitor failed" ],
[ "container-group-3", "Container in group - stop failed" ],
[ "container-group-4", "Container in group - reached migration-threshold" ],
[ "container-is-remote-node", "Place resource within container when container is remote-node" ],
[ "bug-rh-1097457", "Kill user defined container/contents ordering" ],
[ "bug-cl-5247", "Graph loop when recovering m/s resource in a container" ],
[ "bundle-order-startup", "Bundle startup ordering" ],
[ "bundle-order-partial-start",
"Bundle startup ordering when some dependencies are already running" ],
[ "bundle-order-partial-start-2",
"Bundle startup ordering when some dependencies and the container are already running" ],
[ "bundle-order-stop", "Bundle stop ordering" ],
[ "bundle-order-partial-stop", "Bundle startup ordering when some dependencies are already stopped" ],
[ "bundle-order-stop-on-remote", "Stop nested resource after bringing up the connection" ],
[ "bundle-order-startup-clone", "Prevent startup because bundle isn't promoted" ],
[ "bundle-order-startup-clone-2", "Bundle startup with clones" ],
[ "bundle-order-stop-clone", "Stop bundle because clone is stopping" ],
[ "bundle-nested-colocation", "Colocation of nested connection resources" ],
[ "bundle-order-fencing",
"Order pseudo bundle fencing after parent node fencing if both are happening" ],
[ "bundle-probe-order-1", "order 1" ],
[ "bundle-probe-order-2", "order 2" ],
[ "bundle-probe-order-3", "order 3" ],
[ "bundle-probe-remotes", "Ensure remotes get probed too" ],
[ "bundle-replicas-change", "Change bundle from 1 replica to multiple" ],
[ "nested-remote-recovery", "Recover bundle's container hosted on remote node" ],
],
[
[ "whitebox-fail1", "Fail whitebox container rsc" ],
[ "whitebox-fail2", "Fail cluster connection to guest node" ],
[ "whitebox-fail3", "Failed containers should not run nested on remote nodes" ],
[ "whitebox-start", "Start whitebox container with resources assigned to it" ],
[ "whitebox-stop", "Stop whitebox container with resources assigned to it" ],
[ "whitebox-move", "Move whitebox container with resources assigned to it" ],
[ "whitebox-asymmetric", "Verify connection rsc opts-in based on container resource" ],
[ "whitebox-ms-ordering", "Verify promote/demote can not occur before connection is established" ],
[ "whitebox-ms-ordering-move", "Stop/Start cycle within a moving container" ],
[ "whitebox-orphaned", "Properly shutdown orphaned whitebox container" ],
[ "whitebox-orphan-ms", "Properly tear down orphan ms resources on remote-nodes" ],
[ "whitebox-unexpectedly-running", "Recover container nodes the cluster did not start" ],
[ "whitebox-migrate1", "Migrate both container and connection resource" ],
[ "whitebox-imply-stop-on-fence",
"imply stop action on container node rsc when host node is fenced" ],
[ "whitebox-nested-group", "Verify guest remote-node works nested in a group" ],
[ "guest-node-host-dies", "Verify guest node is recovered if host goes away" ],
[ "guest-node-cleanup", "Order guest node connection recovery after container probe" ],
[ "guest-host-not-fenceable", "Actions on guest node are unrunnable if host is unclean and cannot be fenced" ],
],
[
[ "remote-startup-probes", "Baremetal remote-node startup probes" ],
[ "remote-startup", "Startup a newly discovered remote-nodes with no status" ],
[ "remote-fence-unclean", "Fence unclean baremetal remote-node" ],
[ "remote-fence-unclean2",
"Fence baremetal remote-node after cluster node fails and connection can not be recovered" ],
[ "remote-fence-unclean-3", "Probe failed remote nodes (triggers fencing)" ],
[ "remote-move", "Move remote-node connection resource" ],
[ "remote-disable", "Disable a baremetal remote-node" ],
[ "remote-probe-disable", "Probe then stop a baremetal remote-node" ],
[ "remote-orphaned", "Properly shutdown orphaned connection resource" ],
[ "remote-orphaned2",
"verify we can handle orphaned remote connections with active resources on the remote" ],
[ "remote-recover", "Recover connection resource after cluster-node fails" ],
[ "remote-stale-node-entry",
"Make sure we properly handle leftover remote-node entries in the node section" ],
[ "remote-partial-migrate",
"Make sure partial migrations are handled before ops on the remote node" ],
[ "remote-partial-migrate2",
"Make sure partial migration target is prefered for remote connection" ],
[ "remote-recover-fail", "Make sure start failure causes fencing if rsc are active on remote" ],
[ "remote-start-fail",
"Make sure a start failure does not result in fencing if no active resources are on remote" ],
[ "remote-unclean2",
"Make monitor failure always results in fencing, even if no rsc are active on remote" ],
[ "remote-fence-before-reconnect", "Fence before clearing recurring monitor failure" ],
[ "remote-recovery", "Recover remote connections before attempting demotion" ],
[ "remote-recover-connection", "Optimistically recovery of only the connection" ],
[ "remote-recover-all", "Fencing when the connection has no home" ],
[ "remote-recover-no-resources", "Fencing when the connection has no home and no active resources" ],
[ "remote-recover-unknown",
"Fencing when the connection has no home and the remote has no operation history" ],
[ "remote-reconnect-delay", "Waiting for remote reconnect interval to expire" ],
[ "remote-connection-unrecoverable",
"Remote connection host must be fenced, with connection unrecoverable" ],
],
[
[ "resource-discovery", "Exercises resource-discovery location constraint option" ],
[ "rsc-discovery-per-node", "Disable resource discovery per node" ],
+ [ "shutdown-lock", "Ensure shutdown lock works properly" ],
+ [ "shutdown-lock-expiration", "Ensure shutdown lock expiration works properly" ],
],
# @TODO: If pacemaker implements versioned attributes, uncomment these tests
#[
# [ "versioned-resources", "Start resources with #ra-version rules" ],
# [ "restart-versioned", "Restart resources on #ra-version change" ],
# [ "reload-versioned", "Reload resources on #ra-version change" ],
#],
#[
# [ "versioned-operations-1", "Use #ra-version to configure operations of native resources" ],
# [ "versioned-operations-2", "Use #ra-version to configure operations of stonith resources" ],
# [ "versioned-operations-3", "Use #ra-version to configure operations of master/slave resources" ],
# [ "versioned-operations-4", "Use #ra-version to configure operations of groups of the resources" ],
#],
]
TESTS_64BIT = [
[
[ "year-2038", "Check handling of timestamps beyond 2038-01-19 03:14:08 UTC" ],
],
]
# Constants substituted in the build process
class BuildVars(object):
SBINDIR = "@sbindir@"
BUILDDIR = "@abs_top_builddir@"
CRM_SCHEMA_DIRECTORY = "@CRM_SCHEMA_DIRECTORY@"
# These values must be kept in sync with include/crm/crm.h
class CrmExit(object):
OK = 0
ERROR = 1
NOT_INSTALLED = 5
NOINPUT = 66
def is_executable(path):
""" Check whether a file at a given path is executable. """
try:
return os.stat(path)[stat.ST_MODE] & stat.S_IXUSR
except OSError:
return False
def diff(file1, file2, **kwargs):
""" Call diff on two files """
return subprocess.call([ "diff", "-u", "-N", "--ignore-all-space",
"--ignore-blank-lines", file1, file2 ], **kwargs)
def sort_file(filename):
""" Sort a file alphabetically """
with io.open(filename, "rt") as f:
lines = sorted(f)
with io.open(filename, "wt") as f:
f.writelines(lines)
def remove_files(filenames):
""" Remove a list of files """
for filename in filenames:
try:
os.remove(filename)
except OSError:
pass
def normalize(filename):
""" Remove text from a file that isn't important for comparison """
if not hasattr(normalize, "patterns"):
normalize.patterns = [
re.compile(r'crm_feature_set="[^"]*"'),
re.compile(r'batch-limit="[0-9]*"')
]
if os.path.isfile(filename):
with io.open(filename, "rt") as f:
lines = f.readlines()
with io.open(filename, "wt") as f:
for line in lines:
for pattern in normalize.patterns:
line = pattern.sub("", line)
f.write(line)
def cat(filename, dest=sys.stdout):
""" Copy a file to a destination file descriptor """
with io.open(filename, "rt") as f:
shutil.copyfileobj(f, dest)
class CtsScheduler(object):
""" Regression tests for Pacemaker's scheduler """
def _parse_args(self, argv):
""" Parse command-line arguments """
parser = argparse.ArgumentParser(description=DESC)
parser.add_argument('-V', '--verbose', action='count',
help='Display any differences from expected output')
parser.add_argument('--run', metavar='TEST',
help=('Run only single specified test (any further '
'arguments will be passed to crm_simulate)'))
parser.add_argument('--update', action='store_true',
help='Update expected results with actual results')
parser.add_argument('-b', '--binary', metavar='PATH',
help='Specify path to crm_simulate')
parser.add_argument('-i', '--io-dir', metavar='PATH',
help='Specify path to regression test data directory')
parser.add_argument('-o', '--out-dir', metavar='PATH',
help='Specify where intermediate and output files should go')
parser.add_argument('-v', '--valgrind', action='store_true',
help='Run all commands under valgrind')
parser.add_argument('--valgrind-dhat', action='store_true',
help='Run all commands under valgrind with heap analyzer')
parser.add_argument('--valgrind-skip-output', action='store_true',
help='If running under valgrind, do not display output')
parser.add_argument('--testcmd-options', metavar='OPTIONS', default='',
help='Additional options for command under test')
# argparse can't handle "everything after --run TEST", so grab that
self.single_test_args = []
narg = 0
for arg in argv:
narg = narg + 1
if arg == '--run':
(argv, self.single_test_args) = (argv[:narg+1], argv[narg+1:])
break
self.args = parser.parse_args(argv[1:])
def _error(self, s):
print(" * ERROR: %s" % s)
def _failed(self, s):
print(" * FAILED: %s" % s)
def _get_valgrind_cmd(self):
""" Return command arguments needed (or not) to run valgrind """
if self.args.valgrind:
os.environ['G_SLICE'] = "always-malloc"
return [
"valgrind",
"-q",
"--gen-suppressions=all",
"--time-stamp=yes",
"--trace-children=no",
"--show-reachable=no",
"--leak-check=full",
"--num-callers=20",
"--suppressions=%s/valgrind-pcmk.suppressions" % (self.test_home)
]
if self.args.valgrind_dhat:
os.environ['G_SLICE'] = "always-malloc"
return [
"valgrind",
"--tool=exp-dhat",
"--time-stamp=yes",
"--trace-children=no",
"--show-top-n=100",
"--num-callers=4"
]
return []
def _get_simulator_cmd(self):
""" Locate the simulation binary """
if self.args.binary is None:
self.args.binary = BuildVars.BUILDDIR + "/tools/crm_simulate"
if not is_executable(self.args.binary):
self.args.binary = BuildVars.SBINDIR + "/crm_simulate"
if not is_executable(self.args.binary):
# @TODO it would be more pythonic to raise an exception
self._error("Test binary " + self.args.binary + " not found")
sys.exit(CrmExit.NOT_INSTALLED)
return [ self.args.binary ] + shlex.split(self.args.testcmd_options)
def set_schema_env(self):
""" Ensure schema directory environment variable is set, if possible """
try:
return os.environ['PCMK_schema_directory']
except KeyError:
for d in [ os.path.join(BuildVars.BUILDDIR, "xml"),
BuildVars.CRM_SCHEMA_DIRECTORY ]:
if os.path.isdir(d):
os.environ['PCMK_schema_directory'] = d
return d
return None
def __init__(self, argv=sys.argv):
self._parse_args(argv)
# Where this executable lives
self.test_home = os.path.dirname(os.path.realpath(argv[0]))
# Where test data resides
if self.args.io_dir is None:
self.args.io_dir = os.path.join(self.test_home, "scheduler")
# Where to store generated files
if self.args.out_dir is None:
self.args.out_dir = self.args.io_dir
self.failed_filename = os.path.join(self.test_home, ".regression.failed.diff")
else:
self.failed_filename = os.path.join(self.args.out_dir, ".regression.failed.diff")
os.environ['CIB_shadow_dir'] = self.args.out_dir
self.failed_file = None
# Single test mode (if requested)
try:
# User can give test base name or file name of a test input
self.args.run = os.path.splitext(os.path.basename(self.args.run))[0]
except (AttributeError, TypeError):
pass # --run was not specified
self.set_schema_env()
# Arguments needed (or not) to run commands
self.valgrind_args = self._get_valgrind_cmd()
self.simulate_args = self._get_simulator_cmd()
# Test counters
self.num_failed = 0
self.num_tests = 0
def _compare_files(self, filename1, filename2):
""" Add any file differences to failed results """
with io.open("/dev/null", "wt") as dev_null:
if diff(filename1, filename2, stdout=dev_null) != 0:
diff(filename1, filename2, stdout=self.failed_file, stderr=dev_null)
self.failed_file.write("\n");
return True
return False
def run_one(self, test_name, test_desc, test_args=[]):
""" Run one scheduler test """
print(" Test %-25s %s" % ((test_name + ":"), test_desc))
did_fail = False
self.num_tests = self.num_tests + 1
# Test inputs
input_filename = "%s/%s.xml" % (self.args.io_dir, test_name)
expected_filename = "%s/%s.exp" % (self.args.io_dir, test_name)
dot_expected_filename = "%s/%s.dot" % (self.args.io_dir, test_name)
scores_filename = "%s/%s.scores" % (self.args.io_dir, test_name)
summary_filename = "%s/%s.summary" % (self.args.io_dir, test_name)
stderr_expected_filename = "%s/%s.stderr" % (self.args.io_dir, test_name)
# (Intermediate) test outputs
output_filename = "%s/%s.out" % (self.args.out_dir, test_name)
dot_output_filename = "%s/%s.pe.dot" % (self.args.out_dir, test_name)
score_output_filename = "%s/%s.scores.pe" % (self.args.out_dir, test_name)
summary_output_filename = "%s/%s.summary.pe" % (self.args.out_dir, test_name)
stderr_output_filename = "%s/%s.stderr.pe" % (self.args.out_dir, test_name)
valgrind_output_filename = "%s/%s.valgrind" % (self.args.out_dir, test_name)
# Common arguments for running test
test_cmd = []
if self.valgrind_args:
test_cmd = self.valgrind_args + [ "--log-file=%s" % valgrind_output_filename ]
test_cmd = test_cmd + self.simulate_args
# @TODO It would be more pythonic to raise exceptions for errors,
# then perhaps it would be nice to make a single-test class
# Ensure necessary test inputs exist
if not os.path.isfile(input_filename):
self._error("No input")
self.num_failed = self.num_failed + 1
return CrmExit.NOINPUT
if not self.args.update and not os.path.isfile(expected_filename):
self._error("no stored output")
return CrmExit.NOINPUT
# Run simulation to generate summary output
if self.args.run: # Single test mode
test_cmd_full = test_cmd + [ '-x', input_filename, '-S' ] + test_args
print(" ".join(test_cmd_full))
else:
# @TODO Why isn't test_args added here?
test_cmd_full = test_cmd + [ '-x', input_filename, '-S' ]
with io.open(summary_output_filename, "wt") as f:
subprocess.call(test_cmd_full, stdout=f, stderr=subprocess.STDOUT, env=os.environ)
if self.args.run:
cat(summary_output_filename)
# Re-run simulation to generate dot, graph, and scores
test_cmd_full = test_cmd + [
'-x', input_filename,
'-D', dot_output_filename,
'-G', output_filename,
'-sSQ' ] + test_args
with io.open(stderr_output_filename, "wt") as f_stderr, \
io.open(score_output_filename, "wt") as f_score:
rc = subprocess.call(test_cmd_full, stdout=f_score, stderr=f_stderr, env=os.environ)
# Check for test command failure
if rc != CrmExit.OK:
self._failed("Test returned: %d" % rc)
did_fail = True
print(" ".join(test_cmd_full))
# Check for valgrind errors
if self.valgrind_args and not self.args.valgrind_skip_output:
if os.stat(valgrind_output_filename).st_size > 0:
self._failed("Valgrind reported errors")
did_fail = True
cat(valgrind_output_filename)
remove_files([ valgrind_output_filename ])
# Check for core dump
if os.path.isfile("core"):
self._failed("Core-file detected: core." + test_name)
did_fail = True
os.rename("core", "%s/core.%s" % (self.test_home, test_name))
# Check any stderr output
if os.path.isfile(stderr_expected_filename):
if self._compare_files(stderr_expected_filename, stderr_output_filename):
self._failed("stderr changed")
did_fail = True
elif os.stat(stderr_output_filename).st_size > 0:
self._failed("Output was written to stderr")
did_fail = True
cat(stderr_output_filename)
remove_files([ stderr_output_filename ])
# Check whether output graph exists, and normalize it
if (not os.path.isfile(output_filename)
or os.stat(output_filename).st_size == 0):
self._error("No graph produced")
did_fail = True
self.num_failed = self.num_failed + 1
remove_files([ output_filename ])
return CrmExit.ERROR
normalize(output_filename)
# Check whether dot output exists, and sort it
if (not os.path.isfile(dot_output_filename) or
os.stat(dot_output_filename).st_size == 0):
self._error("No dot-file summary produced")
did_fail = True
self.num_failed = self.num_failed + 1
remove_files([ dot_output_filename, output_filename ])
return CrmExit.ERROR
with io.open(dot_output_filename, "rt") as f:
first_line = f.readline() # "digraph" line with opening brace
lines = f.readlines()
last_line = lines[-1] # closing brace
del lines[-1]
lines = sorted(set(lines)) # unique sort
with io.open(dot_output_filename, "wt") as f:
f.write(first_line)
f.writelines(lines)
f.write(last_line)
# Check whether score output exists, and sort it
if (not os.path.isfile(score_output_filename)
or os.stat(score_output_filename).st_size == 0):
self._error("No allocation scores produced")
did_fail = True
self.num_failed = self.num_failed + 1
remove_files([ score_output_filename, output_filename ])
return CrmExit.ERROR
else:
sort_file(score_output_filename)
if self.args.update:
shutil.copyfile(output_filename, expected_filename)
shutil.copyfile(dot_output_filename, dot_expected_filename)
shutil.copyfile(score_output_filename, scores_filename)
shutil.copyfile(summary_output_filename, summary_filename)
print(" Updated expected outputs")
if self._compare_files(summary_filename, summary_output_filename):
self._failed("summary changed")
did_fail = True
if self._compare_files(dot_expected_filename, dot_output_filename):
self._failed("dot-file summary changed")
did_fail = True
else:
remove_files([ dot_output_filename ])
if self._compare_files(expected_filename, output_filename):
self._failed("xml-file changed")
did_fail = True
if self._compare_files(scores_filename, score_output_filename):
self._failed("scores-file changed")
did_fail = True
remove_files([ output_filename,
score_output_filename,
summary_output_filename])
if did_fail:
self.num_failed = self.num_failed + 1
return CrmExit.ERROR
return CrmExit.OK
def run_all(self):
""" Run all defined tests """
if platform.architecture()[0] == "64bit":
TESTS.extend(TESTS_64BIT)
for group in TESTS:
for test in group:
try:
args = test[2]
except IndexError:
args = []
self.run_one(test[0], test[1], args)
print()
def _print_summary(self):
""" Print a summary of parameters for this test run """
print("Test home is:\t" + self.test_home)
print("Test binary is:\t" + self.args.binary)
if 'PCMK_schema_directory' in os.environ:
print("Schema home is:\t" + os.environ['PCMK_schema_directory'])
if self.valgrind_args != []:
print("Activating memory testing with valgrind")
print()
def _test_results(self):
if self.num_failed == 0:
return CrmExit.OK
if os.path.isfile(self.failed_filename) and os.stat(self.failed_filename).st_size != 0:
if self.args.verbose:
self._error("Results of %d failed tests (out of %d):" %
(self.num_failed, self.num_tests))
cat(self.failed_filename)
else:
self._error("Results of %d failed tests (out of %d) are in %s" %
(self.num_failed, self.num_tests, self.failed_filename))
self._error("Use -V to display them after running the tests")
else:
self._error("%d (of %d) tests failed (no diff results)" %
(self.num_failed, self.num_tests))
if os.path.isfile(self.failed_filename):
os.remove(self.failed_filename)
return CrmExit.ERROR
def run(self):
""" Run test(s) as specified """
self._print_summary()
# Zero out the error log
self.failed_file = io.open(self.failed_filename, "wt")
if self.args.run is None:
print("Performing the following tests from " + self.args.io_dir)
print()
self.run_all()
print()
self.failed_file.close()
rc = self._test_results()
else:
rc = self.run_one(self.args.run, "Single shot", self.single_test_args)
self.failed_file.close()
cat(self.failed_filename)
return rc
if __name__ == "__main__":
sys.exit(CtsScheduler().run())
# vim: set filetype=python expandtab tabstop=4 softtabstop=4 shiftwidth=4 textwidth=120:
diff --git a/cts/scheduler/shutdown-lock-expiration.dot b/cts/scheduler/shutdown-lock-expiration.dot
new file mode 100644
index 0000000000..ee9907937a
--- /dev/null
+++ b/cts/scheduler/shutdown-lock-expiration.dot
@@ -0,0 +1,11 @@
+ digraph "g" {
+"Fencing_monitor_120000 node3" [ style=bold color="green" fontcolor="black"]
+"Fencing_start_0 node3" -> "Fencing_monitor_120000 node3" [ style = bold]
+"Fencing_start_0 node3" [ style=bold color="green" fontcolor="black"]
+"Fencing_stop_0 node3" -> "Fencing_start_0 node3" [ style = bold]
+"Fencing_stop_0 node3" [ style=bold color="green" fontcolor="black"]
+"rsc2_lrm_delete_0 node2" [ style=bold color="green" fontcolor="black"]
+"rsc2_monitor_10000 node4" [ style=bold color="green" fontcolor="black"]
+"rsc2_start_0 node4" -> "rsc2_monitor_10000 node4" [ style = bold]
+"rsc2_start_0 node4" [ style=bold color="green" fontcolor="black"]
+}
diff --git a/cts/scheduler/shutdown-lock-expiration.exp b/cts/scheduler/shutdown-lock-expiration.exp
new file mode 100644
index 0000000000..465f12b864
--- /dev/null
+++ b/cts/scheduler/shutdown-lock-expiration.exp
@@ -0,0 +1,68 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/cts/scheduler/shutdown-lock-expiration.scores b/cts/scheduler/shutdown-lock-expiration.scores
new file mode 100644
index 0000000000..e5d435df3b
--- /dev/null
+++ b/cts/scheduler/shutdown-lock-expiration.scores
@@ -0,0 +1,17 @@
+Allocation scores:
+Using the original execution date of: 2020-01-06 22:11:40Z
+native_color: Fencing allocation score on node1: 0
+native_color: Fencing allocation score on node2: 0
+native_color: Fencing allocation score on node3: 0
+native_color: Fencing allocation score on node4: 0
+native_color: Fencing allocation score on node5: 0
+native_color: rsc1 allocation score on node1: INFINITY
+native_color: rsc1 allocation score on node2: -INFINITY
+native_color: rsc1 allocation score on node3: -INFINITY
+native_color: rsc1 allocation score on node4: -INFINITY
+native_color: rsc1 allocation score on node5: -INFINITY
+native_color: rsc2 allocation score on node1: 0
+native_color: rsc2 allocation score on node2: INFINITY
+native_color: rsc2 allocation score on node3: 0
+native_color: rsc2 allocation score on node4: 0
+native_color: rsc2 allocation score on node5: 0
diff --git a/cts/scheduler/shutdown-lock-expiration.summary b/cts/scheduler/shutdown-lock-expiration.summary
new file mode 100644
index 0000000000..08c93aa87f
--- /dev/null
+++ b/cts/scheduler/shutdown-lock-expiration.summary
@@ -0,0 +1,31 @@
+Using the original execution date of: 2020-01-06 22:11:40Z
+
+Current cluster status:
+Online: [ node3 node4 node5 ]
+OFFLINE: [ node1 node2 ]
+
+ Fencing (stonith:fence_xvm): Started node3
+ rsc1 (ocf::pacemaker:Dummy): Stopped node1 (LOCKED)
+ rsc2 (ocf::pacemaker:Dummy): Stopped
+
+Transition Summary:
+ * Restart Fencing ( node3 ) due to resource definition change
+ * Start rsc2 ( node4 )
+
+Executing cluster transition:
+ * Resource action: Fencing stop on node3
+ * Resource action: Fencing start on node3
+ * Resource action: Fencing monitor=120000 on node3
+ * Resource action: rsc2 start on node4
+ * Cluster action: lrm_delete for rsc2 on node2
+ * Resource action: rsc2 monitor=10000 on node4
+Using the original execution date of: 2020-01-06 22:11:40Z
+
+Revised cluster status:
+Online: [ node3 node4 node5 ]
+OFFLINE: [ node1 node2 ]
+
+ Fencing (stonith:fence_xvm): Started node3
+ rsc1 (ocf::pacemaker:Dummy): Stopped node1 (LOCKED)
+ rsc2 (ocf::pacemaker:Dummy): Started node4
+
diff --git a/cts/scheduler/shutdown-lock-expiration.xml b/cts/scheduler/shutdown-lock-expiration.xml
new file mode 100644
index 0000000000..26f720eb6f
--- /dev/null
+++ b/cts/scheduler/shutdown-lock-expiration.xml
@@ -0,0 +1,187 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/cts/scheduler/shutdown-lock.dot b/cts/scheduler/shutdown-lock.dot
new file mode 100644
index 0000000000..0a7d8c34df
--- /dev/null
+++ b/cts/scheduler/shutdown-lock.dot
@@ -0,0 +1,11 @@
+ digraph "g" {
+"Fencing_monitor_120000 node3" [ style=bold color="green" fontcolor="black"]
+"Fencing_start_0 node3" -> "Fencing_monitor_120000 node3" [ style = bold]
+"Fencing_start_0 node3" [ style=bold color="green" fontcolor="black"]
+"Fencing_stop_0 node1" -> "Fencing_start_0 node3" [ style = bold]
+"Fencing_stop_0 node1" -> "do_shutdown node1" [ style = bold]
+"Fencing_stop_0 node1" [ style=bold color="green" fontcolor="black"]
+"do_shutdown node1" [ style=bold color="green" fontcolor="black"]
+"rsc1_stop_0 node1" -> "do_shutdown node1" [ style = bold]
+"rsc1_stop_0 node1" [ style=bold color="green" fontcolor="black"]
+}
diff --git a/cts/scheduler/shutdown-lock.exp b/cts/scheduler/shutdown-lock.exp
new file mode 100644
index 0000000000..e8bf9d869d
--- /dev/null
+++ b/cts/scheduler/shutdown-lock.exp
@@ -0,0 +1,64 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/cts/scheduler/shutdown-lock.scores b/cts/scheduler/shutdown-lock.scores
new file mode 100644
index 0000000000..e09ebfbf7a
--- /dev/null
+++ b/cts/scheduler/shutdown-lock.scores
@@ -0,0 +1,17 @@
+Allocation scores:
+Using the original execution date of: 2020-01-06 21:59:11Z
+native_color: Fencing allocation score on node1: 0
+native_color: Fencing allocation score on node2: 0
+native_color: Fencing allocation score on node3: 0
+native_color: Fencing allocation score on node4: 0
+native_color: Fencing allocation score on node5: 0
+native_color: rsc1 allocation score on node1: INFINITY
+native_color: rsc1 allocation score on node2: -INFINITY
+native_color: rsc1 allocation score on node3: -INFINITY
+native_color: rsc1 allocation score on node4: -INFINITY
+native_color: rsc1 allocation score on node5: -INFINITY
+native_color: rsc2 allocation score on node1: -INFINITY
+native_color: rsc2 allocation score on node2: INFINITY
+native_color: rsc2 allocation score on node3: -INFINITY
+native_color: rsc2 allocation score on node4: -INFINITY
+native_color: rsc2 allocation score on node5: -INFINITY
diff --git a/cts/scheduler/shutdown-lock.summary b/cts/scheduler/shutdown-lock.summary
new file mode 100644
index 0000000000..6ed56d1504
--- /dev/null
+++ b/cts/scheduler/shutdown-lock.summary
@@ -0,0 +1,31 @@
+Using the original execution date of: 2020-01-06 21:59:11Z
+
+Current cluster status:
+Online: [ node1 node3 node4 node5 ]
+OFFLINE: [ node2 ]
+
+ Fencing (stonith:fence_xvm): Started node1
+ rsc1 (ocf::pacemaker:Dummy): Started node1
+ rsc2 (ocf::pacemaker:Dummy): Stopped node2 (LOCKED)
+
+Transition Summary:
+ * Shutdown node1
+ * Move Fencing ( node1 -> node3 )
+ * Stop rsc1 ( node1 ) due to node availability
+
+Executing cluster transition:
+ * Resource action: Fencing stop on node1
+ * Resource action: rsc1 stop on node1
+ * Cluster action: do_shutdown on node1
+ * Resource action: Fencing start on node3
+ * Resource action: Fencing monitor=120000 on node3
+Using the original execution date of: 2020-01-06 21:59:11Z
+
+Revised cluster status:
+Online: [ node1 node3 node4 node5 ]
+OFFLINE: [ node2 ]
+
+ Fencing (stonith:fence_xvm): Started node3
+ rsc1 (ocf::pacemaker:Dummy): Stopped
+ rsc2 (ocf::pacemaker:Dummy): Stopped node2 (LOCKED)
+
diff --git a/cts/scheduler/shutdown-lock.xml b/cts/scheduler/shutdown-lock.xml
new file mode 100644
index 0000000000..ec6db3050e
--- /dev/null
+++ b/cts/scheduler/shutdown-lock.xml
@@ -0,0 +1,186 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/daemons/controld/controld_based.c b/daemons/controld/controld_based.c
index 42e321fa06..0ffc1e8c5e 100644
--- a/daemons/controld/controld_based.c
+++ b/daemons/controld/controld_based.c
@@ -1,245 +1,336 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include /* sleep */
#include
#include
#include
#include
#include
int cib_retries = 0;
static void
do_cib_updated(const char *event, xmlNode * msg)
{
if (pcmk__alert_in_patchset(msg, TRUE)) {
mainloop_set_trigger(config_read);
}
}
static void
do_cib_replaced(const char *event, xmlNode * msg)
{
crm_debug("Updating the CIB after a replace: DC=%s", AM_I_DC ? "true" : "false");
if (AM_I_DC == FALSE) {
return;
} else if (fsa_state == S_FINALIZE_JOIN && is_set(fsa_input_register, R_CIB_ASKED)) {
/* no need to restart the join - we asked for this replace op */
return;
}
/* start the join process again so we get everyone's LRM status */
populate_cib_nodes(node_update_quick|node_update_all, __FUNCTION__);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
}
/* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */
void
do_cib_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
CRM_ASSERT(fsa_cib_conn != NULL);
if (action & A_CIB_STOP) {
if (fsa_cib_conn->state != cib_disconnected && last_resource_update != 0) {
crm_info("Waiting for resource update %d to complete", last_resource_update);
crmd_fsa_stall(FALSE);
return;
}
crm_info("Disconnecting from the CIB manager");
clear_bit(fsa_input_register, R_CIB_CONNECTED);
fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated);
if (fsa_cib_conn->state != cib_disconnected) {
fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local);
fsa_cib_conn->cmds->signoff(fsa_cib_conn);
}
crm_notice("Disconnected from the CIB manager");
}
if (action & A_CIB_START) {
int rc = pcmk_ok;
if (cur_state == S_STOPPING) {
crm_err("Ignoring request to connect to the CIB manager after shutdown");
return;
}
rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking);
if (rc != pcmk_ok) {
/* a short wait that usually avoids stalling the FSA */
sleep(1);
rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking);
}
if (rc != pcmk_ok) {
crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc));
} else if (pcmk_ok !=
fsa_cib_conn->cmds->set_connection_dnotify(fsa_cib_conn,
crmd_cib_connection_destroy)) {
crm_err("Could not set dnotify callback");
} else if (pcmk_ok !=
fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_REPLACE_NOTIFY,
do_cib_replaced)) {
crm_err("Could not set CIB notification callback (replace)");
} else if (pcmk_ok !=
fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY,
do_cib_updated)) {
crm_err("Could not set CIB notification callback (update)");
} else {
set_bit(fsa_input_register, R_CIB_CONNECTED);
cib_retries = 0;
}
if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
cib_retries++;
crm_warn("Couldn't complete CIB registration %d"
" times... pause and retry", cib_retries);
if (cib_retries < 30) {
controld_start_timer(wait_timer);
crmd_fsa_stall(FALSE);
} else {
crm_err("Could not complete CIB"
" registration %d times..." " hard error", cib_retries);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
}
}
/*!
* \internal
* \brief Get CIB call options to use local scope if master unavailable
*
* \return CIB call options
*/
int crmd_cib_smart_opt()
{
int call_opt = cib_quorum_override;
if (fsa_state == S_ELECTION || fsa_state == S_PENDING) {
crm_info("Sending update to local CIB in state: %s", fsa_state2string(fsa_state));
call_opt |= cib_scope_local;
}
return call_opt;
}
/*!
* \internal
* \brief Check whether an action type should be recorded in the CIB
*
* \param[in] action Action type
*
* \return TRUE if action should be recorded, FALSE otherwise
*/
bool
controld_action_is_recordable(const char *action)
{
if (safe_str_eq(action, CRMD_ACTION_CANCEL)
|| safe_str_eq(action, CRMD_ACTION_DELETE)
|| safe_str_eq(action, CRMD_ACTION_NOTIFY)
|| safe_str_eq(action, CRMD_ACTION_METADATA)) {
return FALSE;
}
return TRUE;
}
static void
cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output,
void *user_data)
{
char *desc = user_data;
if (rc == 0) {
crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id);
} else {
crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d",
desc, call_id, pcmk_strerror(rc), rc);
}
}
// Searches for various portions of node_state to delete
// Match a particular node's node_state (takes node name 1x)
#define XPATH_NODE_STATE "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']"
// Node's lrm section (name 1x)
#define XPATH_NODE_LRM XPATH_NODE_STATE "/" XML_CIB_TAG_LRM
+// Node's lrm_rsc_op entries and lrm_resource entries without lock (name 2x)
+#define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" XML_LRM_TAG_RSC_OP \
+ "|" XPATH_NODE_STATE \
+ "//" XML_LRM_TAG_RESOURCE \
+ "[not(@" XML_CONFIG_ATTR_SHUTDOWN_LOCK ")]"
+
// Node's transient_attributes section (name 1x)
#define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" XML_TAG_TRANSIENT_NODEATTRS
// Everything under node_state (name 1x)
#define XPATH_NODE_ALL XPATH_NODE_STATE "/*"
+// Unlocked history + transient attributes (name 3x)
+#define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS
+
/*!
* \internal
* \brief Delete subsection of a node's CIB node_state
*
* \param[in] uname Desired node
* \param[in] section Subsection of node_state to delete
* \param[in] options CIB call options to use
*/
void
controld_delete_node_state(const char *uname, enum controld_section_e section,
int options)
{
char *xpath = NULL;
char *desc = NULL;
CRM_CHECK(uname != NULL, return);
switch (section) {
case controld_section_lrm:
xpath = crm_strdup_printf(XPATH_NODE_LRM, uname);
desc = crm_strdup_printf("resource history for node %s", uname);
break;
+ case controld_section_lrm_unlocked:
+ xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, uname, uname);
+ desc = crm_strdup_printf("resource history (other than shutdown "
+ "locks) for node %s", uname);
+ break;
case controld_section_attrs:
xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname);
desc = crm_strdup_printf("transient attributes for node %s", uname);
break;
case controld_section_all:
xpath = crm_strdup_printf(XPATH_NODE_ALL, uname);
desc = crm_strdup_printf("all state for node %s", uname);
break;
+ case controld_section_all_unlocked:
+ xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED,
+ uname, uname, uname);
+ desc = crm_strdup_printf("all state (other than shutdown locks) "
+ "for node %s", uname);
+ break;
}
if (fsa_cib_conn == NULL) {
crm_warn("Unable to delete %s: no CIB connection", desc);
free(desc);
} else {
int call_id;
- options |= cib_quorum_override|cib_xpath;
+ options |= cib_quorum_override|cib_xpath|cib_multiple;
call_id = fsa_cib_conn->cmds->remove(fsa_cib_conn, xpath, NULL, options);
crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s",
desc, call_id, xpath);
fsa_register_cib_callback(call_id, FALSE, desc, cib_delete_callback);
// CIB library handles freeing desc
}
free(xpath);
}
+
+// Takes node name and resource ID
+#define XPATH_RESOURCE_HISTORY "//" XML_CIB_TAG_STATE \
+ "[@" XML_ATTR_UNAME "='%s'] /" \
+ XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \
+ "/" XML_LRM_TAG_RESOURCE \
+ "[@" XML_ATTR_ID "='%s']"
+// @TODO could add "and @XML_CONFIG_ATTR_SHUTDOWN_LOCK" to limit to locks
+
+/*!
+ * \internal
+ * \brief Clear resource history from CIB for a given resource and node
+ *
+ * \param[in] rsc_id ID of resource to be cleared
+ * \param[in] node Node whose resource history should be cleared
+ * \param[in] user_name ACL user name to use
+ * \param[in] call_options CIB call options
+ *
+ * \return Standard Pacemaker return code
+ */
+int
+controld_delete_resource_history(const char *rsc_id, const char *node,
+ const char *user_name, int call_options)
+{
+ char *desc = NULL;
+ char *xpath = NULL;
+ int rc = pcmk_rc_ok;
+
+ CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL);
+
+ desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node);
+ if (fsa_cib_conn == NULL) {
+ crm_err("Unable to clear %s: no CIB connection", desc);
+ free(desc);
+ return ENOTCONN;
+ }
+
+ // Ask CIB to delete the entry
+ xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id);
+ rc = cib_internal_op(fsa_cib_conn, CIB_OP_DELETE, NULL, xpath, NULL,
+ NULL, call_options|cib_xpath, user_name);
+
+ if (rc < 0) {
+ rc = pcmk_legacy2rc(rc);
+ crm_err("Could not delete resource status of %s on %s%s%s: %s "
+ CRM_XS " rc=%d", rsc_id, node,
+ (user_name? " for user " : ""), (user_name? user_name : ""),
+ pcmk_rc_str(rc), rc);
+ free(desc);
+ free(xpath);
+ return rc;
+ }
+
+ if (is_set(call_options, cib_sync_call)) {
+ if (is_set(call_options, cib_dryrun)) {
+ crm_debug("Deletion of %s would succeed", desc);
+ } else {
+ crm_debug("Deletion of %s succeeded", desc);
+ }
+ free(desc);
+
+ } else {
+ crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s",
+ desc, rc, xpath);
+ fsa_register_cib_callback(rc, FALSE, desc, cib_delete_callback);
+ // CIB library handles freeing desc
+ }
+
+ free(xpath);
+ return pcmk_rc_ok;
+}
diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c
index 6c7f97cbc0..c918a1ef42 100644
--- a/daemons/controld/controld_control.c
+++ b/daemons/controld/controld_control.c
@@ -1,773 +1,780 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
qb_ipcs_service_t *ipcs = NULL;
#if SUPPORT_COROSYNC
extern gboolean crm_connect_corosync(crm_cluster_t * cluster);
#endif
void crm_shutdown(int nsig);
gboolean crm_read_options(gpointer user_data);
gboolean fsa_has_quorum = FALSE;
crm_trigger_t *fsa_source = NULL;
crm_trigger_t *config_read = NULL;
bool no_quorum_suicide_escalation = FALSE;
+bool controld_shutdown_lock_enabled = false;
/* A_HA_CONNECT */
void
do_ha_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
gboolean registered = FALSE;
static crm_cluster_t *cluster = NULL;
if (cluster == NULL) {
cluster = calloc(1, sizeof(crm_cluster_t));
}
if (action & A_HA_DISCONNECT) {
crm_cluster_disconnect(cluster);
crm_info("Disconnected from the cluster");
set_bit(fsa_input_register, R_HA_DISCONNECTED);
}
if (action & A_HA_CONNECT) {
crm_set_status_callback(&peer_update_callback);
crm_set_autoreap(FALSE);
if (is_corosync_cluster()) {
#if SUPPORT_COROSYNC
registered = crm_connect_corosync(cluster);
#endif
}
if (registered == TRUE) {
controld_election_init(cluster->uname);
fsa_our_uname = cluster->uname;
fsa_our_uuid = cluster->uuid;
if(cluster->uuid == NULL) {
crm_err("Could not obtain local uuid");
registered = FALSE;
}
}
if (registered == FALSE) {
set_bit(fsa_input_register, R_HA_DISCONNECTED);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
populate_cib_nodes(node_update_none, __FUNCTION__);
clear_bit(fsa_input_register, R_HA_DISCONNECTED);
crm_info("Connected to the cluster");
}
if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) {
crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__);
}
}
/* A_SHUTDOWN */
void
do_shutdown(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
/* just in case */
set_bit(fsa_input_register, R_SHUTDOWN);
controld_disconnect_fencer(FALSE);
}
/* A_SHUTDOWN_REQ */
void
do_shutdown_req(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
xmlNode *msg = NULL;
set_bit(fsa_input_register, R_SHUTDOWN);
crm_info("Sending shutdown request to all peers (DC is %s)",
(fsa_our_dc? fsa_our_dc : "not set"));
msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
/* set_bit(fsa_input_register, R_STAYDOWN); */
if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
free_xml(msg);
}
extern char *max_generation_from;
extern xmlNode *max_generation_xml;
extern GHashTable *resource_history;
extern GHashTable *voted;
void
crmd_fast_exit(crm_exit_t exit_code)
{
if (is_set(fsa_input_register, R_STAYDOWN)) {
crm_warn("Inhibiting respawn "CRM_XS" remapping exit code %d to %d",
exit_code, CRM_EX_FATAL);
exit_code = CRM_EX_FATAL;
} else if ((exit_code == CRM_EX_OK)
&& is_set(fsa_input_register, R_IN_RECOVERY)) {
crm_err("Could not recover from internal error");
exit_code = CRM_EX_ERROR;
}
crm_exit(exit_code);
}
crm_exit_t
crmd_exit(crm_exit_t exit_code)
{
GListPtr gIter = NULL;
GMainLoop *mloop = crmd_mainloop;
static bool in_progress = FALSE;
if (in_progress && (exit_code == CRM_EX_OK)) {
crm_debug("Exit is already in progress");
return exit_code;
} else if(in_progress) {
crm_notice("Error during shutdown process, exiting now with status %d (%s)",
exit_code, crm_exit_str(exit_code));
crm_write_blackbox(SIGTRAP, NULL);
crmd_fast_exit(exit_code);
}
in_progress = TRUE;
crm_trace("Preparing to exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
/* Suppress secondary errors resulting from us disconnecting everything */
set_bit(fsa_input_register, R_HA_DISCONNECTED);
/* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */
if(ipcs) {
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs);
ipcs = NULL;
}
controld_close_attrd_ipc();
pe_subsystem_free();
controld_disconnect_fencer(TRUE);
if ((exit_code == CRM_EX_OK) && (crmd_mainloop == NULL)) {
crm_debug("No mainloop detected");
exit_code = CRM_EX_ERROR;
}
/* On an error, just get out.
*
* Otherwise, make the effort to have mainloop exit gracefully so
* that it (mostly) cleans up after itself and valgrind has less
* to report on - allowing real errors stand out
*/
if (exit_code != CRM_EX_OK) {
crm_notice("Forcing immediate exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
crm_write_blackbox(SIGTRAP, NULL);
crmd_fast_exit(exit_code);
}
/* Clean up as much memory as possible for valgrind */
for (gIter = fsa_message_queue; gIter != NULL; gIter = gIter->next) {
fsa_data_t *fsa_data = gIter->data;
crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]",
fsa_input2string(fsa_data->fsa_input),
fsa_state2string(fsa_state),
fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin);
delete_fsa_input(fsa_data);
}
clear_bit(fsa_input_register, R_MEMBERSHIP);
g_list_free(fsa_message_queue); fsa_message_queue = NULL;
metadata_cache_fini();
controld_election_fini();
/* Tear down the CIB manager connection, but don't free it yet -- it could
* be used when we drain the mainloop later.
*/
cib_free_callbacks(fsa_cib_conn);
fsa_cib_conn->cmds->signoff(fsa_cib_conn);
verify_stopped(fsa_state, LOG_WARNING);
clear_bit(fsa_input_register, R_LRM_CONNECTED);
lrm_state_destroy_all();
/* This basically will not work, since mainloop has a reference to it */
mainloop_destroy_trigger(fsa_source); fsa_source = NULL;
mainloop_destroy_trigger(config_read); config_read = NULL;
mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL;
crm_client_cleanup();
crm_peer_destroy();
controld_free_fsa_timers();
te_cleanup_stonith_history_sync(NULL, TRUE);
controld_free_sched_timer();
free(fsa_our_dc_version); fsa_our_dc_version = NULL;
free(fsa_our_uname); fsa_our_uname = NULL;
free(fsa_our_uuid); fsa_our_uuid = NULL;
free(fsa_our_dc); fsa_our_dc = NULL;
free(fsa_cluster_name); fsa_cluster_name = NULL;
free(te_uuid); te_uuid = NULL;
free(failed_stop_offset); failed_stop_offset = NULL;
free(failed_start_offset); failed_start_offset = NULL;
free(max_generation_from); max_generation_from = NULL;
free_xml(max_generation_xml); max_generation_xml = NULL;
mainloop_destroy_signal(SIGPIPE);
mainloop_destroy_signal(SIGUSR1);
mainloop_destroy_signal(SIGTERM);
mainloop_destroy_signal(SIGTRAP);
/* leave SIGCHLD engaged as we might still want to drain some service-actions */
if (mloop) {
GMainContext *ctx = g_main_loop_get_context(crmd_mainloop);
/* Don't re-enter this block */
crmd_mainloop = NULL;
/* no signals on final draining anymore */
mainloop_destroy_signal(SIGCHLD);
crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
{
int lpc = 0;
while((g_main_context_pending(ctx) && lpc < 10)) {
lpc++;
crm_trace("Iteration %d", lpc);
g_main_context_dispatch(ctx);
}
}
crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
g_main_loop_quit(mloop);
/* Won't do anything yet, since we're inside it now */
g_main_loop_unref(mloop);
} else {
mainloop_destroy_signal(SIGCHLD);
}
cib_delete(fsa_cib_conn);
fsa_cib_conn = NULL;
throttle_fini();
/* Graceful */
crm_trace("Done preparing for exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
return exit_code;
}
/* A_EXIT_0, A_EXIT_1 */
void
do_exit(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_exit_t exit_code = CRM_EX_OK;
int log_level = LOG_INFO;
const char *exit_type = "gracefully";
if (action & A_EXIT_1) {
log_level = LOG_ERR;
exit_type = "forcefully";
exit_code = CRM_EX_ERROR;
}
verify_stopped(cur_state, LOG_ERR);
do_crm_log(log_level, "Performing %s - %s exiting the controller",
fsa_action2string(action), exit_type);
crm_info("[%s] stopped (%d)", crm_system_name, exit_code);
crmd_exit(exit_code);
}
static void sigpipe_ignore(int nsig) { return; }
/* A_STARTUP */
void
do_startup(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_debug("Registering Signal Handlers");
mainloop_add_signal(SIGTERM, crm_shutdown);
mainloop_add_signal(SIGPIPE, sigpipe_ignore);
fsa_source = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL);
config_read = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL);
transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL);
crm_debug("Creating CIB manager and executor objects");
fsa_cib_conn = cib_new();
lrm_state_init_local();
if (controld_init_fsa_timers() == FALSE) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static int32_t
crmd_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (crm_client_new(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
static void
crmd_ipc_created(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
}
static int32_t
crmd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
crm_client_t *client = crm_client_get(c);
xmlNode *msg = crm_ipcs_recv(client, data, size, &id, &flags);
crm_trace("Invoked: %s", crm_client_name(client));
crm_ipcs_send_ack(client, id, flags, "ack", __FUNCTION__, __LINE__);
if (msg == NULL) {
return 0;
}
#if ENABLE_ACL
CRM_ASSERT(client->user != NULL);
crm_acl_get_set_user(msg, F_CRM_USER, client->user);
#endif
crm_trace("Processing msg from %s", crm_client_name(client));
crm_log_xml_trace(msg, "controller[inbound]");
crm_xml_add(msg, F_CRM_SYS_FROM, client->id);
if (crmd_authorize_message(msg, client, NULL)) {
route_message(C_IPC_MESSAGE, msg);
}
trigger_fsa(fsa_source);
free_xml(msg);
return 0;
}
static int32_t
crmd_ipc_closed(qb_ipcs_connection_t * c)
{
crm_client_t *client = crm_client_get(c);
if (client) {
crm_trace("Disconnecting %sregistered client %s (%p/%p)",
(client->userdata? "" : "un"), crm_client_name(client),
c, client);
free(client->userdata);
crm_client_destroy(client);
trigger_fsa(fsa_source);
}
return 0;
}
static void
crmd_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
crmd_ipc_closed(c);
}
/* A_STOP */
void
do_stop(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs); ipcs = NULL;
register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
}
/* A_STARTED */
void
do_started(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
static struct qb_ipcs_service_handlers crmd_callbacks = {
.connection_accept = crmd_ipc_accept,
.connection_created = crmd_ipc_created,
.msg_process = crmd_ipc_dispatch,
.connection_closed = crmd_ipc_closed,
.connection_destroyed = crmd_ipc_destroy
};
if (cur_state != S_STARTING) {
crm_err("Start cancelled... %s", fsa_state2string(cur_state));
return;
} else if (is_set(fsa_input_register, R_MEMBERSHIP) == FALSE) {
crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_LRM_CONNECTED) == FALSE) {
crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_CIB_CONNECTED) == FALSE) {
crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_READ_CONFIG) == FALSE) {
crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_PEER_DATA) == FALSE) {
crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA);
crmd_fsa_stall(TRUE);
return;
}
crm_debug("Init server comms");
ipcs = crmd_ipc_server_init(&crmd_callbacks);
if (ipcs == NULL) {
crm_err("Failed to create IPC server: shutting down and inhibiting respawn");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else {
crm_notice("Pacemaker controller successfully started and accepting connections");
}
controld_trigger_fencer_connect();
clear_bit(fsa_input_register, R_STARTING);
register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL);
}
/* A_RECOVER */
void
do_recover(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
set_bit(fsa_input_register, R_IN_RECOVERY);
crm_warn("Fast-tracking shutdown in response to errors");
register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
}
/* *INDENT-OFF* */
static pe_cluster_option crmd_opts[] = {
/* name, old-name, validate, values, default, short description, long description */
{ "dc-version", NULL, "string", NULL, "none", NULL,
"Version of Pacemaker on the cluster's DC.",
"Includes the hash which identifies the exact changeset it was built from. Used for diagnostic purposes."
},
{ "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL,
"The messaging stack on which Pacemaker is currently running.",
"Used for informational and diagnostic purposes." },
{ "cluster-name", NULL, "string", NULL, NULL, NULL,
"An arbitrary name for the cluster",
"This optional value is mostly for users' convenience as desired "
"in administration, but may also be used in Pacemaker configuration "
"rules via the #cluster-name node attribute, and by higher-level tools "
"and resource agents."
},
{ XML_CONFIG_ATTR_DC_DEADTIME, NULL, "time", NULL, "20s", &check_time,
"How long to wait for a response from other nodes during startup.",
"The \"correct\" value will depend on the speed/load of your network and the type of switches used."
},
{ XML_CONFIG_ATTR_RECHECK, NULL, "time",
"Zero disables polling. Positive values are an interval in seconds (unless other SI units are specified. eg. 5min)",
"15min", &check_timer,
"Polling interval for time based changes to options, resource parameters and constraints.",
"The Cluster is primarily event driven, however the configuration can have elements that change based on time."
" To ensure these changes take effect, we can optionally poll the cluster's status for changes."
},
{ "load-threshold", NULL, "percentage", NULL, "80%", &check_utilization,
"The maximum amount of system resources that should be used by nodes in the cluster",
"The cluster will slow down its recovery process when the amount of system resources used"
" (currently CPU) approaches this limit",
},
{ "node-action-limit", NULL, "integer", NULL, "0", &check_number,
"The maximum number of jobs that can be scheduled per node. Defaults to 2x cores"},
{ XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL,
"How a cluster node should react if notified of its own fencing",
"A cluster node may receive notification of its own fencing if fencing "
"is misconfigured, or if fabric fencing is in use that doesn't cut "
"cluster communication. Allowed values are \"stop\" to attempt to "
"immediately stop pacemaker and stay stopped, or \"panic\" to attempt "
"to immediately reboot the local node, falling back to stop on failure."
},
{ XML_CONFIG_ATTR_ELECTION_FAIL, NULL, "time", NULL, "2min", &check_timer,
"*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug."
},
{ XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL, "20min", &check_timer,
"*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug."
},
{
"join-integration-timeout", "crmd-integration-timeout",
"time", NULL, "3min", &check_timer,
"*** Advanced Use Only ***",
"If need to adjust this value, it probably indicates the presence of a bug"
},
{
"join-finalization-timeout", "crmd-finalization-timeout",
"time", NULL, "30min", &check_timer,
"*** Advanced Use Only ***",
"If you need to adjust this value, it probably indicates the presence of a bug"
},
{
"transition-delay", "crmd-transition-delay",
"time", NULL, "0s", &check_timer,
"*** Advanced Use Only *** Enabling this option will slow down cluster recovery under all conditions",
"Delay cluster recovery for the configured interval to allow for additional/related events to occur.\n"
"Useful if your configuration is sensitive to the order in which ping updates arrive."
},
{ "stonith-watchdog-timeout", NULL, "time", NULL, NULL, &check_sbd_timeout,
"How long to wait before we can assume nodes are safely down", NULL
},
{ "stonith-max-attempts",NULL,"integer",NULL,"10",&check_positive_number,
"How many times stonith can fail before it will no longer be attempted on a target"
},
+
+ // Already documented in libpe_status (other values must be kept identical)
{ "no-quorum-policy", NULL, "enum", "stop, freeze, ignore, suicide", "stop", &check_quorum, NULL, NULL },
+ { XML_CONFIG_ATTR_SHUTDOWN_LOCK, NULL, "boolean", NULL, "false", &check_boolean, NULL, NULL },
};
/* *INDENT-ON* */
void
crmd_metadata(void)
{
config_metadata("pacemaker-controld", "1.0",
"controller properties",
"Cluster properties used by Pacemaker's controller,"
" formerly known as crmd",
crmd_opts, DIMOF(crmd_opts));
}
static void
verify_crmd_options(GHashTable * options)
{
verify_all_options(options, crmd_opts, DIMOF(crmd_opts));
}
static const char *
crmd_pref(GHashTable * options, const char *name)
{
return get_cluster_pref(options, crmd_opts, DIMOF(crmd_opts), name);
}
static void
config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
const char *value = NULL;
GHashTable *config_hash = NULL;
crm_time_t *now = crm_time_new(NULL);
xmlNode *crmconfig = NULL;
xmlNode *alerts = NULL;
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc));
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
if (rc == -EACCES || rc == -pcmk_err_schema_validation) {
crm_err("The cluster is mis-configured - shutting down and staying down");
set_bit(fsa_input_register, R_STAYDOWN);
}
goto bail;
}
crmconfig = output;
if ((crmconfig) &&
(crm_element_name(crmconfig)) &&
(strcmp(crm_element_name(crmconfig), XML_CIB_TAG_CRMCONFIG) != 0)) {
crmconfig = first_named_child(crmconfig, XML_CIB_TAG_CRMCONFIG);
}
if (!crmconfig) {
fsa_data_t *msg_data = NULL;
crm_err("Local CIB query for " XML_CIB_TAG_CRMCONFIG " section failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
goto bail;
}
crm_debug("Call %d : Parsing CIB options", call_id);
config_hash = crm_str_table_new();
pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL,
config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL);
verify_crmd_options(config_hash);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_DC_DEADTIME);
election_trigger->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "node-action-limit"); /* Also checks migration-limit */
throttle_update_job_max(value);
value = crmd_pref(config_hash, "load-threshold");
if(value) {
throttle_set_load_target(strtof(value, NULL) / 100.0);
}
value = crmd_pref(config_hash, "no-quorum-policy");
if (safe_str_eq(value, "suicide") && pcmk_locate_sbd()) {
no_quorum_suicide_escalation = TRUE;
}
set_fence_reaction(crmd_pref(config_hash, XML_CONFIG_ATTR_FENCE_REACTION));
value = crmd_pref(config_hash,"stonith-max-attempts");
update_stonith_max_attempts(value);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_FORCE_QUIT);
shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
crm_debug("Shutdown escalation occurs if DC has not responded to request in %ums",
shutdown_escalation_timer->period_ms);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_ELECTION_FAIL);
controld_set_election_period(value);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_RECHECK);
recheck_interval_ms = crm_parse_interval_spec(value);
crm_debug("Re-run scheduler after %dms of inactivity", recheck_interval_ms);
value = crmd_pref(config_hash, "transition-delay");
transition_timer->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "join-integration-timeout");
integration_timer->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "join-finalization-timeout");
finalization_timer->period_ms = crm_parse_interval_spec(value);
+ value = crmd_pref(config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK);
+ controld_shutdown_lock_enabled = crm_is_true(value);
+
free(fsa_cluster_name);
fsa_cluster_name = NULL;
value = g_hash_table_lookup(config_hash, "cluster-name");
if (value) {
fsa_cluster_name = strdup(value);
}
alerts = first_named_child(output, XML_CIB_TAG_ALERTS);
crmd_unpack_alerts(alerts);
set_bit(fsa_input_register, R_READ_CONFIG);
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
g_hash_table_destroy(config_hash);
bail:
crm_time_free(now);
}
gboolean
crm_read_options(gpointer user_data)
{
int call_id =
fsa_cib_conn->cmds->query(fsa_cib_conn,
"//" XML_CIB_TAG_CRMCONFIG " | //" XML_CIB_TAG_ALERTS,
NULL, cib_xpath | cib_scope_local);
fsa_register_cib_callback(call_id, FALSE, NULL, config_query_callback);
crm_trace("Querying the CIB... call %d", call_id);
return TRUE;
}
/* A_READCONFIG */
void
do_read_config(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
throttle_init();
mainloop_set_trigger(config_read);
}
void
crm_shutdown(int nsig)
{
if (crmd_mainloop != NULL && g_main_loop_is_running(crmd_mainloop)) {
if (is_set(fsa_input_register, R_SHUTDOWN)) {
crm_err("Escalating the shutdown");
register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL);
} else {
set_bit(fsa_input_register, R_SHUTDOWN);
register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
if (shutdown_escalation_timer->period_ms == 0) {
const char *value = crmd_pref(NULL, XML_CONFIG_ATTR_FORCE_QUIT);
shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
}
/* can't rely on this... */
crm_notice("Shutting down cluster resource manager " CRM_XS
" limit=%ums", shutdown_escalation_timer->period_ms);
controld_start_timer(shutdown_escalation_timer);
}
} else {
crm_info("exit from shutdown");
crmd_exit(CRM_EX_OK);
}
}
diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c
index 16751b9f48..8d25fb88a1 100644
--- a/daemons/controld/controld_execd.c
+++ b/daemons/controld/controld_execd.c
@@ -1,2799 +1,2843 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include // lrmd_event_data_t, lrmd_rsc_info_t, etc.
#include
#include
#include
#include
#include
#include
#define START_DELAY_THRESHOLD 5 * 60 * 1000
#define MAX_LRM_REG_FAILS 30
struct delete_event_s {
int rc;
const char *rsc;
lrm_state_t *lrm_state;
};
static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id);
static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list);
static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data);
-static int delete_rsc_status(lrm_state_t * lrm_state, const char *rsc_id, int call_options,
- const char *user_name);
static lrmd_event_data_t *construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op,
const char *rsc_id, const char *operation);
static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
const char *operation, xmlNode *msg);
-void send_direct_ack(const char *to_host, const char *to_sys,
- lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *rsc_id);
-
static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state,
int log_level);
-static int do_update_resource(const char *node_name, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op);
+static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc,
+ lrmd_event_data_t *op, time_t lock_time);
static void
lrm_connection_destroy(void)
{
if (is_set(fsa_input_register, R_LRM_CONNECTED)) {
crm_crit("Connection to executor failed");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
clear_bit(fsa_input_register, R_LRM_CONNECTED);
} else {
crm_info("Disconnected from executor");
}
}
static char *
make_stop_id(const char *rsc, int call_id)
{
return crm_strdup_printf("%s:%d", rsc, call_id);
}
static void
copy_instance_keys(gpointer key, gpointer value, gpointer user_data)
{
if (strstr(key, CRM_META "_") == NULL) {
g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
}
}
static void
copy_meta_keys(gpointer key, gpointer value, gpointer user_data)
{
if (strstr(key, CRM_META "_") != NULL) {
g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
}
}
/*!
* \internal
* \brief Remove a recurring operation from a resource's history
*
* \param[in,out] history Resource history to modify
* \param[in] op Operation to remove
*
* \return TRUE if the operation was found and removed, FALSE otherwise
*/
static gboolean
history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op)
{
GList *iter;
for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
lrmd_event_data_t *existing = iter->data;
if ((op->interval_ms == existing->interval_ms)
&& crm_str_eq(op->rsc_id, existing->rsc_id, TRUE)
&& safe_str_eq(op->op_type, existing->op_type)) {
history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter);
lrmd_free_event(existing);
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Free all recurring operations in resource history
*
* \param[in,out] history Resource history to modify
*/
static void
history_free_recurring_ops(rsc_history_t *history)
{
GList *iter;
for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
lrmd_free_event(iter->data);
}
g_list_free(history->recurring_op_list);
history->recurring_op_list = NULL;
}
/*!
* \internal
* \brief Free resource history
*
* \param[in,out] history Resource history to free
*/
void
history_free(gpointer data)
{
rsc_history_t *history = (rsc_history_t*)data;
if (history->stop_params) {
g_hash_table_destroy(history->stop_params);
}
/* Don't need to free history->rsc.id because it's set to history->id */
free(history->rsc.type);
free(history->rsc.standard);
free(history->rsc.provider);
lrmd_free_event(history->failed);
lrmd_free_event(history->last);
free(history->id);
history_free_recurring_ops(history);
free(history);
}
static void
update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op)
{
int target_rc = 0;
rsc_history_t *entry = NULL;
if (op->rsc_deleted) {
crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type);
- delete_rsc_status(lrm_state, op->rsc_id, cib_quorum_override, NULL);
+ controld_delete_resource_history(op->rsc_id, lrm_state->node_name,
+ NULL, crmd_cib_smart_opt());
return;
}
if (safe_str_eq(op->op_type, RSC_NOTIFY)) {
return;
}
crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type);
entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id);
if (entry == NULL && rsc) {
entry = calloc(1, sizeof(rsc_history_t));
entry->id = strdup(op->rsc_id);
g_hash_table_insert(lrm_state->resource_history, entry->id, entry);
entry->rsc.id = entry->id;
entry->rsc.type = strdup(rsc->type);
entry->rsc.standard = strdup(rsc->standard);
if (rsc->provider) {
entry->rsc.provider = strdup(rsc->provider);
} else {
entry->rsc.provider = NULL;
}
} else if (entry == NULL) {
crm_info("Resource %s no longer exists, not updating cache", op->rsc_id);
return;
}
entry->last_callid = op->call_id;
target_rc = rsc_op_expected_rc(op);
if (op->op_status == PCMK_LRM_OP_CANCELLED) {
if (op->interval_ms > 0) {
crm_trace("Removing cancelled recurring op: " CRM_OP_FMT,
op->rsc_id, op->op_type, op->interval_ms);
history_remove_recurring_op(entry, op);
return;
} else {
crm_trace("Skipping " CRM_OP_FMT " rc=%d, status=%d",
op->rsc_id, op->op_type, op->interval_ms, op->rc,
op->op_status);
}
} else if (did_rsc_op_fail(op, target_rc)) {
/* Store failed monitors here, otherwise the block below will cause them
* to be forgotten when a stop happens.
*/
if (entry->failed) {
lrmd_free_event(entry->failed);
}
entry->failed = lrmd_copy_event(op);
} else if (op->interval_ms == 0) {
if (entry->last) {
lrmd_free_event(entry->last);
}
entry->last = lrmd_copy_event(op);
if (op->params &&
(safe_str_eq(CRMD_ACTION_START, op->op_type) ||
safe_str_eq("reload", op->op_type) ||
safe_str_eq(CRMD_ACTION_STATUS, op->op_type))) {
if (entry->stop_params) {
g_hash_table_destroy(entry->stop_params);
}
entry->stop_params = crm_str_table_new();
g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params);
}
}
if (op->interval_ms > 0) {
/* Ensure there are no duplicates */
history_remove_recurring_op(entry, op);
crm_trace("Adding recurring op: " CRM_OP_FMT,
op->rsc_id, op->op_type, op->interval_ms);
entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op));
} else if (entry->recurring_op_list && safe_str_eq(op->op_type, RSC_STATUS) == FALSE) {
crm_trace("Dropping %d recurring ops because of: " CRM_OP_FMT,
g_list_length(entry->recurring_op_list), op->rsc_id,
op->op_type, op->interval_ms);
history_free_recurring_ops(entry);
}
}
/*!
* \internal
* \brief Send a direct OK ack for a resource task
*
* \param[in] lrm_state LRM connection
* \param[in] input Input message being ack'ed
* \param[in] rsc_id ID of affected resource
* \param[in] rsc Affected resource (if available)
* \param[in] task Operation task being ack'ed
* \param[in] ack_host Name of host to send ack to
* \param[in] ack_sys IPC system name to ack
*/
static void
send_task_ok_ack(lrm_state_t *lrm_state, ha_msg_input_t *input,
const char *rsc_id, lrmd_rsc_info_t *rsc, const char *task,
const char *ack_host, const char *ack_sys)
{
lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task);
op->rc = PCMK_OCF_OK;
op->op_status = PCMK_LRM_OP_DONE;
- send_direct_ack(ack_host, ack_sys, rsc, op, rsc_id);
+ controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id);
lrmd_free_event(op);
}
static inline const char *
op_node_name(lrmd_event_data_t *op)
{
return op->remote_nodename? op->remote_nodename : fsa_our_uname;
}
void
lrm_op_callback(lrmd_event_data_t * op)
{
CRM_CHECK(op != NULL, return);
switch (op->type) {
case lrmd_event_disconnect:
if (op->remote_nodename == NULL) {
/* If this is the local executor IPC connection, set the right
* bits in the controller when the connection goes down.
*/
lrm_connection_destroy();
}
break;
case lrmd_event_exec_complete:
{
lrm_state_t *lrm_state = lrm_state_find(op_node_name(op));
CRM_ASSERT(lrm_state != NULL);
process_lrm_event(lrm_state, op, NULL, NULL);
}
break;
default:
break;
}
}
/* A_LRM_CONNECT */
void
do_lrm_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
/* This only pertains to local executor connections. Remote connections are
* handled as resources within the scheduler. Connecting and disconnecting
* from remote executor instances is handled differently.
*/
lrm_state_t *lrm_state = NULL;
if(fsa_our_uname == NULL) {
return; /* Nothing to do */
}
lrm_state = lrm_state_find_or_create(fsa_our_uname);
if (lrm_state == NULL) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (action & A_LRM_DISCONNECT) {
if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) {
if (action == A_LRM_DISCONNECT) {
crmd_fsa_stall(FALSE);
return;
}
}
clear_bit(fsa_input_register, R_LRM_CONNECTED);
crm_info("Disconnecting from the executor");
lrm_state_disconnect(lrm_state);
lrm_state_reset_tables(lrm_state, FALSE);
crm_notice("Disconnected from the executor");
}
if (action & A_LRM_CONNECT) {
int ret = pcmk_ok;
crm_debug("Connecting to the executor");
ret = lrm_state_ipc_connect(lrm_state);
if (ret != pcmk_ok) {
if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) {
crm_warn("Failed to connect to the executor %d time%s (%d max)",
lrm_state->num_lrm_register_fails,
pcmk__plural_s(lrm_state->num_lrm_register_fails),
MAX_LRM_REG_FAILS);
controld_start_timer(wait_timer);
crmd_fsa_stall(FALSE);
return;
}
}
if (ret != pcmk_ok) {
crm_err("Failed to connect to the executor the max allowed %d time%s",
lrm_state->num_lrm_register_fails,
pcmk__plural_s(lrm_state->num_lrm_register_fails));
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
set_bit(fsa_input_register, R_LRM_CONNECTED);
crm_info("Connection to the executor established");
}
if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) {
crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__);
}
}
static gboolean
lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level)
{
int counter = 0;
gboolean rc = TRUE;
const char *when = "lrm disconnect";
GHashTableIter gIter;
const char *key = NULL;
rsc_history_t *entry = NULL;
active_op_t *pending = NULL;
crm_debug("Checking for active resources before exit");
if (cur_state == S_TERMINATE) {
log_level = LOG_ERR;
when = "shutdown";
} else if (is_set(fsa_input_register, R_SHUTDOWN)) {
when = "shutdown... waiting";
}
if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) {
guint removed = g_hash_table_foreach_remove(
lrm_state->pending_ops, stop_recurring_actions, lrm_state);
guint nremaining = g_hash_table_size(lrm_state->pending_ops);
if (removed || nremaining) {
crm_notice("Stopped %u recurring operation%s at %s (%u remaining)",
removed, pcmk__plural_s(removed), when, nremaining);
}
}
if (lrm_state->pending_ops) {
g_hash_table_iter_init(&gIter, lrm_state->pending_ops);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) {
/* Ignore recurring actions in the shutdown calculations */
if (pending->interval_ms == 0) {
counter++;
}
}
}
if (counter > 0) {
do_crm_log(log_level, "%d pending executor operation%s at %s",
counter, pcmk__plural_s(counter), when);
if (cur_state == S_TERMINATE || !is_set(fsa_input_register, R_SENT_RSC_STOP)) {
g_hash_table_iter_init(&gIter, lrm_state->pending_ops);
while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) {
do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key);
}
} else {
rc = FALSE;
}
return rc;
}
if (lrm_state->resource_history == NULL) {
return rc;
}
if (is_set(fsa_input_register, R_SHUTDOWN)) {
/* At this point we're not waiting, we're just shutting down */
when = "shutdown";
}
counter = 0;
g_hash_table_iter_init(&gIter, lrm_state->resource_history);
while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) {
if (is_rsc_active(lrm_state, entry->id) == FALSE) {
continue;
}
counter++;
if (log_level == LOG_ERR) {
crm_info("Found %s active at %s", entry->id, when);
} else {
crm_trace("Found %s active at %s", entry->id, when);
}
if (lrm_state->pending_ops) {
GHashTableIter hIter;
g_hash_table_iter_init(&hIter, lrm_state->pending_ops);
while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) {
if (crm_str_eq(entry->id, pending->rsc_id, TRUE)) {
crm_notice("%sction %s (%s) incomplete at %s",
pending->interval_ms == 0 ? "A" : "Recurring a",
key, pending->op_key, when);
}
}
}
}
if (counter) {
crm_err("%d resource%s active at %s",
counter, (counter == 1)? " was" : "s were", when);
}
return rc;
}
static char *
build_parameter_list(const lrmd_event_data_t *op,
const struct ra_metadata_s *metadata,
xmlNode *result, enum ra_param_flags_e param_type,
bool invert_for_xml)
{
int len = 0;
int max = 0;
char *list = NULL;
GList *iter = NULL;
/* Newer resource agents support the "private" parameter attribute to
* indicate sensitive parameters. For backward compatibility with older
* agents, this list is used if the agent doesn't specify any as "private".
*/
const char *secure_terms[] = {
"password",
"passwd",
"user",
};
if (is_not_set(metadata->ra_flags, ra_uses_private)
&& (param_type == ra_param_private)) {
max = DIMOF(secure_terms);
}
for (iter = metadata->ra_params; iter != NULL; iter = iter->next) {
struct ra_param_s *param = (struct ra_param_s *) iter->data;
bool accept = FALSE;
if (is_set(param->rap_flags, param_type)) {
accept = TRUE;
} else if (max) {
for (int lpc = 0; lpc < max; lpc++) {
if (safe_str_eq(secure_terms[lpc], param->rap_name)) {
accept = TRUE;
break;
}
}
}
if (accept) {
int start = len;
crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type));
len += strlen(param->rap_name) + 2; // include spaces around
list = realloc_safe(list, len + 1); // include null terminator
// spaces before and after make parsing simpler
sprintf(list + start, " %s ", param->rap_name);
} else {
crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type));
}
if (result && (invert_for_xml? !accept : accept)) {
const char *v = g_hash_table_lookup(op->params, param->rap_name);
if (v != NULL) {
crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v);
crm_xml_add(result, param->rap_name, v);
}
}
}
return list;
}
static void
append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
char *list = NULL;
char *digest = NULL;
xmlNode *restart = NULL;
CRM_LOG_ASSERT(op->params != NULL);
if (op->interval_ms > 0) {
/* monitors are not reloadable */
return;
}
if (is_set(metadata->ra_flags, ra_supports_reload)) {
restart = create_xml_node(NULL, XML_TAG_PARAMS);
/* Add any parameters with unique="1" to the "op-force-restart" list.
*
* (Currently, we abuse "unique=0" to indicate reloadability. This is
* nonstandard and should eventually be replaced once the OCF standard
* is updated with something better.)
*/
list = build_parameter_list(op, metadata, restart, ra_param_unique,
FALSE);
} else {
/* Resource does not support reloads */
return;
}
digest = calculate_operation_digest(restart, version);
/* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload,
* no matter if it actually supports any parameters with unique="1"). */
crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: "");
crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest);
crm_trace("%s: %s, %s", op->rsc_id, digest, list);
crm_log_xml_trace(restart, "restart digest source");
free_xml(restart);
free(digest);
free(list);
}
static void
append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
char *list = NULL;
char *digest = NULL;
xmlNode *secure = NULL;
CRM_LOG_ASSERT(op->params != NULL);
/*
* To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the
* secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on
* the insecure ones
*/
secure = create_xml_node(NULL, XML_TAG_PARAMS);
list = build_parameter_list(op, metadata, secure, ra_param_private, TRUE);
if (list != NULL) {
digest = calculate_operation_digest(secure, version);
crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list);
crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest);
crm_trace("%s: %s, %s", op->rsc_id, digest, list);
crm_log_xml_trace(secure, "secure digest source");
} else {
crm_trace("%s: no secure parameters", op->rsc_id);
}
free_xml(secure);
free(digest);
free(list);
}
static gboolean
build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op,
const char *node_name, const char *src)
{
int target_rc = 0;
xmlNode *xml_op = NULL;
struct ra_metadata_s *metadata = NULL;
const char *caller_version = NULL;
lrm_state_t *lrm_state = NULL;
if (op == NULL) {
return FALSE;
}
target_rc = rsc_op_expected_rc(op);
/* there is a small risk in formerly mixed clusters that it will
* be sub-optimal.
*
* however with our upgrade policy, the update we send should
* still be completely supported anyway
*/
caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION);
CRM_LOG_ASSERT(caller_version != NULL);
if(caller_version == NULL) {
caller_version = CRM_FEATURE_SET;
}
crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version);
xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc,
fsa_our_uname, src, LOG_DEBUG);
if (xml_op == NULL) {
return TRUE;
}
if ((rsc == NULL) || (op->params == NULL)
|| !crm_op_needs_metadata(rsc->standard, op->op_type)) {
crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)",
op->op_type, op->rsc_id, op->params, rsc);
return TRUE;
}
lrm_state = lrm_state_find(node_name);
if (lrm_state == NULL) {
crm_warn("Cannot calculate digests for operation " CRM_OP_FMT
" because we have no connection to executor for %s",
op->rsc_id, op->op_type, op->interval_ms, node_name);
return TRUE;
}
metadata = metadata_cache_get(lrm_state->metadata_cache, rsc);
if (metadata == NULL) {
/* For now, we always collect resource agent meta-data via a local,
* synchronous, direct execution of the agent. This has multiple issues:
* the executor should execute agents, not the controller; meta-data for
* Pacemaker Remote nodes should be collected on those nodes, not
* locally; and the meta-data call shouldn't eat into the timeout of the
* real action being performed.
*
* These issues are planned to be addressed by having the scheduler
* schedule a meta-data cache check at the beginning of each transition.
* Once that is working, this block will only be a fallback in case the
* initial collection fails.
*/
char *metadata_str = NULL;
int rc = lrm_state_get_metadata(lrm_state, rsc->standard,
rsc->provider, rsc->type,
&metadata_str, 0);
if (rc != pcmk_ok) {
crm_warn("Failed to get metadata for %s (%s:%s:%s)",
rsc->id, rsc->standard, rsc->provider, rsc->type);
return TRUE;
}
metadata = metadata_cache_update(lrm_state->metadata_cache, rsc,
metadata_str);
free(metadata_str);
if (metadata == NULL) {
crm_warn("Failed to update metadata for %s (%s:%s:%s)",
rsc->id, rsc->standard, rsc->provider, rsc->type);
return TRUE;
}
}
#if ENABLE_VERSIONED_ATTRS
crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version);
#endif
crm_trace("Including additional digests for %s::%s:%s", rsc->standard, rsc->provider, rsc->type);
append_restart_list(op, metadata, xml_op, caller_version);
append_secure_list(op, metadata, xml_op, caller_version);
return TRUE;
}
static gboolean
is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id)
{
rsc_history_t *entry = NULL;
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
if (entry == NULL || entry->last == NULL) {
return FALSE;
}
crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type,
entry->last->interval_ms, entry->last->rc);
if (entry->last->rc == PCMK_OCF_OK && safe_str_eq(entry->last->op_type, CRMD_ACTION_STOP)) {
return FALSE;
} else if (entry->last->rc == PCMK_OCF_OK
&& safe_str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE)) {
// A stricter check is too complex ... leave that to the scheduler
return FALSE;
} else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) {
return FALSE;
} else if ((entry->last->interval_ms == 0)
&& (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) {
/* Badly configured resources can't be reliably stopped */
return FALSE;
}
return TRUE;
}
static gboolean
build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list)
{
GHashTableIter iter;
rsc_history_t *entry = NULL;
g_hash_table_iter_init(&iter, lrm_state->resource_history);
while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) {
GList *gIter = NULL;
xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE);
crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id);
crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type);
crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard);
crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider);
if (entry->last && entry->last->params) {
const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER);
if (container) {
crm_trace("Resource %s is a part of container resource %s", entry->id, container);
crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container);
}
}
build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name, __FUNCTION__);
build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name, __FUNCTION__);
for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) {
build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name, __FUNCTION__);
}
}
return FALSE;
}
static xmlNode *
do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags)
{
xmlNode *xml_state = NULL;
xmlNode *xml_data = NULL;
xmlNode *rsc_list = NULL;
crm_node_t *peer = NULL;
peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY);
CRM_CHECK(peer != NULL, return NULL);
xml_state = create_node_state_update(peer, update_flags, NULL,
__FUNCTION__);
if (xml_state == NULL) {
return NULL;
}
xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM);
crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid);
rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES);
/* Build a list of active (not always running) resources */
build_active_RAs(lrm_state, rsc_list);
crm_log_xml_trace(xml_state, "Current executor state");
return xml_state;
}
xmlNode *
controld_query_executor_state(const char *node_name)
{
lrm_state_t *lrm_state = lrm_state_find(node_name);
if (!lrm_state) {
crm_err("Could not find executor state for node %s", node_name);
return NULL;
}
return do_lrm_query_internal(lrm_state,
node_update_cluster|node_update_peer);
}
+/*!
+ * \internal
+ * \brief Map standard Pacemaker return code to operation status and OCF code
+ *
+ * \param[out] event Executor event whose status and return code should be set
+ * \param[in] rc Standard Pacemaker return code
+ */
+void
+controld_rc2event(lrmd_event_data_t *event, int rc)
+{
+ switch (rc) {
+ case pcmk_rc_ok:
+ event->rc = PCMK_OCF_OK;
+ event->op_status = PCMK_LRM_OP_DONE;
+ break;
+ case EACCES:
+ event->rc = PCMK_OCF_INSUFFICIENT_PRIV;
+ event->op_status = PCMK_LRM_OP_ERROR;
+ break;
+ default:
+ event->rc = PCMK_OCF_UNKNOWN_ERROR;
+ event->op_status = PCMK_LRM_OP_ERROR;
+ break;
+ }
+}
+
+/*!
+ * \internal
+ * \brief Trigger a new transition after CIB status was deleted
+ *
+ * If a CIB status delete was not expected (as part of the transition graph),
+ * trigger a new transition by updating the (arbitrary) "last-lrm-refresh"
+ * cluster property.
+ *
+ * \param[in] from_sys IPC name that requested the delete
+ * \param[in] rsc_id Resource whose status was deleted (for logging only)
+ */
+void
+controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id)
+{
+ if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) {
+ char *now_s = crm_strdup_printf("%lld", (long long) time(NULL));
+
+ crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id);
+ update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG,
+ NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s,
+ FALSE, NULL, NULL);
+ free(now_s);
+ }
+}
+
static void
notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc)
{
lrmd_event_data_t *op = NULL;
const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
crm_info("Notifying %s on %s that %s was%s deleted",
from_sys, (from_host? from_host : "localhost"), rsc_id,
((rc == pcmk_ok)? "" : " not"));
-
op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE);
-
- if (rc == pcmk_ok) {
- op->op_status = PCMK_LRM_OP_DONE;
- op->rc = PCMK_OCF_OK;
- } else {
- op->op_status = PCMK_LRM_OP_ERROR;
- op->rc = PCMK_OCF_UNKNOWN_ERROR;
- }
-
- send_direct_ack(from_host, from_sys, NULL, op, rsc_id);
+ controld_rc2event(op, pcmk_legacy2rc(rc));
+ controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id);
lrmd_free_event(op);
-
- if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) {
- /* this isn't expected - trigger a new transition */
- time_t now = time(NULL);
- char *now_s = crm_itoa(now);
-
- crm_debug("Triggering a refresh after %s deleted %s from the executor",
- from_sys, rsc_id);
-
- update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL,
- "last-lrm-refresh", now_s, FALSE, NULL, NULL);
-
- free(now_s);
- }
+ controld_trigger_delete_refresh(from_sys, rsc_id);
}
static gboolean
lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data)
{
struct delete_event_s *event = user_data;
struct pending_deletion_op_s *op = value;
if (crm_str_eq(event->rsc, op->rsc, TRUE)) {
notify_deleted(event->lrm_state, op->input, event->rsc, event->rc);
return TRUE;
}
return FALSE;
}
static gboolean
lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data)
{
const char *rsc = user_data;
active_op_t *pending = value;
if (crm_str_eq(rsc, pending->rsc_id, TRUE)) {
crm_info("Removing op %s:%d for deleted resource %s",
pending->op_key, pending->call_id, rsc);
return TRUE;
}
return FALSE;
}
-/*
- * Remove the rsc from the CIB
- *
- * Avoids refreshing the entire LRM section of this host
- */
-#define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']"
-
-static int
-delete_rsc_status(lrm_state_t * lrm_state, const char *rsc_id, int call_options,
- const char *user_name)
-{
- char *rsc_xpath = NULL;
- int rc = pcmk_ok;
-
- CRM_CHECK(rsc_id != NULL, return -ENXIO);
-
- rsc_xpath = crm_strdup_printf(RSC_TEMPLATE, lrm_state->node_name, rsc_id);
-
- rc = cib_internal_op(fsa_cib_conn, CIB_OP_DELETE, NULL, rsc_xpath,
- NULL, NULL, call_options | cib_xpath, user_name);
-
- free(rsc_xpath);
- return rc;
-}
-
static void
delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id,
GHashTableIter * rsc_gIter, int rc, const char *user_name)
{
struct delete_event_s event;
CRM_CHECK(rsc_id != NULL, return);
if (rc == pcmk_ok) {
char *rsc_id_copy = strdup(rsc_id);
if (rsc_gIter)
g_hash_table_iter_remove(rsc_gIter);
else
g_hash_table_remove(lrm_state->resource_history, rsc_id_copy);
crm_debug("sync: Sending delete op for %s", rsc_id_copy);
- delete_rsc_status(lrm_state, rsc_id_copy, cib_quorum_override, user_name);
+ controld_delete_resource_history(rsc_id_copy, lrm_state->node_name,
+ user_name, crmd_cib_smart_opt());
g_hash_table_foreach_remove(lrm_state->pending_ops, lrm_remove_deleted_op, rsc_id_copy);
free(rsc_id_copy);
}
if (input) {
notify_deleted(lrm_state, input, rsc_id, rc);
}
event.rc = rc;
event.rsc = rsc_id;
event.lrm_state = lrm_state;
g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event);
}
/*!
* \internal
* \brief Erase an LRM history entry from the CIB, given the operation data
*
* \param[in] lrm_state LRM state of the desired node
* \param[in] op Operation whose history should be deleted
*/
static void
erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op)
{
xmlNode *xml_top = NULL;
CRM_CHECK(op != NULL, return);
xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP);
crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id);
crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data);
if (op->interval_ms > 0) {
char *op_id = generate_op_key(op->rsc_id, op->op_type, op->interval_ms);
/* Avoid deleting last_failure too (if it was a result of this recurring op failing) */
crm_xml_add(xml_top, XML_ATTR_ID, op_id);
free(op_id);
}
crm_debug("Erasing resource operation history for " CRM_OP_FMT " (call=%d)",
op->rsc_id, op->op_type, op->interval_ms, op->call_id);
fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top,
cib_quorum_override);
crm_log_xml_trace(xml_top, "op:cancel");
free_xml(xml_top);
}
/* Define xpath to find LRM resource history entry by node and resource */
#define XPATH_HISTORY \
"/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \
"/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \
"/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']" \
"/" XML_LRM_TAG_RSC_OP
/* ... and also by operation key */
#define XPATH_HISTORY_ID XPATH_HISTORY \
"[@" XML_ATTR_ID "='%s']"
/* ... and also by operation key and operation call ID */
#define XPATH_HISTORY_CALL XPATH_HISTORY \
"[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']"
/* ... and also by operation key and original operation key */
#define XPATH_HISTORY_ORIG XPATH_HISTORY \
"[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']"
/*!
* \internal
* \brief Erase an LRM history entry from the CIB, given operation identifiers
*
* \param[in] lrm_state LRM state of the node to clear history for
* \param[in] rsc_id Name of resource to clear history for
* \param[in] key Operation key of operation to clear history for
* \param[in] orig_op If specified, delete only if it has this original op
* \param[in] call_id If specified, delete entry only if it has this call ID
*/
static void
erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id,
const char *key, const char *orig_op, int call_id)
{
char *op_xpath = NULL;
CRM_CHECK((rsc_id != NULL) && (key != NULL), return);
if (call_id > 0) {
op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL,
lrm_state->node_name, rsc_id, key,
call_id);
} else if (orig_op) {
op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG,
lrm_state->node_name, rsc_id, key,
orig_op);
} else {
op_xpath = crm_strdup_printf(XPATH_HISTORY_ID,
lrm_state->node_name, rsc_id, key);
}
crm_debug("Erasing resource operation history for %s on %s (call=%d)",
key, rsc_id, call_id);
fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL,
cib_quorum_override | cib_xpath);
free(op_xpath);
}
static inline gboolean
last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms)
{
if (entry == NULL) {
return FALSE;
}
if (op == NULL) {
return TRUE;
}
return (safe_str_eq(op, entry->failed->op_type)
&& (interval_ms == entry->failed->interval_ms));
}
/*!
* \internal
* \brief Clear a resource's last failure
*
* Erase a resource's last failure on a particular node from both the
* LRM resource history in the CIB, and the resource history remembered
* for the LRM state.
*
* \param[in] rsc_id Resource name
* \param[in] node_name Node name
* \param[in] operation If specified, only clear if matching this operation
* \param[in] interval_ms If operation is specified, it has this interval
*/
void
lrm_clear_last_failure(const char *rsc_id, const char *node_name,
const char *operation, guint interval_ms)
{
char *op_key = NULL;
char *orig_op_key = NULL;
lrm_state_t *lrm_state = NULL;
lrm_state = lrm_state_find(node_name);
if (lrm_state == NULL) {
return;
}
/* Erase from CIB */
op_key = generate_op_key(rsc_id, "last_failure", 0);
if (operation) {
orig_op_key = generate_op_key(rsc_id, operation, interval_ms);
}
erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0);
free(op_key);
free(orig_op_key);
/* Remove from memory */
if (lrm_state->resource_history) {
rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history,
rsc_id);
if (last_failed_matches_op(entry, operation, interval_ms)) {
lrmd_free_event(entry->failed);
entry->failed = NULL;
}
}
}
/* Returns: gboolean - cancellation is in progress */
static gboolean
cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove)
{
int rc = pcmk_ok;
char *local_key = NULL;
active_op_t *pending = NULL;
CRM_CHECK(op != 0, return FALSE);
CRM_CHECK(rsc_id != NULL, return FALSE);
if (key == NULL) {
local_key = make_stop_id(rsc_id, op);
key = local_key;
}
pending = g_hash_table_lookup(lrm_state->pending_ops, key);
if (pending) {
if (remove && is_not_set(pending->flags, active_op_remove)) {
set_bit(pending->flags, active_op_remove);
crm_debug("Scheduling %s for removal", key);
}
if (is_set(pending->flags, active_op_cancelled)) {
crm_debug("Operation %s already cancelled", key);
free(local_key);
return FALSE;
}
set_bit(pending->flags, active_op_cancelled);
} else {
crm_info("No pending op found for %s", key);
free(local_key);
return FALSE;
}
crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key);
rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type,
pending->interval_ms);
if (rc == pcmk_ok) {
crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key);
free(local_key);
return TRUE;
}
crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key);
/* The caller needs to make sure the entry is
* removed from the pending_ops list
*
* Usually by returning TRUE inside the worker function
* supplied to g_hash_table_foreach_remove()
*
* Not removing the entry from pending_ops will block
* the node from shutting down
*/
free(local_key);
return FALSE;
}
struct cancel_data {
gboolean done;
gboolean remove;
const char *key;
lrmd_rsc_info_t *rsc;
lrm_state_t *lrm_state;
};
static gboolean
cancel_action_by_key(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
struct cancel_data *data = user_data;
active_op_t *op = value;
if (crm_str_eq(op->op_key, data->key, TRUE)) {
data->done = TRUE;
remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove);
}
return remove;
}
static gboolean
cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove)
{
guint removed = 0;
struct cancel_data data;
CRM_CHECK(rsc != NULL, return FALSE);
CRM_CHECK(key != NULL, return FALSE);
data.key = key;
data.rsc = rsc;
data.done = FALSE;
data.remove = remove;
data.lrm_state = lrm_state;
removed = g_hash_table_foreach_remove(lrm_state->pending_ops, cancel_action_by_key, &data);
crm_trace("Removed %u op cache entries, new size: %u",
removed, g_hash_table_size(lrm_state->pending_ops));
return data.done;
}
/*!
* \internal
* \brief Retrieve resource information from LRM
*
* \param[in] lrm_state LRM connection to use
* \param[in] rsc_xml XML containing resource configuration
* \param[in] do_create If true, register resource with LRM if not already
* \param[out] rsc_info Where to store resource information obtained from LRM
*
* \retval pcmk_ok Success (and rsc_info holds newly allocated result)
* \retval -EINVAL Required information is missing from arguments
* \retval -ENOTCONN No active connection to LRM
* \retval -ENODEV Resource not found
* \retval -errno Error communicating with executor when registering resource
*
* \note Caller is responsible for freeing result on success.
*/
static int
get_lrm_resource(lrm_state_t *lrm_state, xmlNode *rsc_xml, gboolean do_create,
lrmd_rsc_info_t **rsc_info)
{
const char *id = ID(rsc_xml);
CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL);
CRM_CHECK(id, return -EINVAL);
if (lrm_state_is_connected(lrm_state) == FALSE) {
return -ENOTCONN;
}
crm_trace("Retrieving resource information for %s from the executor", id);
*rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
// If resource isn't known by ID, try clone name, if provided
if (!*rsc_info) {
const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG);
if (long_id) {
*rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0);
}
}
if ((*rsc_info == NULL) && do_create) {
const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS);
const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER);
const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE);
int rc;
crm_trace("Registering resource %s with the executor", id);
rc = lrm_state_register_rsc(lrm_state, id, class, provider, type,
lrmd_opt_drop_recurring);
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
crm_err("Could not register resource %s with the executor on %s: %s "
CRM_XS " rc=%d",
id, lrm_state->node_name, pcmk_strerror(rc), rc);
/* Register this as an internal error if this involves the local
* executor. Otherwise, we're likely dealing with an unresponsive
* remote node, which is not an FSA failure.
*/
if (lrm_state_is_local(lrm_state) == TRUE) {
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
return rc;
}
*rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
}
return *rsc_info? pcmk_ok : -ENODEV;
}
static void
delete_resource(lrm_state_t * lrm_state,
const char *id,
lrmd_rsc_info_t * rsc,
GHashTableIter * gIter,
const char *sys,
const char *host,
const char *user,
ha_msg_input_t * request,
gboolean unregister)
{
int rc = pcmk_ok;
crm_info("Removing resource %s for %s (%s) on %s", id, sys, user ? user : "internal", host);
if (rsc && unregister) {
rc = lrm_state_unregister_rsc(lrm_state, id, 0);
}
if (rc == pcmk_ok) {
crm_trace("Resource '%s' deleted", id);
} else if (rc == -EINPROGRESS) {
crm_info("Deletion of resource '%s' pending", id);
if (request) {
struct pending_deletion_op_s *op = NULL;
char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE);
op = calloc(1, sizeof(struct pending_deletion_op_s));
op->rsc = strdup(rsc->id);
op->input = copy_ha_msg_input(request);
g_hash_table_insert(lrm_state->deletion_ops, ref, op);
}
return;
} else {
crm_warn("Deletion of resource '%s' for %s (%s) on %s failed: %d",
id, sys, user ? user : "internal", host, rc);
}
delete_rsc_entry(lrm_state, request, id, gIter, rc, user);
}
static int
get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id)
{
int call_id = 999999999;
rsc_history_t *entry = NULL;
if(lrm_state) {
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
}
/* Make sure the call id is greater than the last successful operation,
* otherwise the failure will not result in a possible recovery of the resource
* as it could appear the failure occurred before the successful start */
if (entry) {
call_id = entry->last_callid + 1;
}
if (call_id < 0) {
call_id = 1;
}
return call_id;
}
static void
fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status,
enum ocf_exitcode op_exitcode)
{
op->call_id = get_fake_call_id(lrm_state, op->rsc_id);
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
op->op_status = op_status;
op->rc = op_exitcode;
}
static void
force_reprobe(lrm_state_t *lrm_state, const char *from_sys,
const char *from_host, const char *user_name,
gboolean is_remote_node)
{
GHashTableIter gIter;
rsc_history_t *entry = NULL;
crm_info("Clearing resource history on node %s", lrm_state->node_name);
g_hash_table_iter_init(&gIter, lrm_state->resource_history);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) {
/* only unregister the resource during a reprobe if it is not a remote connection
* resource. otherwise unregistering the connection will terminate remote-node
* membership */
gboolean unregister = TRUE;
if (is_remote_lrmd_ra(NULL, NULL, entry->id)) {
lrm_state_t *remote_lrm_state = lrm_state_find(entry->id);
if (remote_lrm_state) {
/* when forcing a reprobe, make sure to clear remote node before
* clearing the remote node's connection resource */
force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE);
}
unregister = FALSE;
}
delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, from_host,
user_name, NULL, unregister);
}
/* Now delete the copy in the CIB */
controld_delete_node_state(lrm_state->node_name, controld_section_lrm,
cib_scope_local);
/* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE
* would result in the scheduler sending us back here again
*/
update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node);
}
/*!
* \internal
* \brief Fail a requested action without actually executing it
*
* For an action that can't be executed, process it similarly to an actual
* execution result, with specified error status (except for notify actions,
* which will always be treated as successful).
*
* \param[in] lrm_state Executor connection that action is for
* \param[in] action Action XML from request
* \param[in] rc Desired return code to use
* \param[in] op_status Desired operation status to use
*/
static void
synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action,
int op_status, enum ocf_exitcode rc)
{
lrmd_event_data_t *op = NULL;
const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK);
const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET);
xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE);
if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) {
/* @TODO Should we do something else, like direct ack? */
crm_info("Can't fake %s failure (%d) on %s without resource configuration",
crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc,
target_node);
return;
} else if(operation == NULL) {
/* This probably came from crm_resource -C, nothing to do */
crm_info("Can't fake %s failure (%d) on %s without operation",
ID(xml_rsc), rc, target_node);
return;
}
op = construct_op(lrm_state, action, ID(xml_rsc), operation);
if (safe_str_eq(operation, RSC_NOTIFY)) { // Notifications can't fail
fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_OK);
} else {
fake_op_status(lrm_state, op, op_status, rc);
}
crm_info("Faking " CRM_OP_FMT " result (%d) on %s",
op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node);
// Process the result as if it came from the LRM
process_lrm_event(lrm_state, op, NULL, action);
lrmd_free_event(op);
}
/*!
* \internal
* \brief Get target of an LRM operation
*
* \param[in] xml LRM operation data XML
*
* \return LRM operation target node name (local node or Pacemaker Remote node)
*/
static const char *
lrm_op_target(xmlNode *xml)
{
const char *target = NULL;
if (xml) {
target = crm_element_value(xml, XML_LRM_ATTR_TARGET);
}
if (target == NULL) {
target = fsa_our_uname;
}
return target;
}
static void
fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name,
const char *from_host, const char *from_sys)
{
lrmd_event_data_t *op = NULL;
lrmd_rsc_info_t *rsc = NULL;
xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE);
CRM_CHECK(xml_rsc != NULL, return);
/* The executor simply executes operations and reports the results, without
* any concept of success or failure, so to fail a resource, we must fake
* what a failure looks like.
*
* To do this, we create a fake executor operation event for the resource,
* and pass that event to the executor client callback so it will be
* processed as if it came from the executor.
*/
op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon");
fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR);
free((char*) op->user_data);
op->user_data = NULL;
op->interval_ms = 0;
#if ENABLE_ACL
if (user_name && is_privileged(user_name) == FALSE) {
crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc));
- send_direct_ack(from_host, from_sys, NULL, op, ID(xml_rsc));
+ controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
lrmd_free_event(op);
return;
}
#endif
if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) {
crm_info("Failing resource %s...", rsc->id);
op->exit_reason = strdup("Simulated failure");
process_lrm_event(lrm_state, op, NULL, xml);
op->op_status = PCMK_LRM_OP_DONE;
op->rc = PCMK_OCF_OK;
lrmd_free_rsc_info(rsc);
} else {
crm_info("Cannot find/create resource in order to fail it...");
crm_log_xml_warn(xml, "bad input");
}
- send_direct_ack(from_host, from_sys, NULL, op, ID(xml_rsc));
+ controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
lrmd_free_event(op);
}
static void
handle_refresh_op(lrm_state_t *lrm_state, const char *user_name,
const char *from_host, const char *from_sys)
{
int rc = pcmk_ok;
xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all);
fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name);
crm_info("Forced a local resource history refresh: call=%d", rc);
if (safe_str_neq(CRM_SYSTEM_CRMD, from_sys)) {
xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host,
from_sys, CRM_SYSTEM_LRMD,
fsa_our_uuid);
crm_debug("ACK'ing refresh from %s (%s)", from_sys, from_host);
if (relay_message(reply, TRUE) == FALSE) {
crm_log_xml_err(reply, "Unable to route reply");
}
free_xml(reply);
}
free_xml(fragment);
}
static void
handle_query_op(xmlNode *msg, lrm_state_t *lrm_state)
{
xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all);
xmlNode *reply = create_reply(msg, data);
if (relay_message(reply, TRUE) == FALSE) {
crm_err("Unable to route reply");
crm_log_xml_err(reply, "reply");
}
free_xml(reply);
free_xml(data);
}
static void
handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys,
const char *from_host, const char *user_name,
gboolean is_remote_node)
{
crm_notice("Forcing the status of all resources to be redetected");
force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node);
if (safe_str_neq(CRM_SYSTEM_PENGINE, from_sys)
&& safe_str_neq(CRM_SYSTEM_TENGINE, from_sys)) {
xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host,
from_sys, CRM_SYSTEM_LRMD,
fsa_our_uuid);
crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host);
if (relay_message(reply, TRUE) == FALSE) {
crm_log_xml_err(reply, "Unable to route reply");
}
free_xml(reply);
}
}
static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state,
lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys)
{
char *op_key = NULL;
char *meta_key = NULL;
int call = 0;
const char *call_id = NULL;
const char *op_task = NULL;
const char *interval_ms_s = NULL;
gboolean in_progress = FALSE;
xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE);
CRM_CHECK(params != NULL, return FALSE);
meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS);
interval_ms_s = crm_element_value(params, meta_key);
free(meta_key);
CRM_CHECK(interval_ms_s != NULL, return FALSE);
meta_key = crm_meta_name(XML_LRM_ATTR_TASK);
op_task = crm_element_value(params, meta_key);
free(meta_key);
CRM_CHECK(op_task != NULL, return FALSE);
meta_key = crm_meta_name(XML_LRM_ATTR_CALLID);
call_id = crm_element_value(params, meta_key);
free(meta_key);
op_key = generate_op_key(rsc->id, op_task, crm_parse_ms(interval_ms_s));
crm_debug("Scheduler requested op %s (call=%s) be cancelled",
op_key, (call_id? call_id : "NA"));
call = crm_parse_int(call_id, "0");
if (call == 0) {
// Normal case when the scheduler cancels a recurring op
in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE);
} else {
// Normal case when the scheduler cancels an orphan op
in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE);
}
// Acknowledge cancellation operation if for a remote connection resource
if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
char *op_id = make_stop_id(rsc->id, call);
if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) {
crm_info("Nothing known about operation %d for %s", call, op_key);
}
erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, call);
send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
from_host, from_sys);
/* needed at least for cancellation of a remote operation */
g_hash_table_remove(lrm_state->pending_ops, op_id);
free(op_id);
} else {
/* No ack is needed since abcdaa8, but peers with older versions
* in a rolling upgrade need one. We didn't bump the feature set
* at that commit, so we can only compare against the previous
* CRM version (3.0.8). If any peers have feature set 3.0.9 but
* not abcdaa8, they will time out waiting for the ack (no
* released versions of Pacemaker are affected).
*/
const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION);
if (compare_version(peer_version, "3.0.8") <= 0) {
crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)",
op_key, from_host, peer_version);
send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
from_host, from_sys);
}
}
free(op_key);
return TRUE;
}
static void
do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state,
lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host,
bool crm_rsc_delete, const char *user_name)
{
gboolean unregister = TRUE;
#if ENABLE_ACL
- int cib_rc = delete_rsc_status(lrm_state, rsc->id,
- cib_dryrun|cib_sync_call, user_name);
+ int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name,
+ user_name,
+ cib_dryrun|cib_sync_call);
- if (cib_rc != pcmk_ok) {
+ if (cib_rc != pcmk_rc_ok) {
lrmd_event_data_t *op = NULL;
- crm_err("Could not delete resource status of %s for %s (user %s) on %s: %s"
- CRM_XS " rc=%d",
- rsc->id, from_sys, (user_name? user_name : "unknown"),
- from_host, pcmk_strerror(cib_rc), cib_rc);
-
op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE);
op->op_status = PCMK_LRM_OP_ERROR;
- if (cib_rc == -EACCES) {
+ if (cib_rc == EACCES) {
op->rc = PCMK_OCF_INSUFFICIENT_PRIV;
} else {
op->rc = PCMK_OCF_UNKNOWN_ERROR;
}
- send_direct_ack(from_host, from_sys, NULL, op, rsc->id);
+ controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id);
lrmd_free_event(op);
return;
}
#endif
if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
unregister = FALSE;
}
delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, from_host,
user_name, input, unregister);
}
/* A_LRM_INVOKE */
void
do_lrm_invoke(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
lrm_state_t *lrm_state = NULL;
const char *crm_op = NULL;
const char *from_sys = NULL;
const char *from_host = NULL;
const char *operation = NULL;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
const char *user_name = NULL;
const char *target_node = NULL;
gboolean is_remote_node = FALSE;
bool crm_rsc_delete = FALSE;
target_node = lrm_op_target(input->xml);
is_remote_node = safe_str_neq(target_node, fsa_our_uname);
lrm_state = lrm_state_find(target_node);
if ((lrm_state == NULL) && is_remote_node) {
crm_err("Failing action because local node has never had connection to remote node %s",
target_node);
synthesize_lrmd_failure(NULL, input->xml, PCMK_LRM_OP_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR);
return;
}
CRM_ASSERT(lrm_state != NULL);
#if ENABLE_ACL
user_name = crm_acl_get_set_user(input->msg, F_CRM_USER, NULL);
crm_trace("Executor command from user '%s'", user_name);
#endif
crm_op = crm_element_value(input->msg, F_CRM_TASK);
from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) {
from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
}
crm_trace("Executor %s command from %s", crm_op, from_sys);
if (safe_str_eq(crm_op, CRM_OP_LRM_DELETE)) {
- crm_rsc_delete = TRUE; // Only crm_resource uses this op
+ if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) {
+ crm_rsc_delete = TRUE; // from crm_resource
+ }
operation = CRMD_ACTION_DELETE;
} else if (safe_str_eq(crm_op, CRM_OP_LRM_FAIL)) {
fail_lrm_resource(input->xml, lrm_state, user_name, from_host,
from_sys);
return;
} else if (input->xml != NULL) {
operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK);
}
if (safe_str_eq(crm_op, CRM_OP_LRM_REFRESH)) {
handle_refresh_op(lrm_state, user_name, from_host, from_sys);
} else if (safe_str_eq(crm_op, CRM_OP_LRM_QUERY)) {
handle_query_op(input->msg, lrm_state);
} else if (safe_str_eq(operation, CRM_OP_PROBED)) {
update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE,
user_name, is_remote_node);
} else if (safe_str_eq(operation, CRM_OP_REPROBE)
|| safe_str_eq(crm_op, CRM_OP_REPROBE)) {
handle_reprobe_op(lrm_state, from_sys, from_host, user_name,
is_remote_node);
} else if (operation != NULL) {
lrmd_rsc_info_t *rsc = NULL;
xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE);
gboolean create_rsc = safe_str_neq(operation, CRMD_ACTION_DELETE);
int rc;
// We can't return anything meaningful without a resource ID
CRM_CHECK(xml_rsc && ID(xml_rsc), return);
rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc);
if (rc == -ENOTCONN) {
synthesize_lrmd_failure(lrm_state, input->xml,
PCMK_LRM_OP_NOT_CONNECTED,
PCMK_OCF_UNKNOWN_ERROR);
return;
} else if ((rc < 0) && !create_rsc) {
/* Delete of malformed or nonexistent resource
* (deleting something that does not exist is a success)
*/
crm_notice("Not registering resource '%s' for a %s event "
CRM_XS " get-rc=%d (%s) transition-key=%s",
ID(xml_rsc), operation,
rc, pcmk_strerror(rc), ID(input->xml));
delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok,
user_name);
send_task_ok_ack(lrm_state, input, ID(xml_rsc), NULL, operation,
from_host, from_sys);
return;
} else if (rc == -EINVAL) {
// Resource operation on malformed resource
crm_err("Invalid resource definition for %s", ID(xml_rsc));
crm_log_xml_warn(input->msg, "invalid resource");
synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR,
PCMK_OCF_NOT_CONFIGURED); // fatal error
return;
} else if (rc < 0) {
// Error communicating with the executor
crm_err("Could not register resource '%s' with executor: %s "
CRM_XS " rc=%d",
ID(xml_rsc), pcmk_strerror(rc), rc);
crm_log_xml_warn(input->msg, "failed registration");
synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR,
PCMK_OCF_INVALID_PARAM); // hard error
return;
}
if (safe_str_eq(operation, CRMD_ACTION_CANCEL)) {
if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) {
crm_log_xml_warn(input->xml, "Bad command");
}
} else if (safe_str_eq(operation, CRMD_ACTION_DELETE)) {
do_lrm_delete(input, lrm_state, rsc, from_sys, from_host,
crm_rsc_delete, user_name);
} else {
do_lrm_rsc_op(lrm_state, rsc, operation, input->xml);
}
lrmd_free_rsc_info(rsc);
} else {
crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op));
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static lrmd_event_data_t *
construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op, const char *rsc_id, const char *operation)
{
lrmd_event_data_t *op = NULL;
const char *op_delay = NULL;
const char *op_timeout = NULL;
const char *interval_ms_s = NULL;
GHashTable *params = NULL;
const char *transition = NULL;
CRM_ASSERT(rsc_id && operation);
- op = calloc(1, sizeof(lrmd_event_data_t));
- CRM_ASSERT(op != NULL);
-
+ op = lrmd_new_event(rsc_id, operation, 0);
op->type = lrmd_event_exec_complete;
- op->op_type = strdup(operation);
op->op_status = PCMK_LRM_OP_PENDING;
op->rc = -1;
- op->rsc_id = strdup(rsc_id);
- op->interval_ms = 0;
op->timeout = 0;
op->start_delay = 0;
if (rsc_op == NULL) {
CRM_LOG_ASSERT(safe_str_eq(CRMD_ACTION_STOP, operation));
op->user_data = NULL;
/* the stop_all_resources() case
* by definition there is no DC (or they'd be shutting
* us down).
* So we should put our version here.
*/
op->params = crm_str_table_new();
g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET));
crm_trace("Constructed %s op for %s", operation, rsc_id);
return op;
}
params = xml2list(rsc_op);
g_hash_table_remove(params, CRM_META "_op_target_rc");
op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY);
op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT);
interval_ms_s = crm_meta_value(params, XML_LRM_ATTR_INTERVAL_MS);
op->interval_ms = crm_parse_ms(interval_ms_s);
op->timeout = crm_parse_int(op_timeout, "0");
op->start_delay = crm_parse_int(op_delay, "0");
#if ENABLE_VERSIONED_ATTRS
// Resolve any versioned parameters
if (lrm_state && safe_str_neq(op->op_type, RSC_METADATA)
&& safe_str_neq(op->op_type, CRMD_ACTION_DELETE)
&& !is_remote_lrmd_ra(NULL, NULL, rsc_id)) {
// Resource info *should* already be cached, so we don't get executor call
lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0);
struct ra_metadata_s *metadata;
metadata = metadata_cache_get(lrm_state->metadata_cache, rsc);
if (metadata) {
xmlNode *versioned_attrs = NULL;
GHashTable *hash = NULL;
char *key = NULL;
char *value = NULL;
GHashTableIter iter;
versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS);
hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
g_hash_table_iter_init(&iter, hash);
while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
g_hash_table_iter_steal(&iter);
g_hash_table_replace(params, key, value);
}
g_hash_table_destroy(hash);
versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META);
hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
g_hash_table_iter_init(&iter, hash);
while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
g_hash_table_replace(params, crm_meta_name(key), strdup(value));
if (safe_str_eq(key, XML_ATTR_TIMEOUT)) {
op->timeout = crm_parse_int(value, "0");
} else if (safe_str_eq(key, XML_OP_ATTR_START_DELAY)) {
op->start_delay = crm_parse_int(value, "0");
}
}
g_hash_table_destroy(hash);
versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS);
hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
g_hash_table_iter_init(&iter, hash);
while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
g_hash_table_iter_steal(&iter);
g_hash_table_replace(params, key, value);
}
g_hash_table_destroy(hash);
}
lrmd_free_rsc_info(rsc);
}
#endif
if (safe_str_neq(operation, RSC_STOP)) {
op->params = params;
} else {
rsc_history_t *entry = NULL;
if (lrm_state) {
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
}
/* If we do not have stop parameters cached, use
* whatever we are given */
if (!entry || !entry->stop_params) {
op->params = params;
} else {
/* Copy the cached parameter list so that we stop the resource
* with the old attributes, not the new ones */
op->params = crm_str_table_new();
g_hash_table_foreach(params, copy_meta_keys, op->params);
g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params);
g_hash_table_destroy(params);
params = NULL;
}
}
/* sanity */
if (op->timeout <= 0) {
op->timeout = op->interval_ms;
}
if (op->start_delay < 0) {
op->start_delay = 0;
}
transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY);
CRM_CHECK(transition != NULL, return op);
op->user_data = strdup(transition);
if (op->interval_ms != 0) {
if (safe_str_eq(operation, CRMD_ACTION_START)
|| safe_str_eq(operation, CRMD_ACTION_STOP)) {
crm_err("Start and Stop actions cannot have an interval: %u",
op->interval_ms);
op->interval_ms = 0;
}
}
crm_trace("Constructed %s op for %s: interval=%u",
operation, rsc_id, op->interval_ms);
return op;
}
+/*!
+ * \internal
+ * \brief Send a (synthesized) event result
+ *
+ * Reply with a synthesized event result directly, as opposed to going through
+ * the executor.
+ *
+ * \param[in] to_host Host to send result to
+ * \param[in] to_sys IPC name to send result to (NULL for transition engine)
+ * \param[in] rsc Type information about resource the result is for
+ * \param[in] op Event with result to send
+ * \param[in] rsc_id ID of resource the result is for
+ */
void
-send_direct_ack(const char *to_host, const char *to_sys,
- lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *rsc_id)
+controld_ack_event_directly(const char *to_host, const char *to_sys,
+ lrmd_rsc_info_t *rsc, lrmd_event_data_t *op,
+ const char *rsc_id)
{
xmlNode *reply = NULL;
xmlNode *update, *iter;
crm_node_t *peer = NULL;
CRM_CHECK(op != NULL, return);
if (op->rsc_id == NULL) {
CRM_ASSERT(rsc_id != NULL);
op->rsc_id = strdup(rsc_id);
}
if (to_sys == NULL) {
to_sys = CRM_SYSTEM_TENGINE;
}
peer = crm_get_peer(0, fsa_our_uname);
update = create_node_state_update(peer, node_update_none, NULL,
__FUNCTION__);
iter = create_xml_node(update, XML_CIB_TAG_LRM);
crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE);
crm_xml_add(iter, XML_ATTR_ID, op->rsc_id);
build_operation_update(iter, rsc, op, fsa_our_uname, __FUNCTION__);
reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL);
crm_log_xml_trace(update, "ACK Update");
crm_debug("ACK'ing resource op " CRM_OP_FMT " from %s: %s",
op->rsc_id, op->op_type, op->interval_ms, op->user_data,
crm_element_value(reply, XML_ATTR_REFERENCE));
if (relay_message(reply, TRUE) == FALSE) {
crm_log_xml_err(reply, "Unable to route reply");
}
free_xml(update);
free_xml(reply);
}
gboolean
verify_stopped(enum crmd_fsa_state cur_state, int log_level)
{
gboolean res = TRUE;
GList *lrm_state_list = lrm_state_get_list();
GList *state_entry;
for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) {
lrm_state_t *lrm_state = state_entry->data;
if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) {
/* keep iterating through all even when false is returned */
res = FALSE;
}
}
set_bit(fsa_input_register, R_SENT_RSC_STOP);
g_list_free(lrm_state_list); lrm_state_list = NULL;
return res;
}
struct stop_recurring_action_s {
lrmd_rsc_info_t *rsc;
lrm_state_t *lrm_state;
};
static gboolean
stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
struct stop_recurring_action_s *event = user_data;
active_op_t *op = value;
if ((op->interval_ms != 0)
&& crm_str_eq(op->rsc_id, event->rsc->id, TRUE)) {
crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key);
remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE);
}
return remove;
}
static gboolean
stop_recurring_actions(gpointer key, gpointer value, gpointer user_data)
{
gboolean remove = FALSE;
lrm_state_t *lrm_state = user_data;
active_op_t *op = value;
if (op->interval_ms != 0) {
crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id,
(const char *) key);
remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE);
}
return remove;
}
static void
record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op)
{
const char *record_pending = NULL;
CRM_CHECK(node_name != NULL, return);
CRM_CHECK(rsc != NULL, return);
CRM_CHECK(op != NULL, return);
// Never record certain operation types as pending
if ((op->op_type == NULL) || (op->params == NULL)
|| !controld_action_is_recordable(op->op_type)) {
return;
}
// defaults to true
record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING);
if (record_pending && !crm_is_true(record_pending)) {
return;
}
op->call_id = -1;
op->op_status = PCMK_LRM_OP_PENDING;
op->rc = PCMK_OCF_UNKNOWN;
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
/* write a "pending" entry to the CIB, inhibit notification */
crm_debug("Recording pending op " CRM_OP_FMT " on %s in the CIB",
op->rsc_id, op->op_type, op->interval_ms, node_name);
- do_update_resource(node_name, rsc, op);
+ do_update_resource(node_name, rsc, op, 0);
}
static void
do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
const char *operation, xmlNode *msg)
{
int call_id = 0;
char *op_id = NULL;
lrmd_event_data_t *op = NULL;
lrmd_key_value_t *params = NULL;
fsa_data_t *msg_data = NULL;
const char *transition = NULL;
gboolean stop_recurring = FALSE;
bool send_nack = FALSE;
CRM_CHECK(rsc != NULL, return);
CRM_CHECK(operation != NULL, return);
if (msg != NULL) {
transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY);
if (transition == NULL) {
crm_log_xml_err(msg, "Missing transition number");
}
}
op = construct_op(lrm_state, msg, rsc->id, operation);
CRM_CHECK(op != NULL, return);
if (is_remote_lrmd_ra(NULL, NULL, rsc->id)
&& (op->interval_ms == 0)
&& strcmp(operation, CRMD_ACTION_MIGRATE) == 0) {
/* pcmk remote connections are a special use case.
* We never ever want to stop monitoring a connection resource until
* the entire migration has completed. If the connection is unexpectedly
* severed, even during a migration, this is an event we must detect.*/
stop_recurring = FALSE;
} else if ((op->interval_ms == 0)
&& strcmp(operation, CRMD_ACTION_STATUS) != 0
&& strcmp(operation, CRMD_ACTION_NOTIFY) != 0) {
/* stop any previous monitor operations before changing the resource state */
stop_recurring = TRUE;
}
if (stop_recurring == TRUE) {
guint removed = 0;
struct stop_recurring_action_s data;
data.rsc = rsc;
data.lrm_state = lrm_state;
removed = g_hash_table_foreach_remove(
lrm_state->pending_ops, stop_recurring_action_by_rsc, &data);
if (removed) {
crm_debug("Stopped %u recurring operation%s in preparation for " CRM_OP_FMT,
removed, pcmk__plural_s(removed),
rsc->id, operation, op->interval_ms);
}
}
/* now do the op */
crm_info("Performing key=%s op=" CRM_OP_FMT,
transition, rsc->id, operation, op->interval_ms);
if (is_set(fsa_input_register, R_SHUTDOWN) && safe_str_eq(operation, RSC_START)) {
register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
send_nack = TRUE;
} else if (fsa_state != S_NOT_DC
&& fsa_state != S_POLICY_ENGINE /* Recalculating */
&& fsa_state != S_TRANSITION_ENGINE
&& safe_str_neq(operation, CRMD_ACTION_STOP)) {
send_nack = TRUE;
}
if(send_nack) {
crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)",
operation, rsc->id, fsa_state2string(fsa_state),
is_set(fsa_input_register, R_SHUTDOWN)?"true":"false");
op->rc = PCMK_OCF_UNKNOWN_ERROR;
op->op_status = PCMK_LRM_OP_INVALID;
- send_direct_ack(NULL, NULL, rsc, op, rsc->id);
+ controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
lrmd_free_event(op);
free(op_id);
return;
}
record_pending_op(lrm_state->node_name, rsc, op);
op_id = generate_op_key(rsc->id, op->op_type, op->interval_ms);
if (op->interval_ms > 0) {
/* cancel it so we can then restart it without conflict */
cancel_op_key(lrm_state, rsc, op_id, FALSE);
}
if (op->params) {
char *key = NULL;
char *value = NULL;
GHashTableIter iter;
g_hash_table_iter_init(&iter, op->params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
params = lrmd_key_value_add(params, key, value);
}
}
call_id = lrm_state_exec(lrm_state, rsc->id, op->op_type, op->user_data,
op->interval_ms, op->timeout, op->start_delay,
params);
if (call_id <= 0 && lrm_state_is_local(lrm_state)) {
crm_err("Operation %s on %s failed: %d", operation, rsc->id, call_id);
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
} else if (call_id <= 0) {
crm_err("Operation %s on resource %s failed to execute on remote node %s: %d",
operation, rsc->id, lrm_state->node_name, call_id);
fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR);
process_lrm_event(lrm_state, op, NULL, NULL);
} else {
/* record all operations so we can wait
* for them to complete during shutdown
*/
char *call_id_s = make_stop_id(rsc->id, call_id);
active_op_t *pending = NULL;
pending = calloc(1, sizeof(active_op_t));
crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s);
pending->call_id = call_id;
pending->interval_ms = op->interval_ms;
pending->op_type = strdup(operation);
pending->op_key = strdup(op_id);
pending->rsc_id = strdup(rsc->id);
pending->start_time = time(NULL);
pending->user_data = op->user_data? strdup(op->user_data) : NULL;
+ if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
+ &(pending->lock_time)) != pcmk_ok) {
+ pending->lock_time = 0;
+ }
g_hash_table_replace(lrm_state->pending_ops, call_id_s, pending);
if ((op->interval_ms > 0)
&& (op->start_delay > START_DELAY_THRESHOLD)) {
int target_rc = 0;
crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id);
decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc);
op->rc = target_rc;
op->op_status = PCMK_LRM_OP_DONE;
- send_direct_ack(NULL, NULL, rsc, op, rsc->id);
+ controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
}
pending->params = op->params;
op->params = NULL;
}
free(op_id);
lrmd_free_event(op);
return;
}
int last_resource_update = 0;
static void
cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
switch (rc) {
case pcmk_ok:
case -pcmk_err_diff_failed:
case -pcmk_err_diff_resync:
crm_trace("Resource update %d complete: rc=%d", call_id, rc);
break;
default:
crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc));
}
if (call_id == last_resource_update) {
last_resource_update = 0;
trigger_fsa(fsa_source);
}
}
+/* Only successful stops, and probes that found the resource inactive, get locks
+ * recorded in the history. This ensures the resource stays locked to the node
+ * until it is active there again after the node comes back up.
+ */
+static bool
+should_preserve_lock(lrmd_event_data_t *op)
+{
+ if (!controld_shutdown_lock_enabled) {
+ return false;
+ }
+ if (!strcmp(op->op_type, RSC_STOP) && (op->rc == PCMK_OCF_OK)) {
+ return true;
+ }
+ if (!strcmp(op->op_type, RSC_STATUS) && (op->rc == PCMK_OCF_NOT_RUNNING)) {
+ return true;
+ }
+ return false;
+}
+
static int
-do_update_resource(const char *node_name, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op)
+do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc,
+ lrmd_event_data_t *op, time_t lock_time)
{
/*
*/
int rc = pcmk_ok;
xmlNode *update, *iter = NULL;
int call_opt = crmd_cib_smart_opt();
const char *uuid = NULL;
CRM_CHECK(op != NULL, return 0);
iter = create_xml_node(iter, XML_CIB_TAG_STATUS);
update = iter;
iter = create_xml_node(iter, XML_CIB_TAG_STATE);
if (safe_str_eq(node_name, fsa_our_uname)) {
uuid = fsa_our_uuid;
} else {
/* remote nodes uuid and uname are equal */
uuid = node_name;
crm_xml_add(iter, XML_NODE_IS_REMOTE, "true");
}
CRM_LOG_ASSERT(uuid != NULL);
if(uuid == NULL) {
rc = -EINVAL;
goto done;
}
crm_xml_add(iter, XML_ATTR_UUID, uuid);
crm_xml_add(iter, XML_ATTR_UNAME, node_name);
crm_xml_add(iter, XML_ATTR_ORIGIN, __FUNCTION__);
iter = create_xml_node(iter, XML_CIB_TAG_LRM);
crm_xml_add(iter, XML_ATTR_ID, uuid);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES);
iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE);
crm_xml_add(iter, XML_ATTR_ID, op->rsc_id);
build_operation_update(iter, rsc, op, node_name, __FUNCTION__);
if (rsc) {
const char *container = NULL;
crm_xml_add(iter, XML_ATTR_TYPE, rsc->type);
crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard);
crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider);
+ if (lock_time != 0) {
+ /* Actions on a locked resource should either preserve the lock by
+ * recording it with the action result, or clear it.
+ */
+ if (!should_preserve_lock(op)) {
+ lock_time = 0;
+ }
+ crm_xml_add_ll(iter, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
+ (long long) lock_time);
+ }
if (op->params) {
container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER);
}
if (container) {
crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container);
crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container);
}
} else {
crm_warn("Resource %s no longer exists in the executor", op->rsc_id);
- send_direct_ack(NULL, NULL, rsc, op, op->rsc_id);
+ controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id);
goto cleanup;
}
crm_log_xml_trace(update, __FUNCTION__);
/* make it an asynchronous call and be done with it
*
* Best case:
* the resource state will be discovered during
* the next signup or election.
*
* Bad case:
* we are shutting down and there is no DC at the time,
* but then why were we shutting down then anyway?
* (probably because of an internal error)
*
* Worst case:
* we get shot for having resources "running" that really weren't
*
* the alternative however means blocking here for too long, which
* isn't acceptable
*/
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL);
if (rc > 0) {
last_resource_update = rc;
}
done:
/* the return code is a call number, not an error code */
crm_trace("Sent resource state update message: %d for %s=%u on %s",
rc, op->op_type, op->interval_ms, op->rsc_id);
fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback);
cleanup:
free_xml(update);
return rc;
}
void
do_lrm_event(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data)
{
CRM_CHECK(FALSE, return);
}
static char *
unescape_newlines(const char *string)
{
char *pch = NULL;
char *ret = NULL;
static const char *escaped_newline = "\\n";
if (!string) {
return NULL;
}
ret = strdup(string);
pch = strstr(ret, escaped_newline);
while (pch != NULL) {
/* Replace newline escape pattern with actual newline (and a space so we
* don't have to shuffle the rest of the buffer)
*/
pch[0] = '\n';
pch[1] = ' ';
pch = strstr(pch, escaped_newline);
}
return ret;
}
static bool
did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id,
const char * op_type, guint interval_ms)
{
rsc_history_t *entry = NULL;
CRM_CHECK(lrm_state != NULL, return FALSE);
CRM_CHECK(rsc_id != NULL, return FALSE);
CRM_CHECK(op_type != NULL, return FALSE);
entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
if (entry == NULL || entry->failed == NULL) {
return FALSE;
}
if (crm_str_eq(entry->failed->rsc_id, rsc_id, TRUE)
&& safe_str_eq(entry->failed->op_type, op_type)
&& entry->failed->interval_ms == interval_ms) {
return TRUE;
}
return FALSE;
}
void
process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op,
active_op_t *pending, xmlNode *action_xml)
{
char *op_id = NULL;
char *op_key = NULL;
int update_id = 0;
gboolean remove = FALSE;
gboolean removed = FALSE;
bool need_direct_ack = FALSE;
lrmd_rsc_info_t *rsc = NULL;
const char *node_name = NULL;
CRM_CHECK(op != NULL, return);
CRM_CHECK(op->rsc_id != NULL, return);
// Remap new status codes for older DCs
if (compare_version(fsa_our_dc_version, "3.2.0") < 0) {
switch (op->op_status) {
case PCMK_LRM_OP_NOT_CONNECTED:
op->op_status = PCMK_LRM_OP_ERROR;
op->rc = PCMK_OCF_CONNECTION_DIED;
break;
case PCMK_LRM_OP_INVALID:
op->op_status = PCMK_LRM_OP_ERROR;
op->rc = CRM_DIRECT_NACK_RC;
break;
default:
break;
}
}
op_id = make_stop_id(op->rsc_id, op->call_id);
op_key = generate_op_key(op->rsc_id, op->op_type, op->interval_ms);
// Get resource info if available (from executor state or action XML)
if (lrm_state) {
rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0);
}
if ((rsc == NULL) && action_xml) {
xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE);
const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS);
const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER);
const char *type = crm_element_value(xml, XML_ATTR_TYPE);
if (standard && type) {
crm_info("%s agent information not cached, using %s%s%s:%s from action XML",
op->rsc_id, standard,
(provider? ":" : ""), (provider? provider : ""), type);
rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type);
} else {
crm_err("Can't process %s result because %s agent information not cached or in XML",
op_key, op->rsc_id);
}
}
// Get node name if available (from executor state or action XML)
if (lrm_state) {
node_name = lrm_state->node_name;
} else if (action_xml) {
node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET);
}
if(pending == NULL) {
remove = TRUE;
if (lrm_state) {
pending = g_hash_table_lookup(lrm_state->pending_ops, op_id);
}
}
if (op->op_status == PCMK_LRM_OP_ERROR) {
switch(op->rc) {
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_MASTER:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_MASTER:
// Leave it to the TE/scheduler to decide if this is an error
op->op_status = PCMK_LRM_OP_DONE;
break;
default:
/* Nothing to do */
break;
}
}
if (op->op_status != PCMK_LRM_OP_CANCELLED) {
/* We might not record the result, so directly acknowledge it to the
* originator instead, so it doesn't time out waiting for the result
* (especially important if part of a transition).
*/
need_direct_ack = TRUE;
if (controld_action_is_recordable(op->op_type)) {
if (node_name && rsc) {
// We should record the result, and happily, we can
- update_id = do_update_resource(node_name, rsc, op);
+ update_id = do_update_resource(node_name, rsc, op,
+ pending? pending->lock_time : 0);
need_direct_ack = FALSE;
} else if (op->rsc_deleted) {
/* We shouldn't record the result (likely the resource was
* refreshed, cleaned, or removed while this operation was
* in flight).
*/
crm_notice("Not recording %s result in CIB because "
"resource information was removed since it was initiated",
op_key);
} else {
/* This shouldn't be possible; the executor didn't consider the
* resource deleted, but we couldn't find resource or node
* information.
*/
crm_err("Unable to record %s result in CIB: %s", op_key,
(node_name? "No resource information" : "No node name"));
}
}
} else if (op->interval_ms == 0) {
/* A non-recurring operation was cancelled. Most likely, the
* never-initiated action was removed from the executor's pending
* operations list upon resource removal.
*/
need_direct_ack = TRUE;
} else if (pending == NULL) {
/* This recurring operation was cancelled, but was not pending. No
* transition actions are waiting on it, nothing needs to be done.
*/
} else if (op->user_data == NULL) {
/* This recurring operation was cancelled and pending, but we don't
* have a transition key. This should never happen.
*/
crm_err("Recurring operation %s was cancelled without transition information",
op_key);
} else if (is_set(pending->flags, active_op_remove)) {
/* This recurring operation was cancelled (by us) and pending, and we
* have been waiting for it to finish.
*/
if (lrm_state) {
erase_lrm_history_by_op(lrm_state, op);
}
/* If the recurring operation had failed, the lrm_rsc_op is recorded as
* "last_failure" which won't get erased from the cib given the logic on
* purpose in erase_lrm_history_by_op(). So that the cancel action won't
* have a chance to get confirmed by DC with process_op_deletion().
* Cluster transition would get stuck waiting for the remaining action
* timer to time out.
*
* Directly acknowledge the cancel operation in this case.
*/
if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id,
pending->op_type, pending->interval_ms)) {
need_direct_ack = TRUE;
}
} else if (op->rsc_deleted) {
/* This recurring operation was cancelled (but not by us, and the
* executor does not have resource information, likely due to resource
* cleanup, refresh, or removal) and pending.
*/
crm_debug("Recurring op %s was cancelled due to resource deletion",
op_key);
need_direct_ack = TRUE;
} else {
/* This recurring operation was cancelled (but not by us, likely by the
* executor before stopping the resource) and pending. We don't need to
* do anything special.
*/
}
if (need_direct_ack) {
- send_direct_ack(NULL, NULL, NULL, op, op->rsc_id);
+ controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id);
}
if(remove == FALSE) {
/* The caller will do this afterwards, but keep the logging consistent */
removed = TRUE;
} else if (lrm_state && ((op->interval_ms == 0)
|| (op->op_status == PCMK_LRM_OP_CANCELLED))) {
gboolean found = g_hash_table_remove(lrm_state->pending_ops, op_id);
if (op->interval_ms != 0) {
removed = TRUE;
} else if (found) {
removed = TRUE;
crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed",
op_key, op->call_id, op_id,
g_hash_table_size(lrm_state->pending_ops));
}
}
if (node_name == NULL) {
node_name = "unknown node"; // for logging
}
switch (op->op_status) {
case PCMK_LRM_OP_CANCELLED:
crm_info("Result of %s operation for %s on %s: %s "
CRM_XS " call=%d key=%s confirmed=%s",
crm_action_str(op->op_type, op->interval_ms),
op->rsc_id, node_name,
services_lrm_status_str(op->op_status),
op->call_id, op_key, (removed? "true" : "false"));
break;
case PCMK_LRM_OP_DONE:
crm_notice("Result of %s operation for %s on %s: %d (%s) "
CRM_XS " call=%d key=%s confirmed=%s cib-update=%d",
crm_action_str(op->op_type, op->interval_ms),
op->rsc_id, node_name,
op->rc, services_ocf_exitcode_str(op->rc),
op->call_id, op_key, (removed? "true" : "false"),
update_id);
break;
case PCMK_LRM_OP_TIMEOUT:
crm_err("Result of %s operation for %s on %s: %s "
CRM_XS " call=%d key=%s timeout=%dms",
crm_action_str(op->op_type, op->interval_ms),
op->rsc_id, node_name,
services_lrm_status_str(op->op_status),
op->call_id, op_key, op->timeout);
break;
default:
crm_err("Result of %s operation for %s on %s: %s "
CRM_XS " call=%d key=%s confirmed=%s status=%d cib-update=%d",
crm_action_str(op->op_type, op->interval_ms),
op->rsc_id, node_name,
services_lrm_status_str(op->op_status), op->call_id, op_key,
(removed? "true" : "false"), op->op_status, update_id);
}
if (op->output) {
char *prefix =
crm_strdup_printf("%s-" CRM_OP_FMT ":%d", node_name,
op->rsc_id, op->op_type, op->interval_ms,
op->call_id);
if (op->rc) {
crm_log_output(LOG_NOTICE, prefix, op->output);
} else {
crm_log_output(LOG_DEBUG, prefix, op->output);
}
free(prefix);
}
if (lrm_state) {
if (safe_str_neq(op->op_type, RSC_METADATA)) {
crmd_alert_resource_op(lrm_state->node_name, op);
} else if (rsc && (op->rc == PCMK_OCF_OK)) {
char *metadata = unescape_newlines(op->output);
metadata_cache_update(lrm_state->metadata_cache, rsc, metadata);
free(metadata);
}
}
if (op->rsc_deleted) {
crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key);
if (lrm_state) {
delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL);
}
}
/* If a shutdown was escalated while operations were pending,
* then the FSA will be stalled right now... allow it to continue
*/
mainloop_set_trigger(fsa_source);
if (lrm_state && rsc) {
update_history_cache(lrm_state, rsc, op);
}
lrmd_free_rsc_info(rsc);
free(op_key);
free(op_id);
}
diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c
index 885b2a95d9..f0eb2a2def 100644
--- a/daemons/controld/controld_join_dc.c
+++ b/daemons/controld/controld_join_dc.c
@@ -1,770 +1,773 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
char *max_epoch = NULL;
char *max_generation_from = NULL;
xmlNode *max_generation_xml = NULL;
void finalize_join_for(gpointer key, gpointer value, gpointer user_data);
void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data);
gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source);
/* Numeric counter used to identify join rounds (an unsigned int would be
* appropriate, except we get and set it in XML as int)
*/
static int current_join_id = 0;
unsigned long long saved_ccm_membership_id = 0;
void
crm_update_peer_join(const char *source, crm_node_t * node, enum crm_join_phase phase)
{
enum crm_join_phase last = 0;
CRM_CHECK(node != NULL, return);
/* Remote nodes do not participate in joins */
if (is_set(node->flags, crm_remote_node)) {
return;
}
last = node->join;
if(phase == last) {
crm_trace("Node %s join-%d phase is still %s "
CRM_XS " nodeid=%u source=%s",
node->uname, current_join_id, crm_join_phase_str(last),
node->id, source);
} else if ((phase <= crm_join_none) || (phase == (last + 1))) {
node->join = phase;
crm_trace("Node %s join-%d phase is now %s (was %s) "
CRM_XS " nodeid=%u source=%s",
node->uname, current_join_id, crm_join_phase_str(phase),
crm_join_phase_str(last), node->id, source);
} else {
crm_warn("Rejecting join-%d phase update for node %s because "
"can't go from %s to %s " CRM_XS " nodeid=%u source=%s",
current_join_id, node->uname, crm_join_phase_str(last),
crm_join_phase_str(phase), node->id, source);
}
}
static void
start_join_round()
{
GHashTableIter iter;
crm_node_t *peer = NULL;
crm_debug("Starting new join round join-%d", current_join_id);
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
crm_update_peer_join(__FUNCTION__, peer, crm_join_none);
}
if (max_generation_from != NULL) {
free(max_generation_from);
max_generation_from = NULL;
}
if (max_generation_xml != NULL) {
free_xml(max_generation_xml);
max_generation_xml = NULL;
}
clear_bit(fsa_input_register, R_HAVE_CIB);
clear_bit(fsa_input_register, R_CIB_ASKED);
}
/*!
* \internal
* \brief Create a join message from the DC
*
* \param[in] join_op Join operation name
* \param[in] host_to Recipient of message
*/
static xmlNode *
create_dc_message(const char *join_op, const char *host_to)
{
xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD,
CRM_SYSTEM_DC, NULL);
/* Identify which election this is a part of */
crm_xml_add_int(msg, F_CRM_JOIN_ID, current_join_id);
/* Add a field specifying whether the DC is shutting down. This keeps the
* joining node from fencing the old DC if it becomes the new DC.
*/
crm_xml_add_boolean(msg, F_CRM_DC_LEAVING,
is_set(fsa_input_register, R_SHUTDOWN));
return msg;
}
static void
join_make_offer(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *offer = NULL;
crm_node_t *member = (crm_node_t *)value;
CRM_ASSERT(member != NULL);
if (crm_is_peer_active(member) == FALSE) {
crm_info("Not making join-%d offer to inactive node %s",
current_join_id,
(member->uname? member->uname : "with unknown name"));
if(member->expected == NULL && safe_str_eq(member->state, CRM_NODE_LOST)) {
/* You would think this unsafe, but in fact this plus an
* active resource is what causes it to be fenced.
*
* Yes, this does mean that any node that dies at the same
* time as the old DC and is not running resource (still)
* won't be fenced.
*
* I'm not happy about this either.
*/
crm_update_peer_expected(__FUNCTION__, member, CRMD_JOINSTATE_DOWN);
}
return;
}
if (member->uname == NULL) {
crm_info("Not making join-%d offer to node uuid %s with unknown name",
current_join_id, member->uuid);
return;
}
if (saved_ccm_membership_id != crm_peer_seq) {
saved_ccm_membership_id = crm_peer_seq;
crm_info("Making join-%d offers based on membership event %llu",
current_join_id, crm_peer_seq);
}
if(user_data && member->join > crm_join_none) {
crm_info("Not making join-%d offer to already known node %s (%s)",
current_join_id, member->uname,
crm_join_phase_str(member->join));
return;
}
crm_update_peer_join(__FUNCTION__, (crm_node_t*)member, crm_join_none);
offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname);
// Advertise our feature set so the joining node can bail if not compatible
crm_xml_add(offer, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
crm_info("Sending join-%d offer to %s", current_join_id, member->uname);
send_cluster_message(member, crm_msg_crmd, offer, TRUE);
free_xml(offer);
crm_update_peer_join(__FUNCTION__, member, crm_join_welcomed);
}
/* A_DC_JOIN_OFFER_ALL */
void
do_dc_join_offer_all(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
int count;
/* Reset everyone's status back to down or in_ccm in the CIB.
* Any nodes that are active in the CIB but not in the cluster membership
* will be seen as offline by the scheduler anyway.
*/
current_join_id++;
start_join_round();
/* do_update_cib_nodes(TRUE, __FUNCTION__); */
update_dc(NULL);
if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) {
crm_info("A new node joined the cluster");
}
g_hash_table_foreach(crm_peer_cache, join_make_offer, NULL);
count = crmd_join_phase_count(crm_join_welcomed);
crm_info("Waiting on join-%d requests from %d outstanding node%s",
current_join_id, count, pcmk__plural_s(count));
// Don't waste time by invoking the scheduler yet
}
/* A_DC_JOIN_OFFER_ONE */
void
do_dc_join_offer_one(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_node_t *member;
ha_msg_input_t *welcome = NULL;
int count;
const char *join_to = NULL;
if (msg_data->data == NULL) {
crm_info("Making join-%d offers to any unconfirmed nodes "
"because an unknown node joined", current_join_id);
g_hash_table_foreach(crm_peer_cache, join_make_offer, &member);
check_join_state(cur_state, __FUNCTION__);
return;
}
welcome = fsa_typed_data(fsa_dt_ha_msg);
if (welcome == NULL) {
// fsa_typed_data() already logged an error
return;
}
join_to = crm_element_value(welcome->msg, F_CRM_HOST_FROM);
if (join_to == NULL) {
crm_err("Can't make join-%d offer to unknown node", current_join_id);
return;
}
member = crm_get_peer(0, join_to);
/* It is possible that a node will have been sick or starting up when the
* original offer was made. However, it will either re-announce itself in
* due course, or we can re-store the original offer on the client.
*/
crm_update_peer_join(__FUNCTION__, member, crm_join_none);
join_make_offer(NULL, member, NULL);
/* If the offer isn't to the local node, make an offer to the local node as
* well, to ensure the correct value for max_generation_from.
*/
if (strcmp(join_to, fsa_our_uname) != 0) {
member = crm_get_peer(0, fsa_our_uname);
join_make_offer(NULL, member, NULL);
}
/* This was a genuine join request; cancel any existing transition and
* invoke the scheduler.
*/
abort_transition(INFINITY, tg_restart, "Node join", NULL);
count = crmd_join_phase_count(crm_join_welcomed);
crm_info("Waiting on join-%d requests from %d outstanding node%s",
current_join_id, count, pcmk__plural_s(count));
// Don't waste time by invoking the scheduler yet
}
static int
compare_int_fields(xmlNode * left, xmlNode * right, const char *field)
{
const char *elem_l = crm_element_value(left, field);
const char *elem_r = crm_element_value(right, field);
int int_elem_l = crm_int_helper(elem_l, NULL);
int int_elem_r = crm_int_helper(elem_r, NULL);
if (int_elem_l < int_elem_r) {
return -1;
} else if (int_elem_l > int_elem_r) {
return 1;
}
return 0;
}
/* A_DC_JOIN_PROCESS_REQ */
void
do_dc_join_filter_offer(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
xmlNode *generation = NULL;
int cmp = 0;
int join_id = -1;
int count = 0;
gboolean ack_nack_bool = TRUE;
ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM);
const char *ref = crm_element_value(join_ack->msg, F_CRM_REFERENCE);
const char *join_version = crm_element_value(join_ack->msg,
XML_ATTR_CRM_VERSION);
crm_node_t *join_node = NULL;
if (join_from == NULL) {
crm_err("Ignoring invalid join request without node name");
return;
}
join_node = crm_get_peer(0, join_from);
crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id);
if (join_id != current_join_id) {
crm_debug("Ignoring join-%d request from %s because we are on join-%d",
join_id, join_from, current_join_id);
check_join_state(cur_state, __FUNCTION__);
return;
}
generation = join_ack->xml;
if (max_generation_xml != NULL && generation != NULL) {
int lpc = 0;
const char *attributes[] = {
XML_ATTR_GENERATION_ADMIN,
XML_ATTR_GENERATION,
XML_ATTR_NUMUPDATES,
};
for (lpc = 0; cmp == 0 && lpc < DIMOF(attributes); lpc++) {
cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]);
}
}
if (ref == NULL) {
ref = "none"; // for logging only
}
if (crm_is_peer_active(join_node) == FALSE) {
crm_err("Rejecting join-%d request from inactive node %s "
CRM_XS " ref=%s", join_id, join_from, ref);
ack_nack_bool = FALSE;
} else if (generation == NULL) {
crm_err("Rejecting invalid join-%d request from node %s "
"missing CIB generation " CRM_XS " ref=%s",
join_id, join_from, ref);
ack_nack_bool = FALSE;
} else if ((join_version == NULL)
|| !feature_set_compatible(CRM_FEATURE_SET, join_version)) {
crm_err("Rejecting join-%d request from node %s because feature set %s"
" is incompatible with ours (%s) " CRM_XS " ref=%s",
join_id, join_from, (join_version? join_version : "pre-3.1.0"),
CRM_FEATURE_SET, ref);
ack_nack_bool = FALSE;
} else if (max_generation_xml == NULL) {
crm_debug("Accepting join-%d request from %s "
"(with first CIB generation) " CRM_XS " ref=%s",
join_id, join_from, ref);
max_generation_xml = copy_xml(generation);
max_generation_from = strdup(join_from);
} else if (cmp < 0 || (cmp == 0 && safe_str_eq(join_from, fsa_our_uname))) {
crm_debug("Accepting join-%d request from %s (with better "
"CIB generation than current best from %s) " CRM_XS " ref=%s",
join_id, join_from, max_generation_from, ref);
crm_log_xml_debug(max_generation_xml, "Old max generation");
crm_log_xml_debug(generation, "New max generation");
free(max_generation_from);
free_xml(max_generation_xml);
max_generation_from = strdup(join_from);
max_generation_xml = copy_xml(join_ack->xml);
} else {
crm_debug("Accepting join-%d request from %s " CRM_XS " ref=%s",
join_id, join_from, ref);
}
if (ack_nack_bool == FALSE) {
crm_update_peer_join(__FUNCTION__, join_node, crm_join_nack);
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_NACK);
} else {
crm_update_peer_join(__FUNCTION__, join_node, crm_join_integrated);
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_MEMBER);
}
count = crmd_join_phase_count(crm_join_integrated);
crm_debug("%d node%s currently integrated in join-%d",
count, pcmk__plural_s(count), join_id);
if (check_join_state(cur_state, __FUNCTION__) == FALSE) {
// Don't waste time by invoking the scheduler yet
count = crmd_join_phase_count(crm_join_welcomed);
crm_debug("Waiting on join-%d requests from %d outstanding node%s",
join_id, count, pcmk__plural_s(count));
}
}
/* A_DC_JOIN_FINALIZE */
void
do_dc_join_finalize(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
char *sync_from = NULL;
int rc = pcmk_ok;
int count_welcomed = crmd_join_phase_count(crm_join_welcomed);
int count_integrated = crmd_join_phase_count(crm_join_integrated);
/* This we can do straight away and avoid clients timing us out
* while we compute the latest CIB
*/
if (count_welcomed != 0) {
crm_debug("Waiting on join-%d requests from %d outstanding node%s "
"before finalizing join", current_join_id, count_welcomed,
pcmk__plural_s(count_welcomed));
crmd_join_phase_log(LOG_DEBUG);
/* crmd_fsa_stall(FALSE); Needed? */
return;
} else if (count_integrated == 0) {
crm_debug("Finalization not needed for join-%d at the current time",
current_join_id);
crmd_join_phase_log(LOG_DEBUG);
check_join_state(fsa_state, __FUNCTION__);
return;
}
clear_bit(fsa_input_register, R_HAVE_CIB);
if (max_generation_from == NULL || safe_str_eq(max_generation_from, fsa_our_uname)) {
set_bit(fsa_input_register, R_HAVE_CIB);
}
if (is_set(fsa_input_register, R_IN_TRANSITION)) {
crm_warn("Delaying join-%d finalization while transition in progress",
current_join_id);
crmd_join_phase_log(LOG_DEBUG);
crmd_fsa_stall(FALSE);
return;
}
if (max_generation_from && is_set(fsa_input_register, R_HAVE_CIB) == FALSE) {
/* ask for the agreed best CIB */
sync_from = strdup(max_generation_from);
set_bit(fsa_input_register, R_CIB_ASKED);
crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)",
current_join_id, count_integrated,
pcmk__plural_s(count_integrated), sync_from);
crm_log_xml_notice(max_generation_xml, "Requested CIB version");
} else {
/* Send _our_ CIB out to everyone */
sync_from = strdup(fsa_our_uname);
crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)",
current_join_id, count_integrated,
pcmk__plural_s(count_integrated));
crm_log_xml_debug(max_generation_xml, "Requested CIB version");
}
crmd_join_phase_log(LOG_DEBUG);
rc = fsa_cib_conn->cmds->sync_from(fsa_cib_conn, sync_from, NULL, cib_quorum_override);
fsa_register_cib_callback(rc, FALSE, sync_from, finalize_sync_callback);
}
void
finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
CRM_LOG_ASSERT(-EPERM != rc);
clear_bit(fsa_input_register, R_CIB_ASKED);
if (rc != pcmk_ok) {
do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR),
"Could not sync CIB from %s in join-%d: %s",
(char *) user_data, current_join_id, pcmk_strerror(rc));
/* restart the whole join process */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __FUNCTION__);
} else if (!AM_I_DC) {
crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id);
} else if (fsa_state != S_FINALIZE_JOIN) {
crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN (%s)",
current_join_id, fsa_state2string(fsa_state));
} else {
set_bit(fsa_input_register, R_HAVE_CIB);
clear_bit(fsa_input_register, R_CIB_ASKED);
/* make sure dc_uuid is re-set to us */
if (check_join_state(fsa_state, __FUNCTION__) == FALSE) {
int count_integrated = crmd_join_phase_count(crm_join_integrated);
crm_debug("Notifying %d node%s of join-%d results",
count_integrated, pcmk__plural_s(count_integrated),
current_join_id);
g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL);
}
}
}
static void
join_update_complete_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if (rc == pcmk_ok) {
crm_debug("join-%d node history update (via CIB call %d) complete",
current_join_id, call_id);
check_join_state(fsa_state, __FUNCTION__);
} else {
crm_err("join-%d node history update (via CIB call %d) failed: %s "
"(next transition may determine resource status incorrectly)",
current_join_id, call_id, pcmk_strerror(rc));
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
/* A_DC_JOIN_PROCESS_ACK */
void
do_dc_join_ack(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
int join_id = -1;
int call_id = 0;
ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
+ enum controld_section_e section = controld_section_lrm;
const char *op = crm_element_value(join_ack->msg, F_CRM_TASK);
const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM);
crm_node_t *peer = NULL;
// Sanity checks
if (join_from == NULL) {
crm_warn("Ignoring message received without node identification");
return;
}
if (op == NULL) {
crm_warn("Ignoring message received from %s without task", join_from);
return;
}
if (strcmp(op, CRM_OP_JOIN_CONFIRM)) {
crm_debug("Ignoring '%s' message from %s while waiting for '%s'",
op, join_from, CRM_OP_JOIN_CONFIRM);
return;
}
if (crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id) != 0) {
crm_warn("Ignoring join confirmation from %s without valid join ID",
join_from);
return;
}
peer = crm_get_peer(0, join_from);
if (peer->join != crm_join_finalized) {
crm_info("Ignoring out-of-sequence join-%d confirmation from %s "
"(currently %s not %s)",
join_id, join_from, crm_join_phase_str(peer->join),
crm_join_phase_str(crm_join_finalized));
return;
}
if (join_id != current_join_id) {
crm_err("Rejecting join-%d confirmation from %s "
"because currently on join-%d",
join_id, join_from, current_join_id);
crm_update_peer_join(__FUNCTION__, peer, crm_join_nack);
return;
}
crm_update_peer_join(__FUNCTION__, peer, crm_join_confirmed);
/* Update CIB with node's current executor state. A new transition will be
* triggered later, when the CIB notifies us of the change.
*/
- controld_delete_node_state(join_from, controld_section_lrm,
- cib_scope_local);
+ if (controld_shutdown_lock_enabled) {
+ section = controld_section_lrm_unlocked;
+ }
+ controld_delete_node_state(join_from, section, cib_scope_local);
if (safe_str_eq(join_from, fsa_our_uname)) {
xmlNode *now_dc_lrmd_state = controld_query_executor_state(fsa_our_uname);
if (now_dc_lrmd_state != NULL) {
fsa_cib_update(XML_CIB_TAG_STATUS, now_dc_lrmd_state,
cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL);
free_xml(now_dc_lrmd_state);
crm_debug("Updating local node history for join-%d "
"from query result (via CIB call %d)", join_id, call_id);
} else {
fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml,
cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL);
crm_warn("Updating local node history from join-%d confirmation "
"because query failed (via CIB call %d)", join_id, call_id);
}
} else {
fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml,
cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL);
crm_debug("Updating node history for %s from join-%d confirmation "
"(via CIB call %d)", join_from, join_id, call_id);
}
fsa_register_cib_callback(call_id, FALSE, NULL, join_update_complete_callback);
}
void
finalize_join_for(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *acknak = NULL;
xmlNode *tmp1 = NULL;
crm_node_t *join_node = value;
const char *join_to = join_node->uname;
if(join_node->join != crm_join_integrated) {
crm_trace("Not updating non-integrated node %s (%s) for join-%d",
join_to, crm_join_phase_str(join_node->join),
current_join_id);
return;
}
crm_trace("Updating node state for %s", join_to);
tmp1 = create_xml_node(NULL, XML_CIB_TAG_NODE);
set_uuid(tmp1, XML_ATTR_UUID, join_node);
crm_xml_add(tmp1, XML_ATTR_UNAME, join_to);
fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1);
free_xml(tmp1);
join_node = crm_get_peer(0, join_to);
if (crm_is_peer_active(join_node) == FALSE) {
/*
* NACK'ing nodes that the membership layer doesn't know about yet
* simply creates more churn
*
* Better to leave them waiting and let the join restart when
* the new membership event comes in
*
* All other NACKs (due to versions etc) should still be processed
*/
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_PENDING);
return;
}
// Acknowledge node's join request
crm_debug("Acknowledging join-%d request from %s",
current_join_id, join_to);
acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to);
crm_xml_add(acknak, CRM_OP_JOIN_ACKNAK, XML_BOOLEAN_TRUE);
crm_update_peer_join(__FUNCTION__, join_node, crm_join_finalized);
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_MEMBER);
send_cluster_message(crm_get_peer(0, join_to), crm_msg_crmd, acknak, TRUE);
free_xml(acknak);
return;
}
gboolean
check_join_state(enum crmd_fsa_state cur_state, const char *source)
{
static unsigned long long highest_seq = 0;
if (saved_ccm_membership_id != crm_peer_seq) {
crm_debug("join-%d: Membership changed from %llu to %llu "
CRM_XS " highest=%llu state=%s for=%s",
current_join_id, saved_ccm_membership_id, crm_peer_seq, highest_seq,
fsa_state2string(cur_state), source);
if(highest_seq < crm_peer_seq) {
/* Don't spam the FSA with duplicates */
highest_seq = crm_peer_seq;
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
} else if (cur_state == S_INTEGRATION) {
if (crmd_join_phase_count(crm_join_welcomed) == 0) {
int count = crmd_join_phase_count(crm_join_integrated);
crm_debug("join-%d: Integration of %d peer%s complete "
CRM_XS " state=%s for=%s",
current_join_id, count, pcmk__plural_s(count),
fsa_state2string(cur_state), source);
register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL);
return TRUE;
}
} else if (cur_state == S_FINALIZE_JOIN) {
if (is_set(fsa_input_register, R_HAVE_CIB) == FALSE) {
crm_debug("join-%d: Delaying finalization until we have CIB "
CRM_XS " state=%s for=%s",
current_join_id, fsa_state2string(cur_state), source);
return TRUE;
} else if (crmd_join_phase_count(crm_join_welcomed) != 0) {
int count = crmd_join_phase_count(crm_join_welcomed);
crm_debug("join-%d: Still waiting on %d welcomed node%s "
CRM_XS " state=%s for=%s",
current_join_id, count, pcmk__plural_s(count),
fsa_state2string(cur_state), source);
crmd_join_phase_log(LOG_DEBUG);
} else if (crmd_join_phase_count(crm_join_integrated) != 0) {
int count = crmd_join_phase_count(crm_join_integrated);
crm_debug("join-%d: Still waiting on %d integrated node%s "
CRM_XS " state=%s for=%s",
current_join_id, count, pcmk__plural_s(count),
fsa_state2string(cur_state), source);
crmd_join_phase_log(LOG_DEBUG);
} else if (crmd_join_phase_count(crm_join_finalized) != 0) {
int count = crmd_join_phase_count(crm_join_finalized);
crm_debug("join-%d: Still waiting on %d finalized node%s "
CRM_XS " state=%s for=%s",
current_join_id, count, pcmk__plural_s(count),
fsa_state2string(cur_state), source);
crmd_join_phase_log(LOG_DEBUG);
} else {
crm_debug("join-%d: Complete " CRM_XS " state=%s for=%s",
current_join_id, fsa_state2string(cur_state), source);
register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL);
return TRUE;
}
}
return FALSE;
}
void
do_dc_join_final(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_debug("Ensuring DC, quorum and node attributes are up-to-date");
crm_update_quorum(crm_have_quorum, TRUE);
}
int crmd_join_phase_count(enum crm_join_phase phase)
{
int count = 0;
crm_node_t *peer;
GHashTableIter iter;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
if(peer->join == phase) {
count++;
}
}
return count;
}
void crmd_join_phase_log(int level)
{
crm_node_t *peer;
GHashTableIter iter;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->uname,
crm_join_phase_str(peer->join));
}
}
diff --git a/daemons/controld/controld_lrm.h b/daemons/controld/controld_lrm.h
index 3ab70484b4..da0582c71b 100644
--- a/daemons/controld/controld_lrm.h
+++ b/daemons/controld/controld_lrm.h
@@ -1,171 +1,181 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
+#ifndef CONTROLD_LRM__H
+# define CONTROLD_LRM__H
#include
#include
extern gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level);
void lrm_clear_last_failure(const char *rsc_id, const char *node_name,
const char *operation, guint interval_ms);
void lrm_op_callback(lrmd_event_data_t * op);
lrmd_t *crmd_local_lrmd_conn(void);
typedef struct resource_history_s {
char *id;
uint32_t last_callid;
lrmd_rsc_info_t rsc;
lrmd_event_data_t *last;
lrmd_event_data_t *failed;
GList *recurring_op_list;
/* Resources must be stopped using the same
* parameters they were started with. This hashtable
* holds the parameters that should be used for the next stop
* cmd on this resource. */
GHashTable *stop_params;
} rsc_history_t;
void history_free(gpointer data);
enum active_op_e {
active_op_remove = (1 << 0),
active_op_cancelled = (1 << 1),
};
// In-flight action (recurring or pending)
typedef struct active_op_s {
guint interval_ms;
int call_id;
uint32_t flags; // bitmask of active_op_e
time_t start_time;
+ time_t lock_time;
char *rsc_id;
char *op_type;
char *op_key;
char *user_data;
GHashTable *params;
} active_op_t;
typedef struct lrm_state_s {
const char *node_name;
void *conn; // Reserved for controld_execd_state.c usage
void *remote_ra_data; // Reserved for controld_remote_ra.c usage
GHashTable *resource_history;
GHashTable *pending_ops;
GHashTable *deletion_ops;
GHashTable *rsc_info_cache;
GHashTable *metadata_cache; // key = class[:provider]:agent, value = ra_metadata_s
int num_lrm_register_fails;
} lrm_state_t;
struct pending_deletion_op_s {
char *rsc;
ha_msg_input_t *input;
};
/*!
* \brief Check whether this the local IPC connection to the executor
*/
gboolean
lrm_state_is_local(lrm_state_t *lrm_state);
/*!
* \brief Clear all state information from a single state entry.
* \note It sometimes useful to save metadata cache when it won't go stale.
* \note This does not close the executor connection
*/
void lrm_state_reset_tables(lrm_state_t * lrm_state, gboolean reset_metadata);
GList *lrm_state_get_list(void);
/*!
* \brief Initiate internal state tables
*/
gboolean lrm_state_init_local(void);
/*!
* \brief Destroy all state entries and internal state tables
*/
void lrm_state_destroy_all(void);
/*!
* \brief Create executor connection entry
*/
lrm_state_t *lrm_state_create(const char *node_name);
/*!
* \brief Destroy executor connection by node name
*/
void lrm_state_destroy(const char *node_name);
/*!
* \brief Find lrm_state data by node name
*/
lrm_state_t *lrm_state_find(const char *node_name);
/*!
* \brief Either find or create a new entry
*/
lrm_state_t *lrm_state_find_or_create(const char *node_name);
/*!
* The functions below are wrappers for the executor API the the controller
* uses. These wrapper functions allow us to treat the controller's remote
* executor connection resources the same as regular resources. Internally,
* regular resources go to the executor, and remote connection resources are
* handled locally in the controller.
*/
void lrm_state_disconnect_only(lrm_state_t * lrm_state);
void lrm_state_disconnect(lrm_state_t * lrm_state);
int lrm_state_ipc_connect(lrm_state_t * lrm_state);
int lrm_state_remote_connect_async(lrm_state_t * lrm_state, const char *server, int port,
int timeout);
int lrm_state_is_connected(lrm_state_t * lrm_state);
int lrm_state_poke_connection(lrm_state_t * lrm_state);
int lrm_state_get_metadata(lrm_state_t * lrm_state,
const char *class,
const char *provider,
const char *agent, char **output, enum lrmd_call_options options);
int lrm_state_cancel(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, guint interval_ms);
int lrm_state_exec(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, const char *userdata, guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
lrmd_key_value_t * params);
lrmd_rsc_info_t *lrm_state_get_rsc_info(lrm_state_t * lrm_state,
const char *rsc_id, enum lrmd_call_options options);
int lrm_state_register_rsc(lrm_state_t * lrm_state,
const char *rsc_id,
const char *class,
const char *provider, const char *agent, enum lrmd_call_options options);
int lrm_state_unregister_rsc(lrm_state_t * lrm_state,
const char *rsc_id, enum lrmd_call_options options);
// Functions used to manage remote executor connection resources
void remote_lrm_op_callback(lrmd_event_data_t * op);
gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id);
lrmd_rsc_info_t *remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id);
int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, guint interval_ms);
int remote_ra_exec(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, const char *userdata, guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
lrmd_key_value_t * params);
void remote_ra_cleanup(lrm_state_t * lrm_state);
void remote_ra_fail(const char *node_name);
void remote_ra_process_pseudo(xmlNode *xml);
gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state);
void remote_ra_process_maintenance_nodes(xmlNode *xml);
gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state);
void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op,
active_op_t *pending, xmlNode *action_xml);
+void controld_ack_event_directly(const char *to_host, const char *to_sys,
+ lrmd_rsc_info_t *rsc, lrmd_event_data_t *op,
+ const char *rsc_id);
+void controld_rc2event(lrmd_event_data_t *event, int rc);
+void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id);
+
+#endif
diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c
index 466c64c778..689e4a0f09 100644
--- a/daemons/controld/controld_messages.c
+++ b/daemons/controld/controld_messages.c
@@ -1,1167 +1,1258 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
GListPtr fsa_message_queue = NULL;
extern void crm_shutdown(int nsig);
void handle_response(xmlNode * stored_msg);
enum crmd_fsa_input handle_request(xmlNode * stored_msg, enum crmd_fsa_cause cause);
enum crmd_fsa_input handle_shutdown_request(xmlNode * stored_msg);
#define ROUTER_RESULT(x) crm_trace("Router result: %s", x)
/* debug only, can wrap all it likes */
int last_data_id = 0;
void
register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input,
fsa_data_t * cur_data, void *new_data, const char *raised_from)
{
/* save the current actions if any */
if (fsa_actions != A_NOTHING) {
register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL,
I_NULL, cur_data ? cur_data->data : NULL,
fsa_actions, TRUE, __FUNCTION__);
}
/* reset the action list */
crm_info("Resetting the current action list");
fsa_dump_actions(fsa_actions, "Drop");
fsa_actions = A_NOTHING;
/* register the error */
register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from);
}
int
register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input,
void *data, long long with_actions,
gboolean prepend, const char *raised_from)
{
unsigned old_len = g_list_length(fsa_message_queue);
fsa_data_t *fsa_data = NULL;
if (raised_from == NULL) {
raised_from = "";
}
if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) {
/* no point doing anything */
crm_err("Cannot add entry to queue: no input and no action");
return 0;
}
if (input == I_WAIT_FOR_EVENT) {
do_fsa_stall = TRUE;
crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d",
raised_from, fsa_cause2string(cause), data, old_len);
if (old_len > 0) {
fsa_dump_queue(LOG_TRACE);
prepend = FALSE;
}
if (data == NULL) {
fsa_actions |= with_actions;
fsa_dump_actions(with_actions, "Restored");
return 0;
}
/* Store everything in the new event and reset fsa_actions */
with_actions |= fsa_actions;
fsa_actions = A_NOTHING;
}
last_data_id++;
crm_trace("%s %s FSA input %d (%s) (cause=%s) %s data",
raised_from, prepend ? "prepended" : "appended", last_data_id,
fsa_input2string(input), fsa_cause2string(cause), data ? "with" : "without");
fsa_data = calloc(1, sizeof(fsa_data_t));
fsa_data->id = last_data_id;
fsa_data->fsa_input = input;
fsa_data->fsa_cause = cause;
fsa_data->origin = raised_from;
fsa_data->data = NULL;
fsa_data->data_type = fsa_dt_none;
fsa_data->actions = with_actions;
if (with_actions != A_NOTHING) {
crm_trace("Adding actions %.16llx to input", with_actions);
}
if (data != NULL) {
switch (cause) {
case C_FSA_INTERNAL:
case C_CRMD_STATUS_CALLBACK:
case C_IPC_MESSAGE:
case C_HA_MESSAGE:
crm_trace("Copying %s data from %s as a HA msg",
fsa_cause2string(cause), raised_from);
CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL,
crm_err("Bogus data from %s", raised_from));
fsa_data->data = copy_ha_msg_input(data);
fsa_data->data_type = fsa_dt_ha_msg;
break;
case C_LRM_OP_CALLBACK:
crm_trace("Copying %s data from %s as lrmd_event_data_t",
fsa_cause2string(cause), raised_from);
fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data);
fsa_data->data_type = fsa_dt_lrm;
break;
case C_TIMER_POPPED:
case C_SHUTDOWN:
case C_UNKNOWN:
case C_STARTUP:
crm_err("Copying %s data (from %s)"
" not yet implemented", fsa_cause2string(cause), raised_from);
crmd_exit(CRM_EX_SOFTWARE);
break;
}
crm_trace("%s data copied", fsa_cause2string(fsa_data->fsa_cause));
}
/* make sure to free it properly later */
if (prepend) {
crm_trace("Prepending input");
fsa_message_queue = g_list_prepend(fsa_message_queue, fsa_data);
} else {
fsa_message_queue = g_list_append(fsa_message_queue, fsa_data);
}
crm_trace("Queue len: %d", g_list_length(fsa_message_queue));
/* fsa_dump_queue(LOG_TRACE); */
if (old_len == g_list_length(fsa_message_queue)) {
crm_err("Couldn't add message to the queue");
}
if (fsa_source && input != I_WAIT_FOR_EVENT) {
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
}
return last_data_id;
}
void
fsa_dump_queue(int log_level)
{
int offset = 0;
GListPtr lpc = NULL;
for (lpc = fsa_message_queue; lpc != NULL; lpc = lpc->next) {
fsa_data_t *data = (fsa_data_t *) lpc->data;
do_crm_log_unlikely(log_level,
"queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)",
offset++, data->id, fsa_input2string(data->fsa_input),
data->origin, data->data, data->data_type,
fsa_cause2string(data->fsa_cause));
}
}
ha_msg_input_t *
copy_ha_msg_input(ha_msg_input_t * orig)
{
ha_msg_input_t *copy = NULL;
xmlNodePtr data = NULL;
if (orig != NULL) {
crm_trace("Copy msg");
data = copy_xml(orig->msg);
} else {
crm_trace("No message to copy");
}
copy = new_ha_msg_input(data);
if (orig && orig->msg != NULL) {
CRM_CHECK(copy->msg != NULL, crm_err("copy failed"));
}
return copy;
}
void
delete_fsa_input(fsa_data_t * fsa_data)
{
lrmd_event_data_t *op = NULL;
xmlNode *foo = NULL;
if (fsa_data == NULL) {
return;
}
crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause));
if (fsa_data->data != NULL) {
switch (fsa_data->data_type) {
case fsa_dt_ha_msg:
delete_ha_msg_input(fsa_data->data);
break;
case fsa_dt_xml:
foo = fsa_data->data;
free_xml(foo);
break;
case fsa_dt_lrm:
op = (lrmd_event_data_t *) fsa_data->data;
lrmd_free_event(op);
break;
case fsa_dt_none:
if (fsa_data->data != NULL) {
crm_err("Don't know how to free %s data from %s",
fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin);
crmd_exit(CRM_EX_SOFTWARE);
}
break;
}
crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause));
}
free(fsa_data);
}
/* returns the next message */
fsa_data_t *
get_message(void)
{
fsa_data_t *message = g_list_nth_data(fsa_message_queue, 0);
fsa_message_queue = g_list_remove(fsa_message_queue, message);
crm_trace("Processing input %d", message->id);
return message;
}
/* returns the current head of the FIFO queue */
gboolean
is_message(void)
{
return fsa_message_queue != NULL;
}
void *
fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller)
{
void *ret_val = NULL;
if (fsa_data == NULL) {
crm_err("%s: No FSA data available", caller);
} else if (fsa_data->data == NULL) {
crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin);
} else if (fsa_data->data_type != a_type) {
crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s",
caller, fsa_data->data_type, a_type, fsa_data->origin);
CRM_ASSERT(fsa_data->data_type == a_type);
} else {
ret_val = fsa_data->data;
}
return ret_val;
}
/* A_MSG_ROUTE */
void
do_msg_route(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
route_message(msg_data->fsa_cause, input->msg);
}
void
route_message(enum crmd_fsa_cause cause, xmlNode * input)
{
ha_msg_input_t fsa_input;
enum crmd_fsa_input result = I_NULL;
fsa_input.msg = input;
CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return);
/* try passing the buck first */
if (relay_message(input, cause == C_IPC_MESSAGE)) {
return;
}
/* handle locally */
result = handle_message(input, cause);
/* done or process later? */
switch (result) {
case I_NULL:
case I_CIB_OP:
case I_ROUTER:
case I_NODE_JOIN:
case I_JOIN_REQUEST:
case I_JOIN_RESULT:
break;
default:
/* Defering local processing of message */
register_fsa_input_later(cause, result, &fsa_input);
return;
}
if (result != I_NULL) {
/* add to the front of the queue */
register_fsa_input(cause, result, &fsa_input);
}
}
gboolean
relay_message(xmlNode * msg, gboolean originated_locally)
{
int dest = 1;
int is_for_dc = 0;
int is_for_dcib = 0;
int is_for_te = 0;
int is_for_crm = 0;
int is_for_cib = 0;
int is_local = 0;
gboolean processing_complete = FALSE;
const char *host_to = crm_element_value(msg, F_CRM_HOST_TO);
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM);
const char *type = crm_element_value(msg, F_TYPE);
const char *task = crm_element_value(msg, F_CRM_TASK);
const char *msg_error = NULL;
crm_trace("Routing message %s", crm_element_value(msg, XML_ATTR_REFERENCE));
if (msg == NULL) {
msg_error = "Cannot route empty message";
} else if (safe_str_eq(task, CRM_OP_HELLO)) {
/* quietly ignore */
processing_complete = TRUE;
} else if (safe_str_neq(type, T_CRM)) {
msg_error = "Bad message type";
} else if (sys_to == NULL) {
msg_error = "Bad message destination: no subsystem";
}
if (msg_error != NULL) {
processing_complete = TRUE;
crm_err("%s", msg_error);
crm_log_xml_warn(msg, "bad msg");
}
if (processing_complete) {
return TRUE;
}
processing_complete = TRUE;
is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0);
is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0);
is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0);
is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0);
is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0);
is_local = 0;
if (host_to == NULL || strlen(host_to) == 0) {
if (is_for_dc || is_for_te) {
is_local = 0;
} else if (is_for_crm) {
if (safe_str_eq(task, CRM_OP_NODE_INFO)) {
/* Node info requests do not specify a host, which is normally
* treated as "all hosts", because the whole point is that the
* client doesn't know the local node name. Always handle these
* requests locally.
*/
is_local = 1;
} else {
is_local = !originated_locally;
}
} else {
is_local = 1;
}
} else if (safe_str_eq(fsa_our_uname, host_to)) {
is_local = 1;
+ } else if (is_for_crm && safe_str_eq(task, CRM_OP_LRM_DELETE)) {
+ xmlNode *msg_data = get_message_xml(msg, F_CRM_DATA);
+ const char *mode = crm_element_value(msg_data, PCMK__XA_MODE);
+
+ if (safe_str_eq(mode, XML_TAG_CIB)) {
+ // Local delete of an offline node's resource history
+ is_local = 1;
+ }
}
if (is_for_dc || is_for_dcib || is_for_te) {
if (AM_I_DC && is_for_te) {
ROUTER_RESULT("Message result: Local relay");
send_msg_via_ipc(msg, sys_to);
} else if (AM_I_DC) {
ROUTER_RESULT("Message result: DC/controller process");
processing_complete = FALSE; /* more to be done by caller */
} else if (originated_locally && safe_str_neq(sys_from, CRM_SYSTEM_PENGINE)
&& safe_str_neq(sys_from, CRM_SYSTEM_TENGINE)) {
/* Neither the TE nor the scheduler should be sending messages
* to DCs on other nodes. By definition, if we are no longer the DC,
* then the scheduler's or TE's data should be discarded.
*/
#if SUPPORT_COROSYNC
if (is_corosync_cluster()) {
dest = text2msg_type(sys_to);
}
#endif
ROUTER_RESULT("Message result: External relay to DC");
send_cluster_message(host_to ? crm_get_peer(0, host_to) : NULL, dest, msg, TRUE);
} else {
/* discard */
ROUTER_RESULT("Message result: Discard, not DC");
}
} else if (is_local && (is_for_crm || is_for_cib)) {
ROUTER_RESULT("Message result: controller process");
processing_complete = FALSE; /* more to be done by caller */
} else if (is_local) {
ROUTER_RESULT("Message result: Local relay");
send_msg_via_ipc(msg, sys_to);
} else {
crm_node_t *node_to = NULL;
#if SUPPORT_COROSYNC
if (is_corosync_cluster()) {
dest = text2msg_type(sys_to);
if (dest == crm_msg_none || dest > crm_msg_stonith_ng) {
dest = crm_msg_crmd;
}
}
#endif
if (host_to) {
node_to = crm_find_peer(0, host_to);
if (node_to == NULL) {
crm_err("Cannot route message to unknown node %s", host_to);
return TRUE;
}
}
ROUTER_RESULT("Message result: External relay");
send_cluster_message(host_to ? node_to : NULL, dest, msg, TRUE);
}
return processing_complete;
}
static gboolean
process_hello_message(xmlNode * hello,
char **client_name, char **major_version, char **minor_version)
{
const char *local_client_name;
const char *local_major_version;
const char *local_minor_version;
*client_name = NULL;
*major_version = NULL;
*minor_version = NULL;
if (hello == NULL) {
return FALSE;
}
local_client_name = crm_element_value(hello, "client_name");
local_major_version = crm_element_value(hello, "major_version");
local_minor_version = crm_element_value(hello, "minor_version");
if (local_client_name == NULL || strlen(local_client_name) == 0) {
crm_err("Hello message was not valid (field %s not found)", "client name");
return FALSE;
} else if (local_major_version == NULL || strlen(local_major_version) == 0) {
crm_err("Hello message was not valid (field %s not found)", "major version");
return FALSE;
} else if (local_minor_version == NULL || strlen(local_minor_version) == 0) {
crm_err("Hello message was not valid (field %s not found)", "minor version");
return FALSE;
}
*client_name = strdup(local_client_name);
*major_version = strdup(local_major_version);
*minor_version = strdup(local_minor_version);
crm_trace("Hello message ok");
return TRUE;
}
gboolean
crmd_authorize_message(xmlNode * client_msg, crm_client_t * curr_client, const char *proxy_session)
{
char *client_name = NULL;
char *major_version = NULL;
char *minor_version = NULL;
gboolean auth_result = FALSE;
xmlNode *xml = NULL;
const char *op = crm_element_value(client_msg, F_CRM_TASK);
const char *uuid = curr_client ? curr_client->id : proxy_session;
if (uuid == NULL) {
crm_warn("Message [%s] not authorized", crm_element_value(client_msg, XML_ATTR_REFERENCE));
return FALSE;
} else if (safe_str_neq(CRM_OP_HELLO, op)) {
return TRUE;
}
xml = get_message_xml(client_msg, F_CRM_DATA);
auth_result = process_hello_message(xml, &client_name, &major_version, &minor_version);
if (auth_result == TRUE) {
if (client_name == NULL) {
crm_err("Bad client details (client_name=%s, uuid=%s)",
crm_str(client_name), uuid);
auth_result = FALSE;
}
}
if (auth_result == TRUE) {
/* check version */
int mav = atoi(major_version);
int miv = atoi(minor_version);
crm_trace("Checking client version number");
if (mav < 0 || miv < 0) {
crm_err("Client version (%d:%d) is not acceptable", mav, miv);
auth_result = FALSE;
}
}
if (auth_result == TRUE) {
crm_trace("Accepted client %s", client_name);
if (curr_client) {
curr_client->userdata = strdup(client_name);
}
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
} else {
crm_warn("Rejected client logon request");
if (curr_client) {
qb_ipcs_disconnect(curr_client->ipcs);
}
}
free(minor_version);
free(major_version);
free(client_name);
/* hello messages should never be processed further */
return FALSE;
}
enum crmd_fsa_input
handle_message(xmlNode * msg, enum crmd_fsa_cause cause)
{
const char *type = NULL;
CRM_CHECK(msg != NULL, return I_NULL);
type = crm_element_value(msg, F_CRM_MSG_TYPE);
if (crm_str_eq(type, XML_ATTR_REQUEST, TRUE)) {
return handle_request(msg, cause);
} else if (crm_str_eq(type, XML_ATTR_RESPONSE, TRUE)) {
handle_response(msg);
return I_NULL;
}
crm_err("Unknown message type: %s", type);
return I_NULL;
}
static enum crmd_fsa_input
handle_failcount_op(xmlNode * stored_msg)
{
const char *rsc = NULL;
const char *uname = NULL;
const char *op = NULL;
const char *interval_ms_s = NULL;
char *interval_spec = NULL;
guint interval_ms = 0;
gboolean is_remote_node = FALSE;
xmlNode *xml_op = get_message_xml(stored_msg, F_CRM_DATA);
if (xml_op) {
xmlNode *xml_rsc = first_named_child(xml_op, XML_CIB_TAG_RESOURCE);
xmlNode *xml_attrs = first_named_child(xml_op, XML_TAG_ATTRS);
if (xml_rsc) {
rsc = ID(xml_rsc);
}
if (xml_attrs) {
op = crm_element_value(xml_attrs,
CRM_META "_" XML_RSC_ATTR_CLEAR_OP);
interval_ms_s = crm_element_value(xml_attrs,
CRM_META "_" XML_RSC_ATTR_CLEAR_INTERVAL);
interval_ms = crm_parse_ms(interval_ms_s);
}
}
uname = crm_element_value(xml_op, XML_LRM_ATTR_TARGET);
if ((rsc == NULL) || (uname == NULL)) {
crm_log_xml_warn(stored_msg, "invalid failcount op");
return I_NULL;
}
if (crm_element_value(xml_op, XML_LRM_ATTR_ROUTER_NODE)) {
is_remote_node = TRUE;
}
if (interval_ms) {
interval_spec = crm_strdup_printf("%ums", interval_ms);
}
update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node);
free(interval_spec);
lrm_clear_last_failure(rsc, uname, op, interval_ms);
return I_NULL;
}
+static enum crmd_fsa_input
+handle_lrm_delete(xmlNode *stored_msg)
+{
+ const char *mode = NULL;
+ xmlNode *msg_data = get_message_xml(stored_msg, F_CRM_DATA);
+
+ CRM_CHECK(msg_data != NULL, return I_NULL);
+
+ /* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to
+ * relay the operation to the affected node, which will unregister the
+ * resource from the local executor, clear the resource's history from the
+ * CIB, and do some bookkeeping in the controller.
+ *
+ * However, if the affected node is offline, the client will specify
+ * mode="cib" which means the controller receiving the operation should
+ * clear the resource's history from the CIB and nothing else. This is used
+ * to clear shutdown locks.
+ */
+ mode = crm_element_value(msg_data, PCMK__XA_MODE);
+ if ((mode == NULL) || strcmp(mode, XML_TAG_CIB)) {
+ // Relay to affected node
+ crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD);
+ return I_ROUTER;
+
+ } else {
+ // Delete CIB history locally (compare with do_lrm_delete())
+ const char *from_sys = NULL;
+ const char *user_name = NULL;
+ const char *rsc_id = NULL;
+ const char *node = NULL;
+ xmlNode *rsc_xml = NULL;
+ int rc = pcmk_rc_ok;
+
+ rsc_xml = first_named_child(msg_data, XML_CIB_TAG_RESOURCE);
+ CRM_CHECK(rsc_xml != NULL, return I_NULL);
+
+ rsc_id = ID(rsc_xml);
+ from_sys = crm_element_value(stored_msg, F_CRM_SYS_FROM);
+ node = crm_element_value(msg_data, XML_LRM_ATTR_TARGET);
+#if ENABLE_ACL
+ user_name = crm_acl_get_set_user(stored_msg, F_CRM_USER, NULL);
+#endif
+ crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s "
+ "(clearing CIB resource history only)", rsc_id, node,
+ (user_name? " for user " : ""), (user_name? user_name : ""));
+#if ENABLE_ACL
+ rc = controld_delete_resource_history(rsc_id, node, user_name,
+ cib_dryrun|cib_sync_call);
+#endif
+ if (rc == pcmk_rc_ok) {
+ rc = controld_delete_resource_history(rsc_id, node, user_name,
+ crmd_cib_smart_opt());
+ }
+
+ // Notify client if not from graph (compare with notify_deleted())
+ if (from_sys && strcmp(from_sys, CRM_SYSTEM_TENGINE)) {
+ lrmd_event_data_t *op = NULL;
+ const char *from_host = crm_element_value(stored_msg,
+ F_CRM_HOST_FROM);
+ const char *transition = crm_element_value(msg_data,
+ XML_ATTR_TRANSITION_KEY);
+
+ crm_info("Notifying %s on %s that %s was%s deleted",
+ from_sys, (from_host? from_host : "local node"), rsc_id,
+ ((rc == pcmk_rc_ok)? "" : " not"));
+ op = lrmd_new_event(rsc_id, CRMD_ACTION_DELETE, 0);
+ op->type = lrmd_event_exec_complete;
+ op->user_data = strdup(transition? transition : FAKE_TE_ID);
+ op->params = crm_str_table_new();
+ g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION),
+ strdup(CRM_FEATURE_SET));
+ controld_rc2event(op, rc);
+ controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id);
+ lrmd_free_event(op);
+ controld_trigger_delete_refresh(from_sys, rsc_id);
+ }
+ return I_NULL;
+ }
+}
+
/*!
* \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_remote_state(xmlNode *msg)
{
const char *remote_uname = ID(msg);
const char *remote_is_up = crm_element_value(msg, XML_NODE_IN_CLUSTER);
crm_node_t *remote_peer;
CRM_CHECK(remote_uname && remote_is_up, return I_NULL);
remote_peer = crm_remote_peer_get(remote_uname);
CRM_CHECK(remote_peer, return I_NULL);
crm_update_peer_state(__FUNCTION__, remote_peer,
crm_is_true(remote_is_up)?
CRM_NODE_MEMBER : CRM_NODE_LOST, 0);
return I_NULL;
}
/*!
* \brief Handle a CRM_OP_PING message
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_ping(xmlNode *msg)
{
const char *value = NULL;
xmlNode *ping = NULL;
// Build reply
ping = create_xml_node(NULL, XML_CRM_TAG_PING);
value = crm_element_value(msg, F_CRM_SYS_TO);
crm_xml_add(ping, XML_PING_ATTR_SYSFROM, value);
// Add controller state
value = fsa_state2string(fsa_state);
crm_xml_add(ping, XML_PING_ATTR_CRMDSTATE, value);
crm_notice("Current ping state: %s", value); // CTS needs this
// Add controller health
// @TODO maybe do some checks to determine meaningful status
crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok");
// Send reply
msg = create_reply(msg, ping);
free_xml(ping);
if (msg) {
(void) relay_message(msg, TRUE);
free_xml(msg);
}
// Nothing further to do
return I_NULL;
}
/*!
* \brief Handle a CRM_OP_NODE_INFO request
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_node_info_request(xmlNode *msg)
{
const char *value = NULL;
crm_node_t *node = NULL;
int node_id = 0;
xmlNode *reply = NULL;
// Build reply
reply = create_xml_node(NULL, XML_CIB_TAG_NODE);
crm_xml_add(reply, XML_PING_ATTR_SYSFROM, CRM_SYSTEM_CRMD);
// Add whether current partition has quorum
crm_xml_add_boolean(reply, XML_ATTR_HAVE_QUORUM, fsa_has_quorum);
// Check whether client requested node info by ID and/or name
crm_element_value_int(msg, XML_ATTR_ID, &node_id);
if (node_id < 0) {
node_id = 0;
}
value = crm_element_value(msg, XML_ATTR_UNAME);
// Default to local node if none given
if ((node_id == 0) && (value == NULL)) {
value = fsa_our_uname;
}
node = crm_find_peer_full(node_id, value, CRM_GET_PEER_ANY);
if (node) {
crm_xml_add_int(reply, XML_ATTR_ID, node->id);
crm_xml_add(reply, XML_ATTR_UUID, node->uuid);
crm_xml_add(reply, XML_ATTR_UNAME, node->uname);
crm_xml_add(reply, XML_NODE_IS_PEER, node->state);
crm_xml_add_boolean(reply, XML_NODE_IS_REMOTE,
node->flags & crm_remote_node);
}
// Send reply
msg = create_reply(msg, reply);
free_xml(reply);
if (msg) {
(void) relay_message(msg, TRUE);
free_xml(msg);
}
// Nothing further to do
return I_NULL;
}
static void
verify_feature_set(xmlNode *msg)
{
const char *dc_version = crm_element_value(msg, XML_ATTR_CRM_VERSION);
if (dc_version == NULL) {
/* All we really know is that the DC feature set is older than 3.1.0,
* but that's also all that really matters.
*/
dc_version = "3.0.14";
}
if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) {
crm_trace("Local feature set (%s) is compatible with DC's (%s)",
CRM_FEATURE_SET, dc_version);
} else {
crm_err("Local feature set (%s) is incompatible with DC's (%s)",
CRM_FEATURE_SET, dc_version);
// Nothing is likely to improve without administrator involvement
set_bit(fsa_input_register, R_STAYDOWN);
crmd_exit(CRM_EX_FATAL);
}
}
enum crmd_fsa_input
handle_request(xmlNode * stored_msg, enum crmd_fsa_cause cause)
{
xmlNode *msg = NULL;
const char *op = crm_element_value(stored_msg, F_CRM_TASK);
/* Optimize this for the DC - it has the most to do */
if (op == NULL) {
crm_log_xml_err(stored_msg, "Bad message");
return I_NULL;
}
if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
const char *from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
crm_node_t *node = crm_find_peer(0, from);
crm_update_peer_expected(__FUNCTION__, node, CRMD_JOINSTATE_DOWN);
if(AM_I_DC == FALSE) {
return I_NULL; /* Done */
}
}
/*========== DC-Only Actions ==========*/
if (AM_I_DC) {
if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) {
return I_NODE_JOIN;
} else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) {
return I_JOIN_REQUEST;
} else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) {
return I_JOIN_RESULT;
} else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) {
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
gboolean dc_match = safe_str_eq(host_from, fsa_our_dc);
if (is_set(fsa_input_register, R_SHUTDOWN)) {
crm_info("Shutting ourselves down (DC)");
return I_STOP;
} else if (dc_match) {
crm_err("We didn't ask to be shut down, yet our"
" TE is telling us to. Better get out now!");
return I_TERMINATE;
} else if (fsa_state != S_STOPPING) {
crm_err("Another node is asking us to shutdown" " but we think we're ok.");
return I_ELECTION;
}
} else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
/* a slave wants to shut down */
/* create cib fragment and add to message */
return handle_shutdown_request(stored_msg);
} else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) {
/* a remote connection host is letting us know the node state */
return handle_remote_state(stored_msg);
}
}
/*========== common actions ==========*/
if (strcmp(op, CRM_OP_NOVOTE) == 0) {
ha_msg_input_t fsa_input;
fsa_input.msg = stored_msg;
register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input,
A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__);
} else if (strcmp(op, CRM_OP_THROTTLE) == 0) {
throttle_update(stored_msg);
if (AM_I_DC && transition_graph != NULL) {
if (transition_graph->complete == FALSE) {
crm_debug("The throttle changed. Trigger a graph.");
trigger_graph();
}
}
return I_NULL;
} else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) {
return handle_failcount_op(stored_msg);
} else if (strcmp(op, CRM_OP_VOTE) == 0) {
/* count the vote and decide what to do after that */
ha_msg_input_t fsa_input;
fsa_input.msg = stored_msg;
register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input,
A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__);
/* Sometimes we _must_ go into S_ELECTION */
if (fsa_state == S_HALT) {
crm_debug("Forcing an election from S_HALT");
return I_ELECTION;
#if 0
} else if (AM_I_DC) {
/* This is the old way of doing things but what is gained? */
return I_ELECTION;
#endif
}
} else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) {
verify_feature_set(stored_msg);
crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID));
return I_JOIN_OFFER;
} else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) {
crm_debug("Raising I_JOIN_RESULT: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID));
return I_JOIN_RESULT;
- } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0
- || strcmp(op, CRM_OP_LRM_FAIL) == 0
- || strcmp(op, CRM_OP_LRM_REFRESH) == 0 || strcmp(op, CRM_OP_REPROBE) == 0) {
+ } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) {
+ return handle_lrm_delete(stored_msg);
+
+ } else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0)
+ || (strcmp(op, CRM_OP_LRM_REFRESH) == 0)
+ || (strcmp(op, CRM_OP_REPROBE) == 0)) {
crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD);
return I_ROUTER;
} else if (strcmp(op, CRM_OP_NOOP) == 0) {
return I_NULL;
} else if (strcmp(op, CRM_OP_LOCAL_SHUTDOWN) == 0) {
crm_shutdown(SIGTERM);
/*return I_SHUTDOWN; */
return I_NULL;
} else if (strcmp(op, CRM_OP_PING) == 0) {
return handle_ping(stored_msg);
} else if (strcmp(op, CRM_OP_NODE_INFO) == 0) {
return handle_node_info_request(stored_msg);
} else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) {
int id = 0;
const char *name = NULL;
crm_element_value_int(stored_msg, XML_ATTR_ID, &id);
name = crm_element_value(stored_msg, XML_ATTR_UNAME);
if(cause == C_IPC_MESSAGE) {
msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) {
crm_err("Could not instruct peers to remove references to node %s/%u", name, id);
} else {
crm_notice("Instructing peers to remove references to node %s/%u", name, id);
}
free_xml(msg);
} else {
reap_crm_member(id, name);
/* If we're forgetting this node, also forget any failures to fence
* it, so we don't carry that over to any node added later with the
* same name.
*/
st_fail_count_reset(name);
}
} else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) {
xmlNode *xml = get_message_xml(stored_msg, F_CRM_DATA);
remote_ra_process_maintenance_nodes(xml);
/*========== (NOT_DC)-Only Actions ==========*/
} else if (AM_I_DC == FALSE && strcmp(op, CRM_OP_SHUTDOWN) == 0) {
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
gboolean dc_match = safe_str_eq(host_from, fsa_our_dc);
if (dc_match || fsa_our_dc == NULL) {
if (is_set(fsa_input_register, R_SHUTDOWN) == FALSE) {
crm_err("We didn't ask to be shut down, yet our DC is telling us to.");
set_bit(fsa_input_register, R_STAYDOWN);
return I_STOP;
}
crm_info("Shutting down");
return I_STOP;
} else {
crm_warn("Discarding %s op from %s", op, host_from);
}
} else {
crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node");
crm_log_xml_err(stored_msg, "Unexpected");
}
return I_NULL;
}
void
handle_response(xmlNode * stored_msg)
{
const char *op = crm_element_value(stored_msg, F_CRM_TASK);
if (op == NULL) {
crm_log_xml_err(stored_msg, "Bad message");
} else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) {
// Check whether scheduler answer been superseded by subsequent request
const char *msg_ref = crm_element_value(stored_msg, XML_ATTR_REFERENCE);
if (msg_ref == NULL) {
crm_err("%s - Ignoring calculation with no reference", op);
} else if (safe_str_eq(msg_ref, fsa_pe_ref)) {
ha_msg_input_t fsa_input;
controld_stop_sched_timer();
fsa_input.msg = stored_msg;
register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input);
} else {
crm_info("%s calculation %s is obsolete", op, msg_ref);
}
} else if (strcmp(op, CRM_OP_VOTE) == 0
|| strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) {
} else {
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
crm_err("Unexpected response (op=%s, src=%s) sent to the %s",
op, host_from, AM_I_DC ? "DC" : "controller");
}
}
enum crmd_fsa_input
handle_shutdown_request(xmlNode * stored_msg)
{
/* handle here to avoid potential version issues
* where the shutdown message/procedure may have
* been changed in later versions.
*
* This way the DC is always in control of the shutdown
*/
char *now_s = NULL;
time_t now = time(NULL);
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
if (host_from == NULL) {
/* we're shutting down and the DC */
host_from = fsa_our_uname;
}
crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state));
crm_log_xml_trace(stored_msg, "message");
now_s = crm_itoa(now);
update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, FALSE);
free(now_s);
/* will be picked up by the TE as long as its running */
return I_NULL;
}
/* msg is deleted by the time this returns */
extern gboolean process_te_message(xmlNode * msg, xmlNode * xml_data);
gboolean
send_msg_via_ipc(xmlNode * msg, const char *sys)
{
gboolean send_ok = TRUE;
crm_client_t *client_channel = crm_client_get_by_id(sys);
if (crm_element_value(msg, F_CRM_HOST_FROM) == NULL) {
crm_xml_add(msg, F_CRM_HOST_FROM, fsa_our_uname);
}
if (client_channel != NULL) {
/* Transient clients such as crmadmin */
send_ok = crm_ipcs_send(client_channel, 0, msg, crm_ipc_server_event);
} else if (sys != NULL && strcmp(sys, CRM_SYSTEM_TENGINE) == 0) {
xmlNode *data = get_message_xml(msg, F_CRM_DATA);
process_te_message(msg, data);
} else if (sys != NULL && strcmp(sys, CRM_SYSTEM_LRMD) == 0) {
fsa_data_t fsa_data;
ha_msg_input_t fsa_input;
fsa_input.msg = msg;
fsa_input.xml = get_message_xml(msg, F_CRM_DATA);
fsa_data.id = 0;
fsa_data.actions = 0;
fsa_data.data = &fsa_input;
fsa_data.fsa_input = I_MESSAGE;
fsa_data.fsa_cause = C_IPC_MESSAGE;
fsa_data.origin = __FUNCTION__;
fsa_data.data_type = fsa_dt_ha_msg;
#ifdef FSA_TRACE
crm_trace("Invoking action A_LRM_INVOKE (%.16llx)", A_LRM_INVOKE);
#endif
do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, fsa_state, I_MESSAGE, &fsa_data);
} else if (sys != NULL && crmd_is_proxy_session(sys)) {
crmd_proxy_send(sys, msg);
} else {
crm_debug("Unknown Sub-system (%s)... discarding message.", crm_str(sys));
send_ok = FALSE;
}
return send_ok;
}
ha_msg_input_t *
new_ha_msg_input(xmlNode * orig)
{
ha_msg_input_t *input_copy = NULL;
input_copy = calloc(1, sizeof(ha_msg_input_t));
input_copy->msg = orig;
input_copy->xml = get_message_xml(input_copy->msg, F_CRM_DATA);
return input_copy;
}
void
delete_ha_msg_input(ha_msg_input_t * orig)
{
if (orig == NULL) {
return;
}
free_xml(orig->msg);
free(orig);
}
/*!
* \internal
* \brief Notify the DC of a remote node state change
*
* \param[in] node_name Node's name
* \param[in] node_up TRUE if node is up, FALSE if down
*/
void
send_remote_state_message(const char *node_name, gboolean node_up)
{
/* If we don't have a DC, or the message fails, we have a failsafe:
* the DC will eventually pick up the change via the CIB node state.
* The message allows it to happen sooner if possible.
*/
if (fsa_our_dc) {
xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_info("Notifying DC %s of pacemaker_remote node %s %s",
fsa_our_dc, node_name, (node_up? "coming up" : "going down"));
crm_xml_add(msg, XML_ATTR_ID, node_name);
crm_xml_add_boolean(msg, XML_NODE_IN_CLUSTER, node_up);
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, msg,
TRUE);
free_xml(msg);
} else {
crm_debug("No DC to notify of pacemaker_remote node %s %s",
node_name, (node_up? "coming up" : "going down"));
}
}
diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c
index 2d3dfa73ea..a81c354630 100644
--- a/daemons/controld/controld_remote_ra.c
+++ b/daemons/controld/controld_remote_ra.c
@@ -1,1295 +1,1299 @@
/*
- * Copyright 2013-2019 the Pacemaker project contributors
+ * Copyright 2013-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#define REMOTE_LRMD_RA "remote"
/* The max start timeout before cmd retry */
#define MAX_START_TIMEOUT_MS 10000
typedef struct remote_ra_cmd_s {
/*! the local node the cmd is issued from */
char *owner;
/*! the remote node the cmd is executed on */
char *rsc_id;
/*! the action to execute */
char *action;
/*! some string the client wants us to give it back */
char *userdata;
char *exit_reason; // descriptive text on error
/*! start delay in ms */
int start_delay;
/*! timer id used for start delay. */
int delay_id;
/*! timeout in ms for cmd */
int timeout;
int remaining_timeout;
/*! recurring interval in ms */
guint interval_ms;
/*! interval timer id */
int interval_id;
int reported_success;
int monitor_timeout_id;
int takeover_timeout_id;
/*! action parameters */
lrmd_key_value_t *params;
/*! executed rc */
int rc;
int op_status;
int call_id;
time_t start_time;
gboolean cancel;
} remote_ra_cmd_t;
enum remote_migration_status {
expect_takeover = 1,
takeover_complete,
};
typedef struct remote_ra_data_s {
crm_trigger_t *work;
remote_ra_cmd_t *cur_cmd;
GList *cmds;
GList *recurring_cmds;
enum remote_migration_status migrate_status;
gboolean active;
/* Maintenance mode is difficult to determine from the controller's context,
* so we have it signalled back with the transition from the scheduler.
*/
gboolean is_maintenance;
/* Similar for whether we are controlling a guest node or remote node.
* Fortunately there is a meta-attribute in the transition already and
* as the situation doesn't change over time we can use the
* resource start for noting down the information for later use when
* the attributes aren't at hand.
*/
gboolean controlling_guest;
} remote_ra_data_t;
static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms);
static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd);
static GList *fail_all_monitor_cmds(GList * list);
static void
free_cmd(gpointer user_data)
{
remote_ra_cmd_t *cmd = user_data;
if (!cmd) {
return;
}
if (cmd->delay_id) {
g_source_remove(cmd->delay_id);
}
if (cmd->interval_id) {
g_source_remove(cmd->interval_id);
}
if (cmd->monitor_timeout_id) {
g_source_remove(cmd->monitor_timeout_id);
}
if (cmd->takeover_timeout_id) {
g_source_remove(cmd->takeover_timeout_id);
}
free(cmd->owner);
free(cmd->rsc_id);
free(cmd->action);
free(cmd->userdata);
free(cmd->exit_reason);
lrmd_key_value_freeall(cmd->params);
free(cmd);
}
static int
generate_callid(void)
{
static int remote_ra_callid = 0;
remote_ra_callid++;
if (remote_ra_callid <= 0) {
remote_ra_callid = 1;
}
return remote_ra_callid;
}
static gboolean
recurring_helper(gpointer data)
{
remote_ra_cmd_t *cmd = data;
lrm_state_t *connection_rsc = NULL;
cmd->interval_id = 0;
connection_rsc = lrm_state_find(cmd->rsc_id);
if (connection_rsc && connection_rsc->remote_ra_data) {
remote_ra_data_t *ra_data = connection_rsc->remote_ra_data;
ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd);
ra_data->cmds = g_list_append(ra_data->cmds, cmd);
mainloop_set_trigger(ra_data->work);
}
return FALSE;
}
static gboolean
start_delay_helper(gpointer data)
{
remote_ra_cmd_t *cmd = data;
lrm_state_t *connection_rsc = NULL;
cmd->delay_id = 0;
connection_rsc = lrm_state_find(cmd->rsc_id);
if (connection_rsc && connection_rsc->remote_ra_data) {
remote_ra_data_t *ra_data = connection_rsc->remote_ra_data;
mainloop_set_trigger(ra_data->work);
}
return FALSE;
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node joining
*
* \param[in] node_name Name of newly integrated pacemaker_remote node
*/
static void
remote_node_up(const char *node_name)
{
int call_opt, call_id = 0;
xmlNode *update, *state;
crm_node_t *node;
+ enum controld_section_e section = controld_section_all;
CRM_CHECK(node_name != NULL, return);
crm_info("Announcing pacemaker_remote node %s", node_name);
- /* Clear node's entire state (resource history and transient attributes).
- * The transient attributes should and normally will be cleared when the
- * node leaves, but since remote node state has a number of corner cases,
- * clear them here as well, to be sure.
+ /* Clear node's entire state (resource history and transient attributes)
+ * other than shutdown locks. The transient attributes should and normally
+ * will be cleared when the node leaves, but since remote node state has a
+ * number of corner cases, clear them here as well, to be sure.
*/
call_opt = crmd_cib_smart_opt();
- controld_delete_node_state(node_name, controld_section_all, call_opt);
+ if (controld_shutdown_lock_enabled) {
+ section = controld_section_all_unlocked;
+ }
+ controld_delete_node_state(node_name, section, call_opt);
/* Clear node's probed attribute */
update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE);
/* Ensure node is in the remote peer cache with member status */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0);
/* pacemaker_remote nodes don't participate in the membership layer,
* so cluster nodes don't automatically get notified when they come and go.
* We send a cluster message to the DC, and update the CIB node state entry,
* so the DC will get it sooner (via message) or later (via CIB refresh),
* and any other interested parties can query the CIB.
*/
send_remote_state_message(node_name, TRUE);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_cluster, update,
__FUNCTION__);
/* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever
* needs to be fenced, this flag will allow various actions to determine
* whether the fencing has happened yet.
*/
crm_xml_add(state, XML_NODE_IS_FENCED, "0");
/* TODO: If the remote connection drops, and this (async) CIB update either
* failed or has not yet completed, later actions could mistakenly think the
* node has already been fenced (if the XML_NODE_IS_FENCED attribute was
* previously set, because it won't have been cleared). This could prevent
* actual fencing or allow recurring monitor failures to be cleared too
* soon. Ideally, we wouldn't rely on the CIB for the fenced status.
*/
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_WARNING, "%s CIB node state setup", node_name);
}
free_xml(update);
}
enum down_opts {
DOWN_KEEP_LRM,
DOWN_ERASE_LRM
};
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node leaving
*
* \param[in] node_name Name of lost node
* \param[in] opts Whether to keep or erase LRM history
*/
static void
remote_node_down(const char *node_name, const enum down_opts opts)
{
xmlNode *update;
int call_id = 0;
int call_opt = crmd_cib_smart_opt();
crm_node_t *node;
/* Purge node from attrd's memory */
update_attrd_remote_node_removed(node_name, NULL);
/* Normally, only node attributes should be erased, and the resource history
* should be kept until the node comes back up. However, after a successful
* fence, we want to clear the history as well, so we don't think resources
* are still running on the node.
*/
if (opts == DOWN_ERASE_LRM) {
controld_delete_node_state(node_name, controld_section_all, call_opt);
} else {
controld_delete_node_state(node_name, controld_section_attrs, call_opt);
}
/* Ensure node is in the remote peer cache with lost state */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0);
/* Notify DC */
send_remote_state_message(node_name, FALSE);
/* Update CIB node state */
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
create_node_state_update(node, node_update_cluster, update, __FUNCTION__);
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_ERR, "%s CIB node state update", node_name);
}
free_xml(update);
}
/*!
* \internal
* \brief Handle effects of a remote RA command on node state
*
* \param[in] cmd Completed remote RA command
*/
static void
check_remote_node_state(remote_ra_cmd_t *cmd)
{
/* Only successful actions can change node state */
if (cmd->rc != PCMK_OCF_OK) {
return;
}
if (safe_str_eq(cmd->action, "start")) {
remote_node_up(cmd->rsc_id);
} else if (safe_str_eq(cmd->action, "migrate_from")) {
/* After a successful migration, we don't need to do remote_node_up()
* because the DC already knows the node is up, and we don't want to
* clear LRM history etc. We do need to add the remote node to this
* host's remote peer cache, because (unless it happens to be DC)
* it hasn't been tracking the remote node, and other code relies on
* the cache to distinguish remote nodes from unseen cluster nodes.
*/
crm_node_t *node = crm_remote_peer_get(cmd->rsc_id);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0);
} else if (safe_str_eq(cmd->action, "stop")) {
lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id);
remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL;
if (ra_data) {
if (ra_data->migrate_status != takeover_complete) {
/* Stop means down if we didn't successfully migrate elsewhere */
remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM);
} else if (AM_I_DC == FALSE) {
/* Only the connection host and DC track node state,
* so if the connection migrated elsewhere and we aren't DC,
* un-cache the node, so we don't have stale info
*/
crm_remote_peer_cache_remove(cmd->rsc_id);
}
}
}
/* We don't do anything for successful monitors, which is correct for
* routine recurring monitors, and for monitors on nodes where the
* connection isn't supposed to be (the cluster will stop the connection in
* that case). However, if the initial probe finds the connection already
* active on the node where we want it, we probably should do
* remote_node_up(). Unfortunately, we can't distinguish that case here.
* Given that connections have to be initiated by the cluster, the chance of
* that should be close to zero.
*/
}
static void
report_remote_ra_result(remote_ra_cmd_t * cmd)
{
lrmd_event_data_t op = { 0, };
check_remote_node_state(cmd);
op.type = lrmd_event_exec_complete;
op.rsc_id = cmd->rsc_id;
op.op_type = cmd->action;
op.user_data = cmd->userdata;
op.exit_reason = cmd->exit_reason;
op.timeout = cmd->timeout;
op.interval_ms = cmd->interval_ms;
op.rc = cmd->rc;
op.op_status = cmd->op_status;
op.t_run = (unsigned int) cmd->start_time;
op.t_rcchange = (unsigned int) cmd->start_time;
if (cmd->reported_success && cmd->rc != PCMK_OCF_OK) {
op.t_rcchange = (unsigned int) time(NULL);
/* This edge case will likely never ever occur, but if it does the
* result is that a failure will not be processed correctly. This is only
* remotely possible because we are able to detect a connection resource's tcp
* connection has failed at any moment after start has completed. The actual
* recurring operation is just a connectivity ping.
*
* basically, we are not guaranteed that the first successful monitor op and
* a subsequent failed monitor op will not occur in the same timestamp. We have to
* make it look like the operations occurred at separate times though. */
if (op.t_rcchange == op.t_run) {
op.t_rcchange++;
}
}
if (cmd->params) {
lrmd_key_value_t *tmp;
op.params = crm_str_table_new();
for (tmp = cmd->params; tmp; tmp = tmp->next) {
g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value));
}
}
op.call_id = cmd->call_id;
op.remote_nodename = cmd->owner;
lrm_op_callback(&op);
if (op.params) {
g_hash_table_destroy(op.params);
}
}
static void
update_remaining_timeout(remote_ra_cmd_t * cmd)
{
cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000;
}
static gboolean
retry_start_cmd_cb(gpointer data)
{
lrm_state_t *lrm_state = data;
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
remote_ra_cmd_t *cmd = NULL;
int rc = -1;
if (!ra_data || !ra_data->cur_cmd) {
return FALSE;
}
cmd = ra_data->cur_cmd;
if (safe_str_neq(cmd->action, "start") && safe_str_neq(cmd->action, "migrate_from")) {
return FALSE;
}
update_remaining_timeout(cmd);
if (cmd->remaining_timeout > 0) {
rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout);
}
if (rc != 0) {
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
cmd->op_status = PCMK_LRM_OP_ERROR;
report_remote_ra_result(cmd);
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
ra_data->cur_cmd = NULL;
free_cmd(cmd);
} else {
/* wait for connection event */
}
return FALSE;
}
static gboolean
connection_takeover_timeout_cb(gpointer data)
{
lrm_state_t *lrm_state = NULL;
remote_ra_cmd_t *cmd = data;
crm_info("takeover event timed out for node %s", cmd->rsc_id);
cmd->takeover_timeout_id = 0;
lrm_state = lrm_state_find(cmd->rsc_id);
handle_remote_ra_stop(lrm_state, cmd);
free_cmd(cmd);
return FALSE;
}
static gboolean
monitor_timeout_cb(gpointer data)
{
lrm_state_t *lrm_state = NULL;
remote_ra_cmd_t *cmd = data;
lrm_state = lrm_state_find(cmd->rsc_id);
crm_info("Timed out waiting for remote poke response from %s%s",
cmd->rsc_id, (lrm_state? "" : " (no LRM state)"));
cmd->monitor_timeout_id = 0;
cmd->op_status = PCMK_LRM_OP_TIMEOUT;
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
if (lrm_state && lrm_state->remote_ra_data) {
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
if (ra_data->cur_cmd == cmd) {
ra_data->cur_cmd = NULL;
}
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
}
report_remote_ra_result(cmd);
free_cmd(cmd);
if(lrm_state) {
lrm_state_disconnect(lrm_state);
}
return FALSE;
}
static void
synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type)
{
lrmd_event_data_t op = { 0, };
if (lrm_state == NULL) {
/* if lrm_state not given assume local */
lrm_state = lrm_state_find(fsa_our_uname);
}
CRM_ASSERT(lrm_state != NULL);
op.type = lrmd_event_exec_complete;
op.rsc_id = rsc_id;
op.op_type = op_type;
op.rc = PCMK_OCF_OK;
op.op_status = PCMK_LRM_OP_DONE;
op.t_run = (unsigned int) time(NULL);
op.t_rcchange = op.t_run;
op.call_id = generate_callid();
process_lrm_event(lrm_state, &op, NULL, NULL);
}
void
remote_lrm_op_callback(lrmd_event_data_t * op)
{
gboolean cmd_handled = FALSE;
lrm_state_t *lrm_state = NULL;
remote_ra_data_t *ra_data = NULL;
remote_ra_cmd_t *cmd = NULL;
crm_debug("Processing '%s%s%s' event on remote connection to %s: %s "
"(%d) status=%s (%d)",
(op->op_type? op->op_type : ""), (op->op_type? " " : ""),
lrmd_event_type2str(op->type), op->remote_nodename,
services_ocf_exitcode_str(op->rc), op->rc,
services_lrm_status_str(op->op_status), op->op_status);
lrm_state = lrm_state_find(op->remote_nodename);
if (!lrm_state || !lrm_state->remote_ra_data) {
crm_debug("No state information found for remote connection event");
return;
}
ra_data = lrm_state->remote_ra_data;
if (op->type == lrmd_event_new_client) {
// Another client has connected to the remote daemon
if (ra_data->migrate_status == expect_takeover) {
// Great, we knew this was coming
ra_data->migrate_status = takeover_complete;
} else {
crm_err("Unexpected pacemaker_remote client takeover for %s. Disconnecting", op->remote_nodename);
/* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */
/* Do not free lrm_state->conn yet. */
/* It'll be freed in the following stop action. */
lrm_state_disconnect_only(lrm_state);
}
return;
}
/* filter all EXEC events up */
if (op->type == lrmd_event_exec_complete) {
if (ra_data->migrate_status == takeover_complete) {
crm_debug("ignoring event, this connection is taken over by another node");
} else {
lrm_op_callback(op);
}
return;
}
if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) {
if (ra_data->active == FALSE) {
crm_debug("Disconnection from Pacemaker Remote node %s complete",
lrm_state->node_name);
} else if (!remote_ra_is_in_maintenance(lrm_state)) {
crm_err("Lost connection to Pacemaker Remote node %s",
lrm_state->node_name);
ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds);
ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds);
} else {
crm_notice("Unmanaged Pacemaker Remote node %s disconnected",
lrm_state->node_name);
/* Do roughly what a 'stop' on the remote-resource would do */
handle_remote_ra_stop(lrm_state, NULL);
remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM);
/* now fake the reply of a successful 'stop' */
synthesize_lrmd_success(NULL, lrm_state->node_name, "stop");
}
return;
}
if (!ra_data->cur_cmd) {
crm_debug("no event to match");
return;
}
cmd = ra_data->cur_cmd;
/* Start actions and migrate from actions complete after connection
* comes back to us. */
if (op->type == lrmd_event_connect && (safe_str_eq(cmd->action, "start") ||
safe_str_eq(cmd->action, "migrate_from"))) {
if (op->connection_rc < 0) {
update_remaining_timeout(cmd);
if (op->connection_rc == -ENOKEY) {
// Hard error, don't retry
cmd->op_status = PCMK_LRM_OP_ERROR;
cmd->rc = PCMK_OCF_INVALID_PARAM;
cmd->exit_reason = strdup("Authentication key not readable");
} else if (cmd->remaining_timeout > 3000) {
crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout);
g_timeout_add(1000, retry_start_cmd_cb, lrm_state);
return;
} else {
crm_trace("can't reschedule start, remaining timeout too small %d",
cmd->remaining_timeout);
cmd->op_status = PCMK_LRM_OP_TIMEOUT;
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
}
} else {
lrm_state_reset_tables(lrm_state, TRUE);
cmd->rc = PCMK_OCF_OK;
cmd->op_status = PCMK_LRM_OP_DONE;
ra_data->active = TRUE;
}
crm_debug("Remote connection event matched %s action", cmd->action);
report_remote_ra_result(cmd);
cmd_handled = TRUE;
} else if (op->type == lrmd_event_poke && safe_str_eq(cmd->action, "monitor")) {
if (cmd->monitor_timeout_id) {
g_source_remove(cmd->monitor_timeout_id);
cmd->monitor_timeout_id = 0;
}
/* Only report success the first time, after that only worry about failures.
* For this function, if we get the poke pack, it is always a success. Pokes
* only fail if the send fails, or the response times out. */
if (!cmd->reported_success) {
cmd->rc = PCMK_OCF_OK;
cmd->op_status = PCMK_LRM_OP_DONE;
report_remote_ra_result(cmd);
cmd->reported_success = 1;
}
crm_debug("Remote poke event matched %s action", cmd->action);
/* success, keep rescheduling if interval is present. */
if (cmd->interval_ms && (cmd->cancel == FALSE)) {
ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd);
cmd->interval_id = g_timeout_add(cmd->interval_ms,
recurring_helper, cmd);
cmd = NULL; /* prevent free */
}
cmd_handled = TRUE;
} else if (op->type == lrmd_event_disconnect && safe_str_eq(cmd->action, "monitor")) {
if (ra_data->active == TRUE && (cmd->cancel == FALSE)) {
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
cmd->op_status = PCMK_LRM_OP_ERROR;
report_remote_ra_result(cmd);
crm_err("Remote connection to %s unexpectedly dropped during monitor",
lrm_state->node_name);
}
cmd_handled = TRUE;
} else if (op->type == lrmd_event_new_client && safe_str_eq(cmd->action, "stop")) {
handle_remote_ra_stop(lrm_state, cmd);
cmd_handled = TRUE;
} else {
crm_debug("Event did not match %s action", ra_data->cur_cmd->action);
}
if (cmd_handled) {
ra_data->cur_cmd = NULL;
if (ra_data->cmds) {
mainloop_set_trigger(ra_data->work);
}
free_cmd(cmd);
}
}
static void
handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd)
{
remote_ra_data_t *ra_data = NULL;
CRM_ASSERT(lrm_state);
ra_data = lrm_state->remote_ra_data;
if (ra_data->migrate_status != takeover_complete) {
/* delete pending ops when ever the remote connection is intentionally stopped */
g_hash_table_remove_all(lrm_state->pending_ops);
} else {
/* we no longer hold the history if this connection has been migrated,
* however, we keep metadata cache for future use */
lrm_state_reset_tables(lrm_state, FALSE);
}
ra_data->active = FALSE;
lrm_state_disconnect(lrm_state);
if (ra_data->cmds) {
g_list_free_full(ra_data->cmds, free_cmd);
}
if (ra_data->recurring_cmds) {
g_list_free_full(ra_data->recurring_cmds, free_cmd);
}
ra_data->cmds = NULL;
ra_data->recurring_cmds = NULL;
ra_data->cur_cmd = NULL;
if (cmd) {
cmd->rc = PCMK_OCF_OK;
cmd->op_status = PCMK_LRM_OP_DONE;
report_remote_ra_result(cmd);
}
}
static int
handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms)
{
const char *server = NULL;
lrmd_key_value_t *tmp = NULL;
int port = 0;
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms;
for (tmp = cmd->params; tmp; tmp = tmp->next) {
if (safe_str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR) ||
safe_str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_SERVER)) {
server = tmp->value;
} else if (safe_str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT)) {
port = atoi(tmp->value);
} else if (safe_str_eq(tmp->key, CRM_META"_"XML_RSC_ATTR_CONTAINER)) {
ra_data->controlling_guest = TRUE;
}
}
return lrm_state_remote_connect_async(lrm_state, server, port, timeout_used);
}
static gboolean
handle_remote_ra_exec(gpointer user_data)
{
int rc = 0;
lrm_state_t *lrm_state = user_data;
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
remote_ra_cmd_t *cmd;
GList *first = NULL;
if (ra_data->cur_cmd) {
/* still waiting on previous cmd */
return TRUE;
}
while (ra_data->cmds) {
first = ra_data->cmds;
cmd = first->data;
if (cmd->delay_id) {
/* still waiting for start delay timer to trip */
return TRUE;
}
ra_data->cmds = g_list_remove_link(ra_data->cmds, first);
g_list_free_1(first);
if (!strcmp(cmd->action, "start") || !strcmp(cmd->action, "migrate_from")) {
ra_data->migrate_status = 0;
rc = handle_remote_ra_start(lrm_state, cmd, cmd->timeout);
if (rc == 0) {
/* take care of this later when we get async connection result */
crm_debug("Initiated async remote connection, %s action will complete after connect event",
cmd->action);
ra_data->cur_cmd = cmd;
return TRUE;
} else {
crm_debug("Could not initiate remote connection for %s action",
cmd->action);
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
cmd->op_status = PCMK_LRM_OP_ERROR;
}
report_remote_ra_result(cmd);
} else if (!strcmp(cmd->action, "monitor")) {
if (lrm_state_is_connected(lrm_state) == TRUE) {
rc = lrm_state_poke_connection(lrm_state);
if (rc < 0) {
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
cmd->op_status = PCMK_LRM_OP_ERROR;
}
} else {
rc = -1;
cmd->op_status = PCMK_LRM_OP_DONE;
cmd->rc = PCMK_OCF_NOT_RUNNING;
}
if (rc == 0) {
crm_debug("Poked Pacemaker Remote at node %s, waiting for async response",
cmd->rsc_id);
ra_data->cur_cmd = cmd;
cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd);
return TRUE;
}
report_remote_ra_result(cmd);
} else if (!strcmp(cmd->action, "stop")) {
if (ra_data->migrate_status == expect_takeover) {
/* briefly wait on stop for the takeover event to occur. If the
* takeover event does not occur during the wait period, that's fine.
* It just means that the remote-node's lrm_status section is going to get
* cleared which will require all the resources running in the remote-node
* to be explicitly re-detected via probe actions. If the takeover does occur
* successfully, then we can leave the status section intact. */
cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd);
ra_data->cur_cmd = cmd;
return TRUE;
}
handle_remote_ra_stop(lrm_state, cmd);
} else if (!strcmp(cmd->action, "migrate_to")) {
ra_data->migrate_status = expect_takeover;
cmd->rc = PCMK_OCF_OK;
cmd->op_status = PCMK_LRM_OP_DONE;
report_remote_ra_result(cmd);
} else if (!strcmp(cmd->action, "reload")) {
/* reloads are a no-op right now, add logic here when they become important */
cmd->rc = PCMK_OCF_OK;
cmd->op_status = PCMK_LRM_OP_DONE;
report_remote_ra_result(cmd);
}
free_cmd(cmd);
}
return TRUE;
}
static void
remote_ra_data_init(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = NULL;
if (lrm_state->remote_ra_data) {
return;
}
ra_data = calloc(1, sizeof(remote_ra_data_t));
ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state);
lrm_state->remote_ra_data = ra_data;
}
void
remote_ra_cleanup(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
if (!ra_data) {
return;
}
if (ra_data->cmds) {
g_list_free_full(ra_data->cmds, free_cmd);
}
if (ra_data->recurring_cmds) {
g_list_free_full(ra_data->recurring_cmds, free_cmd);
}
mainloop_destroy_trigger(ra_data->work);
free(ra_data);
lrm_state->remote_ra_data = NULL;
}
gboolean
is_remote_lrmd_ra(const char *agent, const char *provider, const char *id)
{
if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) {
return TRUE;
}
if (id && lrm_state_find(id) && safe_str_neq(id, fsa_our_uname)) {
return TRUE;
}
return FALSE;
}
lrmd_rsc_info_t *
remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id)
{
lrmd_rsc_info_t *info = NULL;
if ((lrm_state_find(rsc_id))) {
info = calloc(1, sizeof(lrmd_rsc_info_t));
info->id = strdup(rsc_id);
info->type = strdup(REMOTE_LRMD_RA);
info->standard = strdup(PCMK_RESOURCE_CLASS_OCF);
info->provider = strdup("pacemaker");
}
return info;
}
static gboolean
is_remote_ra_supported_action(const char *action)
{
if (!action) {
return FALSE;
} else if (strcmp(action, "start") &&
strcmp(action, "stop") &&
strcmp(action, "reload") &&
strcmp(action, "migrate_to") &&
strcmp(action, "migrate_from") && strcmp(action, "monitor")) {
return FALSE;
}
return TRUE;
}
static GList *
fail_all_monitor_cmds(GList * list)
{
GList *rm_list = NULL;
remote_ra_cmd_t *cmd = NULL;
GListPtr gIter = NULL;
for (gIter = list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms > 0) && safe_str_eq(cmd->action, "monitor")) {
rm_list = g_list_append(rm_list, cmd);
}
}
for (gIter = rm_list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
cmd->rc = PCMK_OCF_UNKNOWN_ERROR;
cmd->op_status = PCMK_LRM_OP_ERROR;
crm_trace("Pre-emptively failing %s %s (interval=%u, %s)",
cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata);
report_remote_ra_result(cmd);
list = g_list_remove(list, cmd);
free_cmd(cmd);
}
/* frees only the list data, not the cmds */
g_list_free(rm_list);
return list;
}
static GList *
remove_cmd(GList * list, const char *action, guint interval_ms)
{
remote_ra_cmd_t *cmd = NULL;
GListPtr gIter = NULL;
for (gIter = list; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& safe_str_eq(cmd->action, action)) {
break;
}
cmd = NULL;
}
if (cmd) {
list = g_list_remove(list, cmd);
free_cmd(cmd);
}
return list;
}
int
remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id,
const char *action, guint interval_ms)
{
lrm_state_t *connection_rsc = NULL;
remote_ra_data_t *ra_data = NULL;
connection_rsc = lrm_state_find(rsc_id);
if (!connection_rsc || !connection_rsc->remote_ra_data) {
return -EINVAL;
}
ra_data = connection_rsc->remote_ra_data;
ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms);
ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action,
interval_ms);
if (ra_data->cur_cmd &&
(ra_data->cur_cmd->interval_ms == interval_ms) &&
(safe_str_eq(ra_data->cur_cmd->action, action))) {
ra_data->cur_cmd->cancel = TRUE;
}
return 0;
}
static remote_ra_cmd_t *
handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms,
const char *userdata)
{
GList *gIter = NULL;
remote_ra_cmd_t *cmd = NULL;
/* there are 3 places a potential duplicate monitor operation
* could exist.
* 1. recurring_cmds list. where the op is waiting for its next interval
* 2. cmds list, where the op is queued to get executed immediately
* 3. cur_cmd, which means the monitor op is in flight right now.
*/
if (interval_ms == 0) {
return NULL;
}
if (ra_data->cur_cmd &&
ra_data->cur_cmd->cancel == FALSE &&
(ra_data->cur_cmd->interval_ms == interval_ms) &&
safe_str_eq(ra_data->cur_cmd->action, "monitor")) {
cmd = ra_data->cur_cmd;
goto handle_dup;
}
for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& safe_str_eq(cmd->action, "monitor")) {
goto handle_dup;
}
}
for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) {
cmd = gIter->data;
if ((cmd->interval_ms == interval_ms)
&& safe_str_eq(cmd->action, "monitor")) {
goto handle_dup;
}
}
return NULL;
handle_dup:
crm_trace("merging duplicate monitor cmd " CRM_OP_FMT,
cmd->rsc_id, "monitor", interval_ms);
/* update the userdata */
if (userdata) {
free(cmd->userdata);
cmd->userdata = strdup(userdata);
}
/* if we've already reported success, generate a new call id */
if (cmd->reported_success) {
cmd->start_time = time(NULL);
cmd->call_id = generate_callid();
cmd->reported_success = 0;
}
/* if we have an interval_id set, that means we are in the process of
* waiting for this cmd's next interval. instead of waiting, cancel
* the timer and execute the action immediately */
if (cmd->interval_id) {
g_source_remove(cmd->interval_id);
cmd->interval_id = 0;
recurring_helper(cmd);
}
return cmd;
}
int
remote_ra_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action,
const char *userdata, guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
lrmd_key_value_t * params)
{
int rc = 0;
lrm_state_t *connection_rsc = NULL;
remote_ra_cmd_t *cmd = NULL;
remote_ra_data_t *ra_data = NULL;
if (is_remote_ra_supported_action(action) == FALSE) {
rc = -EINVAL;
goto exec_done;
}
connection_rsc = lrm_state_find(rsc_id);
if (!connection_rsc) {
rc = -EINVAL;
goto exec_done;
}
remote_ra_data_init(connection_rsc);
ra_data = connection_rsc->remote_ra_data;
cmd = handle_dup_monitor(ra_data, interval_ms, userdata);
if (cmd) {
rc = cmd->call_id;
goto exec_done;
}
cmd = calloc(1, sizeof(remote_ra_cmd_t));
cmd->owner = strdup(lrm_state->node_name);
cmd->rsc_id = strdup(rsc_id);
cmd->action = strdup(action);
cmd->userdata = strdup(userdata);
cmd->interval_ms = interval_ms;
cmd->timeout = timeout;
cmd->start_delay = start_delay;
cmd->params = params;
cmd->start_time = time(NULL);
cmd->call_id = generate_callid();
if (cmd->start_delay) {
cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd);
}
ra_data->cmds = g_list_append(ra_data->cmds, cmd);
mainloop_set_trigger(ra_data->work);
return cmd->call_id;
exec_done:
lrmd_key_value_freeall(params);
return rc;
}
/*!
* \internal
* \brief Immediately fail all monitors of a remote node, if proxied here
*
* \param[in] node_name Name of pacemaker_remote node
*/
void
remote_ra_fail(const char *node_name)
{
lrm_state_t *lrm_state = lrm_state_find(node_name);
if (lrm_state && lrm_state_is_connected(lrm_state)) {
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
crm_info("Failing monitors on pacemaker_remote node %s", node_name);
ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds);
ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds);
}
}
/* A guest node fencing implied by host fencing looks like:
*
*
*
*
*
*
*
*/
#define XPATH_PSEUDO_FENCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \
"[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \
"/" XML_CIB_TAG_NODE
/*!
* \internal
* \brief Check a pseudo-action for Pacemaker Remote node side effects
*
* \param[in] xml XML of pseudo-action to check
*/
void
remote_ra_process_pseudo(xmlNode *xml)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE);
if (numXpathResults(search) == 1) {
xmlNode *result = getXpathResult(search, 0);
/* Normally, we handle the necessary side effects of a guest node stop
* action when reporting the remote agent's result. However, if the stop
* is implied due to fencing, it will be a fencing pseudo-event, and
* there won't be a result to report. Handle that case here.
*
* This will result in a duplicate call to remote_node_down() if the
* guest stop was real instead of implied, but that shouldn't hurt.
*
* There is still one corner case that isn't handled: if a guest node
* isn't running any resources when its host is fenced, it will appear
* to be cleanly stopped, so there will be no pseudo-fence, and our
* peer cache state will be incorrect unless and until the guest is
* recovered.
*/
if (result) {
const char *remote = ID(result);
if (remote) {
remote_node_down(remote, DOWN_ERASE_LRM);
}
}
}
freeXpathObject(search);
}
static void
remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
xmlNode *update, *state;
int call_opt, call_id = 0;
crm_node_t *node;
call_opt = crmd_cib_smart_opt();
node = crm_remote_peer_get(lrm_state->node_name);
CRM_CHECK(node != NULL, return);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_none, update,
__FUNCTION__);
crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0");
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_WARNING, "%s CIB node state update failed", lrm_state->node_name);
} else {
/* TODO: still not 100% sure that async update will succeed ... */
ra_data->is_maintenance = maintenance;
}
free_xml(update);
}
#define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \
"[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \
XML_GRAPH_TAG_MAINTENANCE
/*!
* \internal
* \brief Check a pseudo-action holding updates for maintenance state
*
* \param[in] xml XML of pseudo-action to check
*/
void
remote_ra_process_maintenance_nodes(xmlNode *xml)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE);
if (numXpathResults(search) == 1) {
xmlNode *node;
int cnt = 0, cnt_remote = 0;
for (node =
first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE);
node; node = __xml_next(node)) {
lrm_state_t *lrm_state = lrm_state_find(ID(node));
cnt++;
if (lrm_state && lrm_state->remote_ra_data &&
((remote_ra_data_t *) lrm_state->remote_ra_data)->active) {
cnt_remote++;
remote_ra_maintenance(lrm_state,
crm_atoi(crm_element_value(node,
XML_NODE_IS_MAINTENANCE), "0"));
}
}
crm_trace("Action holds %d nodes (%d remotes found) "
"adjusting maintenance-mode", cnt, cnt_remote);
}
freeXpathObject(search);
}
gboolean
remote_ra_is_in_maintenance(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
return ra_data->is_maintenance;
}
gboolean
remote_ra_controlling_guest(lrm_state_t * lrm_state)
{
remote_ra_data_t *ra_data = lrm_state->remote_ra_data;
return ra_data->controlling_guest;
}
diff --git a/daemons/controld/controld_te_actions.c b/daemons/controld/controld_te_actions.c
index 948bd64245..59e0b5a697 100644
--- a/daemons/controld/controld_te_actions.c
+++ b/daemons/controld/controld_te_actions.c
@@ -1,635 +1,642 @@
/*
* Copyright 2004-2019 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include // lrmd_event_data_t, lrmd_free_event()
#include
#include
#include
#include
#include
char *te_uuid = NULL;
GHashTable *te_targets = NULL;
void send_rsc_command(crm_action_t * action);
static void te_update_job_count(crm_action_t * action, int offset);
static void
te_start_action_timer(crm_graph_t * graph, crm_action_t * action)
{
action->timer = calloc(1, sizeof(crm_action_timer_t));
action->timer->timeout = action->timeout;
action->timer->action = action;
action->timer->source_id = g_timeout_add(action->timer->timeout + graph->network_delay,
action_timer_callback, (void *)action->timer);
CRM_ASSERT(action->timer->source_id != 0);
}
static gboolean
te_pseudo_action(crm_graph_t * graph, crm_action_t * pseudo)
{
const char *task = crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK);
/* send to peers as well? */
if (safe_str_eq(task, CRM_OP_MAINTENANCE_NODES)) {
GHashTableIter iter;
crm_node_t *node = NULL;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
xmlNode *cmd = NULL;
if (safe_str_eq(fsa_our_uname, node->uname)) {
continue;
}
cmd = create_request(task, pseudo->xml, node->uname,
CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL);
send_cluster_message(node, crm_msg_crmd, cmd, FALSE);
free_xml(cmd);
}
remote_ra_process_maintenance_nodes(pseudo->xml);
} else {
/* Check action for Pacemaker Remote node side effects */
remote_ra_process_pseudo(pseudo->xml);
}
crm_debug("Pseudo-action %d (%s) fired and confirmed", pseudo->id,
crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK_KEY));
te_action_confirmed(pseudo, graph);
return TRUE;
}
static int
get_target_rc(crm_action_t * action)
{
const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC);
if (target_rc_s != NULL) {
return crm_parse_int(target_rc_s, "0");
}
return 0;
}
static gboolean
te_crm_command(crm_graph_t * graph, crm_action_t * action)
{
char *counter = NULL;
xmlNode *cmd = NULL;
gboolean is_local = FALSE;
const char *id = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
const char *router_node = NULL;
gboolean rc = TRUE;
gboolean no_wait = FALSE;
id = ID(action->xml);
task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE);
if (!router_node) {
router_node = on_node;
+ if (safe_str_eq(task, CRM_OP_LRM_DELETE)) {
+ const char *mode = crm_element_value(action->xml, PCMK__XA_MODE);
+
+ if (safe_str_eq(mode, XML_TAG_CIB)) {
+ router_node = fsa_our_uname;
+ }
+ }
}
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
crm_err("Corrupted command (id=%s) %s: no node", crm_str(id), crm_str(task));
return FALSE);
if (safe_str_eq(router_node, fsa_our_uname)) {
is_local = TRUE;
}
value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT);
if (crm_is_true(value)) {
no_wait = TRUE;
}
crm_info("Executing crm-event (%s)%s%s: %s on %s",
crm_str(id), (is_local? " locally" : ""),
(no_wait? " without waiting" : ""), crm_str(task), on_node);
if (is_local && safe_str_eq(task, CRM_OP_SHUTDOWN)) {
/* defer until everything else completes */
crm_info("crm-event (%s) is a local shutdown", crm_str(id));
graph->completion_action = tg_shutdown;
graph->abort_reason = "local shutdown";
te_action_confirmed(action, graph);
return TRUE;
} else if (safe_str_eq(task, CRM_OP_SHUTDOWN)) {
crm_node_t *peer = crm_get_peer(0, router_node);
crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN);
}
cmd = create_request(task, action->xml, router_node, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL);
counter =
generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid);
crm_xml_add(cmd, XML_ATTR_TRANSITION_KEY, counter);
rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_crmd, cmd, TRUE);
free(counter);
free_xml(cmd);
if (rc == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if (no_wait) {
te_action_confirmed(action, graph);
} else {
if (action->timeout <= 0) {
crm_err("Action %d: %s on %s had an invalid timeout (%dms). Using %ums instead",
action->id, task, on_node, action->timeout, graph->network_delay);
action->timeout = (int) graph->network_delay;
}
te_start_action_timer(graph, action);
}
return TRUE;
}
void
controld_record_action_timeout(crm_action_t *action)
{
lrmd_event_data_t *op = NULL;
xmlNode *state = NULL;
xmlNode *rsc = NULL;
xmlNode *xml_op = NULL;
xmlNode *action_rsc = NULL;
int rc = pcmk_ok;
const char *rsc_id = NULL;
const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
const char *target_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
int call_options = cib_quorum_override | cib_scope_local;
int target_rc = get_target_rc(action);
crm_warn("%s %d: %s on %s timed out",
crm_element_name(action->xml), action->id, task_uuid, target);
action_rsc = find_xml_node(action->xml, XML_CIB_TAG_RESOURCE, TRUE);
if (action_rsc == NULL) {
return;
}
rsc_id = ID(action_rsc);
CRM_CHECK(rsc_id != NULL,
crm_log_xml_err(action->xml, "Bad:action"); return);
/*
update the CIB
*/
state = create_xml_node(NULL, XML_CIB_TAG_STATE);
crm_xml_add(state, XML_ATTR_UUID, target_uuid);
crm_xml_add(state, XML_ATTR_UNAME, target);
rsc = create_xml_node(state, XML_CIB_TAG_LRM);
crm_xml_add(rsc, XML_ATTR_ID, target_uuid);
rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCES);
rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCE);
crm_xml_add(rsc, XML_ATTR_ID, rsc_id);
crm_copy_xml_element(action_rsc, rsc, XML_ATTR_TYPE);
crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_CLASS);
crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_PROVIDER);
/* If the executor gets a timeout while waiting for the action to complete,
* that will be reported via the usual callback. This timeout means that we
* didn't hear from the executor or the controller that relayed the action
* to the executor.
*
* @TODO Using PCMK_OCF_UNKNOWN_ERROR instead of PCMK_OCF_TIMEOUT is one way
* to distinguish those situations, but perhaps PCMK_OCF_TIMEOUT would be
* preferable anyway.
*/
op = convert_graph_action(NULL, action, PCMK_LRM_OP_TIMEOUT,
PCMK_OCF_UNKNOWN_ERROR);
op->call_id = -1;
op->user_data = generate_transition_key(transition_graph->id, action->id, target_rc, te_uuid);
xml_op = pcmk__create_history_xml(rsc, op, CRM_FEATURE_SET, target_rc,
target, __FUNCTION__, LOG_INFO);
lrmd_free_event(op);
crm_log_xml_trace(xml_op, "Action timeout");
rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, state, call_options);
fsa_register_cib_callback(rc, FALSE, NULL, cib_action_updated);
free_xml(state);
crm_trace("Sent CIB update (call ID %d) for timeout of action %d (%s on %s)",
rc, action->id, task_uuid, target);
action->sent_update = TRUE;
}
static gboolean
te_rsc_command(crm_graph_t * graph, crm_action_t * action)
{
/* never overwrite stop actions in the CIB with
* anything other than completed results
*
* Writing pending stops makes it look like the
* resource is running again
*/
xmlNode *cmd = NULL;
xmlNode *rsc_op = NULL;
gboolean rc = TRUE;
gboolean no_wait = FALSE;
gboolean is_local = FALSE;
char *counter = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
const char *router_node = NULL;
const char *task_uuid = NULL;
CRM_ASSERT(action != NULL);
CRM_ASSERT(action->xml != NULL);
action->executed = FALSE;
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
crm_err("Corrupted command(id=%s) %s: no node", ID(action->xml), crm_str(task));
return FALSE);
rsc_op = action->xml;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
router_node = crm_element_value(rsc_op, XML_LRM_ATTR_ROUTER_NODE);
if (!router_node) {
router_node = on_node;
}
counter =
generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid);
crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter);
if (safe_str_eq(router_node, fsa_our_uname)) {
is_local = TRUE;
}
value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT);
if (crm_is_true(value)) {
no_wait = TRUE;
}
crm_notice("Initiating %s operation %s%s on %s%s "CRM_XS" action %d",
task, task_uuid, (is_local? " locally" : ""), on_node,
(no_wait? " without waiting" : ""), action->id);
cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, router_node,
CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL);
if (is_local) {
/* shortcut local resource commands */
ha_msg_input_t data = {
.msg = cmd,
.xml = rsc_op,
};
fsa_data_t msg = {
.id = 0,
.data = &data,
.data_type = fsa_dt_ha_msg,
.fsa_input = I_NULL,
.fsa_cause = C_FSA_INTERNAL,
.actions = A_LRM_INVOKE,
.origin = __FUNCTION__,
};
do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg);
} else {
rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_lrmd, cmd, TRUE);
}
free(counter);
free_xml(cmd);
action->executed = TRUE;
if (rc == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if (no_wait) {
crm_info("Action %d confirmed - no wait", action->id);
action->confirmed = TRUE; /* Just mark confirmed.
* Don't bump the job count only to immediately decrement it
*/
update_graph(transition_graph, action);
trigger_graph();
} else if (action->confirmed == TRUE) {
crm_debug("Action %d: %s %s on %s(timeout %dms) was already confirmed.",
action->id, task, task_uuid, on_node, action->timeout);
} else {
if (action->timeout <= 0) {
crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %ums instead",
action->id, task, task_uuid, on_node, action->timeout, graph->network_delay);
action->timeout = (int) graph->network_delay;
}
te_update_job_count(action, 1);
te_start_action_timer(graph, action);
}
return TRUE;
}
struct te_peer_s
{
char *name;
int jobs;
int migrate_jobs;
};
static void te_peer_free(gpointer p)
{
struct te_peer_s *peer = p;
free(peer->name);
free(peer);
}
void te_reset_job_counts(void)
{
GHashTableIter iter;
struct te_peer_s *peer = NULL;
if(te_targets == NULL) {
te_targets = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, te_peer_free);
}
g_hash_table_iter_init(&iter, te_targets);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & peer)) {
peer->jobs = 0;
peer->migrate_jobs = 0;
}
}
static void
te_update_job_count_on(const char *target, int offset, bool migrate)
{
struct te_peer_s *r = NULL;
if(target == NULL || te_targets == NULL) {
return;
}
r = g_hash_table_lookup(te_targets, target);
if(r == NULL) {
r = calloc(1, sizeof(struct te_peer_s));
r->name = strdup(target);
g_hash_table_insert(te_targets, r->name, r);
}
r->jobs += offset;
if(migrate) {
r->migrate_jobs += offset;
}
crm_trace("jobs[%s] = %d", target, r->jobs);
}
static void
te_update_job_count(crm_action_t * action, int offset)
{
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
if (action->type != action_type_rsc || target == NULL) {
/* No limit on these */
return;
}
/* if we have a router node, this means the action is performing
* on a remote node. For now, we count all actions occurring on a
* remote node against the job list on the cluster node hosting
* the connection resources */
target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE);
if ((target == NULL) &&
(safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED))) {
const char *t1 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE);
const char *t2 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET);
te_update_job_count_on(t1, offset, TRUE);
te_update_job_count_on(t2, offset, TRUE);
return;
} else if (target == NULL) {
target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
}
te_update_job_count_on(target, offset, FALSE);
}
static gboolean
te_should_perform_action_on(crm_graph_t * graph, crm_action_t * action, const char *target)
{
int limit = 0;
struct te_peer_s *r = NULL;
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
const char *id = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
if(target == NULL) {
/* No limit on these */
return TRUE;
} else if(te_targets == NULL) {
return FALSE;
}
r = g_hash_table_lookup(te_targets, target);
limit = throttle_get_job_limit(target);
if(r == NULL) {
r = calloc(1, sizeof(struct te_peer_s));
r->name = strdup(target);
g_hash_table_insert(te_targets, r->name, r);
}
if(limit <= r->jobs) {
crm_trace("Peer %s is over their job limit of %d (%d): deferring %s",
target, limit, r->jobs, id);
return FALSE;
} else if(graph->migration_limit > 0 && r->migrate_jobs >= graph->migration_limit) {
if (safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED)) {
crm_trace("Peer %s is over their migration job limit of %d (%d): deferring %s",
target, graph->migration_limit, r->migrate_jobs, id);
return FALSE;
}
}
crm_trace("Peer %s has not hit their limit yet. current jobs = %d limit= %d limit", target, r->jobs, limit);
return TRUE;
}
static gboolean
te_should_perform_action(crm_graph_t * graph, crm_action_t * action)
{
const char *target = NULL;
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
if (action->type != action_type_rsc) {
/* No limit on these */
return TRUE;
}
/* if we have a router node, this means the action is performing
* on a remote node. For now, we count all actions occurring on a
* remote node against the job list on the cluster node hosting
* the connection resources */
target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE);
if ((target == NULL) &&
(safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED))) {
target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE);
if(te_should_perform_action_on(graph, action, target) == FALSE) {
return FALSE;
}
target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET);
} else if (target == NULL) {
target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
}
return te_should_perform_action_on(graph, action, target);
}
/*!
* \brief Confirm a graph action (and optionally update graph)
*
* \param[in] action Action to confirm
* \param[in] graph Update and trigger this graph (if non-NULL)
*/
void
te_action_confirmed(crm_action_t *action, crm_graph_t *graph)
{
if (action->confirmed == FALSE) {
if ((action->type == action_type_rsc)
&& (crm_element_value(action->xml, XML_LRM_ATTR_TARGET) != NULL)) {
te_update_job_count(action, -1);
}
action->confirmed = TRUE;
}
if (graph) {
update_graph(graph, action);
trigger_graph();
}
}
crm_graph_functions_t te_graph_fns = {
te_pseudo_action,
te_rsc_command,
te_crm_command,
te_fence_node,
te_should_perform_action,
};
void
notify_crmd(crm_graph_t * graph)
{
const char *type = "unknown";
enum crmd_fsa_input event = I_NULL;
crm_debug("Processing transition completion in state %s", fsa_state2string(fsa_state));
if (graph->complete == FALSE) {
CRM_CHECK(graph->complete,);
graph->complete = TRUE;
}
switch (graph->completion_action) {
case tg_stop:
type = "stop";
if (fsa_state == S_TRANSITION_ENGINE) {
event = I_TE_SUCCESS;
}
break;
case tg_done:
type = "done";
if (fsa_state == S_TRANSITION_ENGINE) {
event = I_TE_SUCCESS;
}
break;
case tg_restart:
type = "restart";
if (fsa_state == S_TRANSITION_ENGINE) {
if (transition_timer->period_ms > 0) {
controld_stop_timer(transition_timer);
controld_start_timer(transition_timer);
} else {
event = I_PE_CALC;
}
} else if (fsa_state == S_POLICY_ENGINE) {
register_fsa_action(A_PE_INVOKE);
}
break;
case tg_shutdown:
type = "shutdown";
if (is_set(fsa_input_register, R_SHUTDOWN)) {
event = I_STOP;
} else {
crm_err("We didn't ask to be shut down, yet the scheduler is telling us to");
event = I_TERMINATE;
}
}
crm_debug("Transition %d status: %s - %s", graph->id, type, crm_str(graph->abort_reason));
graph->abort_reason = NULL;
graph->completion_action = tg_done;
clear_bit(fsa_input_register, R_IN_TRANSITION);
if (event != I_NULL) {
register_fsa_input(C_FSA_INTERNAL, event, NULL);
} else if (fsa_source) {
mainloop_set_trigger(fsa_source);
}
}
diff --git a/daemons/controld/controld_te_callbacks.c b/daemons/controld/controld_te_callbacks.c
index 25f0ab2aaa..8506f26232 100644
--- a/daemons/controld/controld_te_callbacks.c
+++ b/daemons/controld/controld_te_callbacks.c
@@ -1,690 +1,715 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include /* For ONLINESTATUS etc */
#include
void te_update_confirm(const char *event, xmlNode * msg);
extern char *te_uuid;
gboolean shuttingdown = FALSE;
crm_graph_t *transition_graph;
crm_trigger_t *transition_trigger = NULL;
/* #define RSC_OP_TEMPLATE "//"XML_TAG_DIFF_ADDED"//"XML_TAG_CIB"//"XML_CIB_TAG_STATE"[@uname='%s']"//"XML_LRM_TAG_RSC_OP"[@id='%s]" */
#define RSC_OP_TEMPLATE "//"XML_TAG_DIFF_ADDED"//"XML_TAG_CIB"//"XML_LRM_TAG_RSC_OP"[@id='%s']"
+// An explicit shutdown-lock of 0 means the lock has been cleared
+static bool
+shutdown_lock_cleared(xmlNode *lrm_resource)
+{
+ time_t shutdown_lock = 0;
+
+ return (crm_element_value_epoch(lrm_resource, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
+ &shutdown_lock) == pcmk_ok)
+ && (shutdown_lock == 0);
+}
+
static void
te_update_diff_v1(const char *event, xmlNode *diff)
{
int lpc, max;
xmlXPathObject *xpathObj = NULL;
CRM_CHECK(diff != NULL, return);
xml_log_patchset(LOG_TRACE, __FUNCTION__, diff);
if (cib_config_changed(NULL, NULL, &diff)) {
abort_transition(INFINITY, tg_restart, "Non-status change", diff);
goto bail; /* configuration changed */
}
/* Tickets Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: update", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Tickets Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Transient Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_TAG_TRANSIENT_NODEATTRS "//" XML_CIB_TAG_NVPAIR);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *attr = getXpathResult(xpathObj, lpc);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
const char *value = NULL;
if (safe_str_eq(CRM_OP_PROBED, name)) {
value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
}
if (crm_is_true(value) == FALSE) {
abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr);
crm_log_xml_trace(attr, "Abort");
goto bail;
}
}
freeXpathObject(xpathObj);
/* Transient Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//"
XML_TAG_TRANSIENT_NODEATTRS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
+ // Check for lrm_resource entries
+ xpathObj = xpath_search(diff,
+ "//" F_CIB_UPDATE_RESULT
+ "//" XML_TAG_DIFF_ADDED
+ "//" XML_LRM_TAG_RESOURCE);
+ max = numXpathResults(xpathObj);
+
/*
- * Updates by, or in response to, TE actions will never contain updates
- * for more than one resource at a time, so such updates indicate an
- * LRM refresh.
- *
- * In that case, start a new transition rather than check each result
- * individually, which can result in _huge_ speedups in large clusters.
+ * Updates by, or in response to, graph actions will never affect more than
+ * one resource at a time, so such updates indicate an LRM refresh. In that
+ * case, start a new transition rather than check each result individually,
+ * which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
- if (transition_graph->pending == 0) {
- xpathObj = xpath_search(diff,
- "//" F_CIB_UPDATE_RESULT
- "//" XML_TAG_DIFF_ADDED
- "//" XML_LRM_TAG_RESOURCE);
- max = numXpathResults(xpathObj);
- if (max > 1) {
- crm_debug("Ignoring resource operation updates due to history refresh of %d resources",
- max);
- crm_log_xml_trace(diff, "lrm-refresh");
- abort_transition(INFINITY, tg_restart, "History refresh", NULL);
- goto bail;
+ if ((transition_graph->pending == 0) && (max > 1)) {
+ crm_debug("Ignoring resource operation updates due to history refresh of %d resources",
+ max);
+ crm_log_xml_trace(diff, "lrm-refresh");
+ abort_transition(INFINITY, tg_restart, "History refresh", NULL);
+ goto bail;
+ }
+
+ if (max == 1) {
+ xmlNode *lrm_resource = getXpathResult(xpathObj, 0);
+
+ if (shutdown_lock_cleared(lrm_resource)) {
+ // @TODO would be more efficient to abort once after transition done
+ abort_transition(INFINITY, tg_restart, "Shutdown lock cleared",
+ lrm_resource);
+ // Still process results, so we stop timers and update failcounts
}
- freeXpathObject(xpathObj);
}
+ freeXpathObject(xpathObj);
/* Process operation updates */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RSC_OP);
max = numXpathResults(xpathObj);
if (max > 0) {
int lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
}
freeXpathObject(xpathObj);
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff, "//" XML_TAG_DIFF_REMOVED "//" XML_LRM_TAG_RSC_OP);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
int path_max = 0;
const char *op_id = NULL;
char *rsc_op_xpath = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if(match == NULL) { continue; };
op_id = ID(match);
path_max = strlen(RSC_OP_TEMPLATE) + strlen(op_id) + 1;
rsc_op_xpath = calloc(1, path_max);
snprintf(rsc_op_xpath, path_max, RSC_OP_TEMPLATE, op_id);
op_match = xpath_search(diff, rsc_op_xpath);
if (numXpathResults(op_match) == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
crm_action_t *cancelled = get_cancel_action(op_id, node);
if (cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id,
node);
abort_transition(INFINITY, tg_restart, "Resource op removal", match);
freeXpathObject(op_match);
free(rsc_op_xpath);
goto bail;
} else {
crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d",
op_id, node, cancelled->id);
}
}
freeXpathObject(op_match);
free(rsc_op_xpath);
}
bail:
freeXpathObject(xpathObj);
}
static void
process_lrm_resource_diff(xmlNode *lrm_resource, const char *node)
{
for (xmlNode *rsc_op = __xml_first_child(lrm_resource); rsc_op != NULL;
rsc_op = __xml_next(rsc_op)) {
process_graph_event(rsc_op, node);
}
+ if (shutdown_lock_cleared(lrm_resource)) {
+ // @TODO would be more efficient to abort once after transition done
+ abort_transition(INFINITY, tg_restart, "Shutdown lock cleared",
+ lrm_resource);
+ }
}
static void
process_resource_updates(const char *node, xmlNode *xml, xmlNode *change,
const char *op, const char *xpath)
{
xmlNode *rsc = NULL;
if (xml == NULL) {
return;
} else if (strcmp((const char*)xml->name, XML_CIB_TAG_LRM) == 0) {
xml = first_named_child(xml, XML_LRM_TAG_RESOURCES);
crm_trace("Got %p in %s", xml, XML_CIB_TAG_LRM);
}
CRM_ASSERT(strcmp((const char*)xml->name, XML_LRM_TAG_RESOURCES) == 0);
/*
* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time, so such updates indicate an
* LRM refresh.
*
* In that case, start a new transition rather than check each result
* individually, which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
if ((transition_graph->pending == 0)
&& xml->children && xml->children->next) {
crm_log_xml_trace(change, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "History refresh", NULL);
return;
}
for (rsc = __xml_first_child(xml); rsc != NULL; rsc = __xml_next(rsc)) {
crm_trace("Processing %s", ID(rsc));
process_lrm_resource_diff(rsc, node);
}
}
#define NODE_PATT "/lrm[@id="
static char *get_node_from_xpath(const char *xpath)
{
char *nodeid = NULL;
char *tmp = strstr(xpath, NODE_PATT);
if(tmp) {
tmp += strlen(NODE_PATT);
tmp += 1;
nodeid = strdup(tmp);
tmp = strstr(nodeid, "\'");
CRM_ASSERT(tmp);
tmp[0] = 0;
}
return nodeid;
}
static char *extract_node_uuid(const char *xpath)
{
char *mutable_path = strdup(xpath);
char *node_uuid = NULL;
char *search = NULL;
char *match = NULL;
match = strstr(mutable_path, "node_state[@id=\'");
if (match == NULL) {
free(mutable_path);
return NULL;
}
match += strlen("node_state[@id=\'");
search = strchr(match, '\'');
if (search == NULL) {
free(mutable_path);
return NULL;
}
search[0] = 0;
node_uuid = strdup(match);
free(mutable_path);
return node_uuid;
}
static void
abort_unless_down(const char *xpath, const char *op, xmlNode *change,
const char *reason)
{
char *node_uuid = NULL;
crm_action_t *down = NULL;
if(safe_str_neq(op, "delete")) {
abort_transition(INFINITY, tg_restart, reason, change);
return;
}
node_uuid = extract_node_uuid(xpath);
if(node_uuid == NULL) {
crm_err("Could not extract node ID from %s", xpath);
abort_transition(INFINITY, tg_restart, reason, change);
return;
}
down = match_down_event(node_uuid);
if (down == NULL) {
crm_trace("Not expecting %s to be down (%s)", node_uuid, xpath);
abort_transition(INFINITY, tg_restart, reason, change);
} else {
crm_trace("Expecting changes to %s (%s)", node_uuid, xpath);
}
free(node_uuid);
}
static void
process_op_deletion(const char *xpath, xmlNode *change)
{
char *mutable_key = strdup(xpath);
char *key;
char *node_uuid;
// Extract the part of xpath between last pair of single quotes
key = strrchr(mutable_key, '\'');
if (key != NULL) {
*key = '\0';
key = strrchr(mutable_key, '\'');
}
if (key == NULL) {
crm_warn("Ignoring malformed CIB update (resource deletion of %s)",
xpath);
free(mutable_key);
return;
}
++key;
node_uuid = extract_node_uuid(xpath);
if (confirm_cancel_action(key, node_uuid) == FALSE) {
abort_transition(INFINITY, tg_restart, "Resource operation removal",
change);
}
free(mutable_key);
free(node_uuid);
}
static void
process_delete_diff(const char *xpath, const char *op, xmlNode *change)
{
if (strstr(xpath, "/" XML_LRM_TAG_RSC_OP "[")) {
process_op_deletion(xpath, change);
} else if (strstr(xpath, "/" XML_CIB_TAG_LRM "[")) {
abort_unless_down(xpath, op, change, "Resource state removal");
} else if (strstr(xpath, "/" XML_CIB_TAG_STATE "[")) {
abort_unless_down(xpath, op, change, "Node state removal");
} else {
crm_trace("Ignoring delete of %s", xpath);
}
}
static void
process_node_state_diff(xmlNode *state, xmlNode *change, const char *op,
const char *xpath)
{
xmlNode *lrm = first_named_child(state, XML_CIB_TAG_LRM);
process_resource_updates(ID(state), lrm, change, op, xpath);
}
static void
process_status_diff(xmlNode *status, xmlNode *change, const char *op,
const char *xpath)
{
for (xmlNode *state = __xml_first_child(status); state != NULL;
state = __xml_next(state)) {
process_node_state_diff(state, change, op, xpath);
}
}
static void
process_cib_diff(xmlNode *cib, xmlNode *change, const char *op,
const char *xpath)
{
xmlNode *status = first_named_child(cib, XML_CIB_TAG_STATUS);
xmlNode *config = first_named_child(cib, XML_CIB_TAG_CONFIGURATION);
if (status) {
process_status_diff(status, change, op, xpath);
}
if (config) {
abort_transition(INFINITY, tg_restart,
"Non-status-only change", change);
}
}
static void
te_update_diff_v2(xmlNode *diff)
{
crm_log_xml_trace(diff, "Patch:Raw");
for (xmlNode *change = __xml_first_child(diff); change != NULL;
change = __xml_next(change)) {
xmlNode *match = NULL;
const char *name = NULL;
const char *xpath = crm_element_value(change, XML_DIFF_PATH);
// Possible ops: create, modify, delete, move
const char *op = crm_element_value(change, XML_DIFF_OP);
// Ignore uninteresting updates
if (op == NULL) {
continue;
} else if (xpath == NULL) {
crm_trace("Ignoring %s change for version field", op);
continue;
} else if (strcmp(op, "move") == 0) {
crm_trace("Ignoring move change at %s", xpath);
continue;
}
// Find the result of create/modify ops
if (strcmp(op, "create") == 0) {
match = change->children;
} else if (strcmp(op, "modify") == 0) {
match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
match = match->children;
}
} else if (strcmp(op, "delete") != 0) {
crm_warn("Ignoring malformed CIB update (%s operation on %s is unrecognized)",
op, xpath);
continue;
}
if (match) {
if (match->type == XML_COMMENT_NODE) {
crm_trace("Ignoring %s operation for comment at %s", op, xpath);
continue;
}
name = (const char *)match->name;
}
crm_trace("Handling %s operation for %s%s%s",
op, (xpath? xpath : "CIB"),
(name? " matched by " : ""), (name? name : ""));
if (strstr(xpath, "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION)) {
abort_transition(INFINITY, tg_restart, "Configuration change",
change);
break; // Won't be packaged with operation results we may be waiting for
} else if (strstr(xpath, "/" XML_CIB_TAG_TICKETS)
|| safe_str_eq(name, XML_CIB_TAG_TICKETS)) {
abort_transition(INFINITY, tg_restart, "Ticket attribute change", change);
break; // Won't be packaged with operation results we may be waiting for
} else if (strstr(xpath, "/" XML_TAG_TRANSIENT_NODEATTRS "[")
|| safe_str_eq(name, XML_TAG_TRANSIENT_NODEATTRS)) {
abort_unless_down(xpath, op, change, "Transient attribute change");
break; // Won't be packaged with operation results we may be waiting for
} else if (strcmp(op, "delete") == 0) {
process_delete_diff(xpath, op, change);
} else if (name == NULL) {
crm_warn("Ignoring malformed CIB update (%s at %s has no result)",
op, xpath);
} else if (strcmp(name, XML_TAG_CIB) == 0) {
process_cib_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_STATUS) == 0) {
process_status_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_STATE) == 0) {
process_node_state_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_LRM) == 0) {
process_resource_updates(ID(match), match, change, op, xpath);
} else if (strcmp(name, XML_LRM_TAG_RESOURCES) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_resource_updates(local_node, match, change, op, xpath);
free(local_node);
} else if (strcmp(name, XML_LRM_TAG_RESOURCE) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_lrm_resource_diff(match, local_node);
free(local_node);
} else if (strcmp(name, XML_LRM_TAG_RSC_OP) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_graph_event(match, local_node);
free(local_node);
} else {
crm_warn("Ignoring malformed CIB update (%s at %s has unrecognized result %s)",
op, xpath, name);
}
}
}
void
te_update_diff(const char *event, xmlNode * msg)
{
xmlNode *diff = NULL;
const char *op = NULL;
int rc = -EINVAL;
int format = 1;
int p_add[] = { 0, 0, 0 };
int p_del[] = { 0, 0, 0 };
CRM_CHECK(msg != NULL, return);
crm_element_value_int(msg, F_CIB_RC, &rc);
if (transition_graph == NULL) {
crm_trace("No graph");
return;
} else if (rc < pcmk_ok) {
crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc));
return;
} else if (transition_graph->complete
&& fsa_state != S_IDLE
&& fsa_state != S_TRANSITION_ENGINE
&& fsa_state != S_POLICY_ENGINE) {
crm_trace("Filter state=%s, complete=%d", fsa_state2string(fsa_state),
transition_graph->complete);
return;
}
op = crm_element_value(msg, F_CIB_OPERATION);
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
xml_patch_versions(diff, p_add, p_del);
crm_debug("Processing (%s) diff: %d.%d.%d -> %d.%d.%d (%s)", op,
p_del[0], p_del[1], p_del[2], p_add[0], p_add[1], p_add[2],
fsa_state2string(fsa_state));
crm_element_value_int(diff, "format", &format);
switch (format) {
case 1:
te_update_diff_v1(event, diff);
break;
case 2:
te_update_diff_v2(diff);
break;
default:
crm_warn("Ignoring malformed CIB update (unknown patch format %d)",
format);
}
}
gboolean
process_te_message(xmlNode * msg, xmlNode * xml_data)
{
const char *from = crm_element_value(msg, F_ORIG);
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM);
const char *ref = crm_element_value(msg, F_CRM_REFERENCE);
const char *op = crm_element_value(msg, F_CRM_TASK);
const char *type = crm_element_value(msg, F_CRM_MSG_TYPE);
crm_trace("Processing %s (%s) message", op, ref);
crm_log_xml_trace(msg, "ipc");
if (op == NULL) {
/* error */
} else if (sys_to == NULL || strcasecmp(sys_to, CRM_SYSTEM_TENGINE) != 0) {
crm_trace("Bad sys-to %s", crm_str(sys_to));
return FALSE;
} else if (safe_str_eq(op, CRM_OP_INVOKE_LRM)
&& safe_str_eq(sys_from, CRM_SYSTEM_LRMD)
/* && safe_str_eq(type, XML_ATTR_RESPONSE) */
) {
xmlXPathObject *xpathObj = NULL;
crm_log_xml_trace(msg, "Processing (N)ACK");
crm_debug("Processing (N)ACK %s from %s", crm_element_value(msg, F_CRM_REFERENCE), from);
xpathObj = xpath_search(xml_data, "//" XML_LRM_TAG_RSC_OP);
if (numXpathResults(xpathObj)) {
int lpc = 0, max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
freeXpathObject(xpathObj);
} else {
crm_log_xml_err(msg, "Invalid (N)ACK");
freeXpathObject(xpathObj);
return FALSE;
}
} else {
crm_err("Unknown command: %s::%s from %s", type, op, sys_from);
}
crm_trace("finished processing message");
return TRUE;
}
void
cib_action_updated(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc < pcmk_ok) {
crm_err("Update %d FAILED: %s", call_id, pcmk_strerror(rc));
}
}
/*!
* \brief Handle a timeout in node-to-node communication
*
* \param[in] data Pointer to action timer
*
* \return FALSE (indicating that source should be not be re-added)
*/
gboolean
action_timer_callback(gpointer data)
{
crm_action_timer_t *timer = NULL;
const char *task = NULL;
const char *on_node = NULL;
const char *via_node = NULL;
CRM_CHECK(data != NULL, return FALSE);
timer = (crm_action_timer_t *) data;
stop_te_timer(timer);
CRM_CHECK(timer->action != NULL, return FALSE);
task = crm_element_value(timer->action->xml, XML_LRM_ATTR_TASK);
on_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_TARGET);
via_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_ROUTER_NODE);
if (transition_graph->complete) {
crm_notice("Node %s did not send %s result (via %s) within %dms "
"(ignoring because transition not in progress)",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"), timer->timeout);
} else {
/* fail the action */
crm_err("Node %s did not send %s result (via %s) within %dms "
"(action timeout plus cluster-delay)",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"), timer->timeout);
print_action(LOG_ERR, "Aborting transition, action lost: ", timer->action);
timer->action->failed = TRUE;
te_action_confirmed(timer->action, transition_graph);
abort_transition(INFINITY, tg_restart, "Action lost", NULL);
// Record timeout in the CIB if appropriate
if ((timer->action->type == action_type_rsc)
&& controld_action_is_recordable(task)) {
controld_record_action_timeout(timer->action);
}
}
return FALSE;
}
diff --git a/daemons/controld/controld_utils.h b/daemons/controld/controld_utils.h
index f902361bd9..5549636bae 100644
--- a/daemons/controld/controld_utils.h
+++ b/daemons/controld/controld_utils.h
@@ -1,111 +1,116 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef CRMD_UTILS__H
# define CRMD_UTILS__H
# include
# include
# include // CIB_OP_MODIFY
# include // fsa_cib_conn
# include
# define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
# define fsa_cib_update(section, data, options, call_id, user_name) \
if(fsa_cib_conn != NULL) { \
call_id = cib_internal_op( \
fsa_cib_conn, CIB_OP_MODIFY, NULL, section, data, \
NULL, options, user_name); \
\
} else { \
crm_err("No CIB manager connection available"); \
}
static inline void
fsa_cib_anon_update(const char *section, xmlNode *data) {
if (fsa_cib_conn == NULL) {
crm_err("No CIB connection available");
} else {
int opts = cib_scope_local | cib_quorum_override | cib_can_create;
fsa_cib_conn->cmds->modify(fsa_cib_conn, section, data, opts);
}
}
extern gboolean fsa_has_quorum;
+extern bool controld_shutdown_lock_enabled;
extern int last_peer_update;
extern int last_resource_update;
enum node_update_flags {
node_update_none = 0x0000,
node_update_quick = 0x0001,
node_update_cluster = 0x0010,
node_update_peer = 0x0020,
node_update_join = 0x0040,
node_update_expected = 0x0100,
node_update_all = node_update_cluster|node_update_peer|node_update_join|node_update_expected,
};
crm_exit_t crmd_exit(crm_exit_t exit_code);
_Noreturn void crmd_fast_exit(crm_exit_t exit_code);
void pe_subsystem_free(void);
void controld_stop_sched_timer(void);
void controld_free_sched_timer(void);
void controld_expect_sched_reply(xmlNode *msg);
void fsa_dump_actions(long long action, const char *text);
void fsa_dump_inputs(int log_level, const char *text, long long input_register);
gboolean update_dc(xmlNode * msg);
void crm_update_peer_join(const char *source, crm_node_t * node, enum crm_join_phase phase);
xmlNode *create_node_state_update(crm_node_t *node, int flags,
xmlNode *parent, const char *source);
void populate_cib_nodes(enum node_update_flags flags, const char *source);
void crm_update_quorum(gboolean quorum, gboolean force_update);
void controld_close_attrd_ipc(void);
void update_attrd(const char *host, const char *name, const char *value, const char *user_name, gboolean is_remote_node);
void update_attrd_remote_node_removed(const char *host, const char *user_name);
void update_attrd_clear_failures(const char *host, const char *rsc,
const char *op, const char *interval_spec,
gboolean is_remote_node);
int crmd_join_phase_count(enum crm_join_phase phase);
void crmd_join_phase_log(int level);
void crmd_peer_down(crm_node_t *peer, bool full);
unsigned int cib_op_timeout(void);
bool feature_set_compatible(const char *dc_version, const char *join_version);
bool controld_action_is_recordable(const char *action);
// Subsections of node_state
enum controld_section_e {
controld_section_lrm,
+ controld_section_lrm_unlocked,
controld_section_attrs,
controld_section_all,
+ controld_section_all_unlocked
};
void controld_delete_node_state(const char *uname,
enum controld_section_e section, int options);
+int controld_delete_resource_history(const char *rsc_id, const char *node,
+ const char *user_name, int call_options);
const char *get_node_id(xmlNode *lrm_rsc_op);
/* Convenience macro for registering a CIB callback
* (assumes that data can be freed with free())
*/
# define fsa_register_cib_callback(id, flag, data, fn) do { \
CRM_ASSERT(fsa_cib_conn); \
fsa_cib_conn->cmds->register_callback_full( \
fsa_cib_conn, id, cib_op_timeout(), \
flag, data, #fn, fn, free); \
} while(0)
#endif
diff --git a/doc/Pacemaker_Explained/en-US/Ch-Options.txt b/doc/Pacemaker_Explained/en-US/Ch-Options.txt
index f864987010..35856aaf64 100644
--- a/doc/Pacemaker_Explained/en-US/Ch-Options.txt
+++ b/doc/Pacemaker_Explained/en-US/Ch-Options.txt
@@ -1,438 +1,465 @@
:compat-mode: legacy
= Cluster-Wide Configuration =
== Configuration Layout ==
The cluster is defined by the Cluster Information Base (CIB),
which uses XML notation. The simplest CIB, an empty one, looks like this:
.An empty configuration
======
[source,XML]
-------
-------
======
The empty configuration above contains the major sections that make up a CIB:
* +cib+: The entire CIB is enclosed with a +cib+ tag. Certain fundamental settings
are defined as attributes of this tag.
** +configuration+: This section -- the primary focus of this document --
contains traditional configuration information such as what resources the
cluster serves and the relationships among them.
*** +crm_config+: cluster-wide configuration options
*** +nodes+: the machines that host the cluster
*** +resources+: the services run by the cluster
*** +constraints+: indications of how resources should be placed
** +status+: This section contains the history of each resource on each node.
Based on this data, the cluster can construct the complete current
state of the cluster. The authoritative source for this section
is the local executor (pacemaker-execd process) on each cluster node, and
the cluster will occasionally repopulate the entire section. For this
reason, it is never written to disk, and administrators are advised
against modifying it in any way.
In this document, configuration settings will be described as 'properties' or 'options'
based on how they are defined in the CIB:
* Properties are XML attributes of an XML element.
* Options are name-value pairs expressed as +nvpair+ child elements of an XML element.
Normally, you will use command-line tools that abstract the XML, so the
distinction will be unimportant; both properties and options are
cluster settings you can tweak.
== CIB Properties ==
Certain settings are defined by CIB properties (that is, attributes of the
+cib+ tag) rather than with the rest of the cluster configuration in the
+configuration+ section.
The reason is simply a matter of parsing. These options are used by the
configuration database which is, by design, mostly ignorant of the content it
holds. So the decision was made to place them in an easy-to-find location.
.CIB Properties
[width="95%",cols="2m,<5",options="header",align="center"]
|=========================================================
|Field |Description
| admin_epoch |
indexterm:[Configuration Version,Cluster]
indexterm:[Cluster,Option,Configuration Version]
indexterm:[admin_epoch,Cluster Option]
indexterm:[Cluster,Option,admin_epoch]
When a node joins the cluster, the cluster performs a check to see
which node has the best configuration. It asks the node with the highest
(+admin_epoch+, +epoch+, +num_updates+) tuple to replace the configuration on
all the nodes -- which makes setting them, and setting them correctly, very
important. +admin_epoch+ is never modified by the cluster; you can use this
to make the configurations on any inactive nodes obsolete. _Never set this
value to zero_. In such cases, the cluster cannot tell the difference between
your configuration and the "empty" one used when nothing is found on disk.
| epoch |
indexterm:[epoch,Cluster Option]
indexterm:[Cluster,Option,epoch]
The cluster increments this every time the configuration is updated (usually by
the administrator).
| num_updates |
indexterm:[num_updates,Cluster Option]
indexterm:[Cluster,Option,num_updates]
The cluster increments this every time the configuration or status is updated
(usually by the cluster) and resets it to 0 when epoch changes.
| validate-with |
indexterm:[validate-with,Cluster Option]
indexterm:[Cluster,Option,validate-with]
Determines the type of XML validation that will be done on the configuration.
If set to +none+, the cluster will not verify that updates conform to the
DTD (nor reject ones that don't). This option can be useful when
operating a mixed-version cluster during an upgrade.
|cib-last-written |
indexterm:[cib-last-written,Cluster Property]
indexterm:[Cluster,Property,cib-last-written]
Indicates when the configuration was last written to disk. Maintained by the
cluster; for informational purposes only.
|have-quorum |
indexterm:[have-quorum,Cluster Property]
indexterm:[Cluster,Property,have-quorum]
Indicates if the cluster has quorum. If false, this may mean that the
cluster cannot start resources or fence other nodes (see
+no-quorum-policy+ below). Maintained by the cluster.
|dc-uuid |
indexterm:[dc-uuid,Cluster Property]
indexterm:[Cluster,Property,dc-uuid]
Indicates which cluster node is the current leader. Used by the
cluster when placing resources and determining the order of some
events. Maintained by the cluster.
|=========================================================
[[s-cluster-options]]
== Cluster Options ==
Cluster options, as you might expect, control how the cluster behaves
when confronted with certain situations.
They are grouped into sets within the +crm_config+ section, and, in advanced
configurations, there may be more than one set. (This will be described later
in the section on <> where we will show how to have the cluster use
different sets of options during working hours than during weekends.) For now,
we will describe the simple case where each option is present at most once.
You can obtain an up-to-date list of cluster options, including
their default values, by running the `man pacemaker-schedulerd` and
`man pacemaker-controld` commands.
.Cluster Options
[width="95%",cols="5m,2,<11",options="header",align="center"]
|=========================================================
|Option |Default |Description
| cluster-name | |
indexterm:[cluster-name,Cluster Property]
indexterm:[Cluster,Property,cluster-name]
An (optional) name for the cluster as a whole. This is mostly for users'
convenience for use as desired in administration, but this can be used
in the Pacemaker configuration in <> (as the
+#cluster-name+ <>). It may
also be used by higher-level tools when displaying cluster information, and by
certain resource agents (for example, the +ocf:heartbeat:GFS2+ agent stores the
cluster name in filesystem meta-data).
| dc-version | |
indexterm:[dc-version,Cluster Property]
indexterm:[Cluster,Property,dc-version]
Version of Pacemaker on the cluster's DC.
Determined automatically by the cluster.
Often includes the hash which identifies the exact Git changeset it was built
from. Used for diagnostic purposes.
| cluster-infrastructure | |
indexterm:[cluster-infrastructure,Cluster Property]
indexterm:[Cluster,Property,cluster-infrastructure]
The messaging stack on which Pacemaker is currently running.
Determined automatically by the cluster.
Used for informational and diagnostic purposes.
| no-quorum-policy | stop
a|
indexterm:[no-quorum-policy,Cluster Option]
indexterm:[Cluster,Option,no-quorum-policy]
What to do when the cluster does not have quorum. Allowed values:
* +ignore:+ continue all resource management
* +freeze:+ continue resource management, but don't recover resources from nodes not in the affected partition
* +stop:+ stop all resources in the affected cluster partition
* +suicide:+ fence all nodes in the affected cluster partition
| batch-limit | 0 |
indexterm:[batch-limit,Cluster Option]
indexterm:[Cluster,Option,batch-limit]
The maximum number of actions that the cluster may execute in parallel across
all nodes. The "correct" value will depend on the speed and load of your
network and cluster nodes. If zero, the cluster will impose a dynamically
calculated limit only when any node has high load.
| migration-limit | -1 |
indexterm:[migration-limit,Cluster Option]
indexterm:[Cluster,Option,migration-limit]
The number of <> actions that the cluster
is allowed to execute in parallel on a node. A value of -1 means unlimited.
| symmetric-cluster | TRUE |
indexterm:[symmetric-cluster,Cluster Option]
indexterm:[Cluster,Option,symmetric-cluster]
Can all resources run on any node by default?
| stop-all-resources | FALSE |
indexterm:[stop-all-resources,Cluster Option]
indexterm:[Cluster,Option,stop-all-resources]
Should the cluster stop all resources?
| stop-orphan-resources | TRUE |
indexterm:[stop-orphan-resources,Cluster Option]
indexterm:[Cluster,Option,stop-orphan-resources]
Should deleted resources be stopped? This value takes precedence over
+is-managed+ (i.e. even unmanaged resources will be stopped if deleted from
the configuration when this value is TRUE).
| stop-orphan-actions | TRUE |
indexterm:[stop-orphan-actions,Cluster Option]
indexterm:[Cluster,Option,stop-orphan-actions]
Should deleted actions be cancelled?
| start-failure-is-fatal | TRUE |
indexterm:[start-failure-is-fatal,Cluster Option]
indexterm:[Cluster,Option,start-failure-is-fatal]
Should a failure to start a resource on a particular node prevent further start
attempts on that node? If FALSE, the cluster will decide whether the same
node is still eligible based on the resource's current failure count
and +migration-threshold+ (see <>).
| enable-startup-probes | TRUE |
indexterm:[enable-startup-probes,Cluster Option]
indexterm:[Cluster,Option,enable-startup-probes]
Should the cluster check for active resources during startup?
| maintenance-mode | FALSE |
indexterm:[maintenance-mode,Cluster Option]
indexterm:[Cluster,Option,maintenance-mode]
Should the cluster refrain from monitoring, starting and stopping resources?
| stonith-enabled | TRUE |
indexterm:[stonith-enabled,Cluster Option]
indexterm:[Cluster,Option,stonith-enabled]
Should failed nodes and nodes with resources that can't be stopped be
shot? If you value your data, set up a STONITH device and enable this.
If true, or unset, the cluster will refuse to start resources unless
one or more STONITH resources have been configured.
If false, unresponsive nodes are immediately assumed to be running no
resources, and resource takeover to online nodes starts without any
further protection (which means _data loss_ if the unresponsive node
still accesses shared storage, for example). See also the +requires+
meta-attribute in <>.
| stonith-action | reboot |
indexterm:[stonith-action,Cluster Option]
indexterm:[Cluster,Option,stonith-action]
Action to send to STONITH device. Allowed values are +reboot+ and +off+.
The value +poweroff+ is also allowed, but is only used for
legacy devices.
| stonith-timeout | 60s |
indexterm:[stonith-timeout,Cluster Option]
indexterm:[Cluster,Option,stonith-timeout]
How long to wait for STONITH actions (reboot, on, off) to complete
| stonith-max-attempts | 10 |
indexterm:[stonith-max-attempts,Cluster Option]
indexterm:[Cluster,Option,stonith-max-attempts]
How many times fencing can fail for a target before the cluster will no longer
immediately re-attempt it.
| stonith-watchdog-timeout | 0 |
indexterm:[stonith-watchdog-timeout,Cluster Option]
indexterm:[Cluster,Option,stonith-watchdog-timeout]
If nonzero, rely on hardware watchdog self-fencing. If positive, assume unseen
nodes self-fence within this much time. If negative, and the
SBD_WATCHDOG_TIMEOUT environment variable is set, use twice that value.
| concurrent-fencing | FALSE |
indexterm:[concurrent-fencing,Cluster Option]
indexterm:[Cluster,Option,concurrent-fencing]
Is the cluster allowed to initiate multiple fence actions concurrently?
| fence-reaction | stop |
indexterm:[fence-reaction,Cluster Option]
indexterm:[Cluster,Option,fence-reaction]
How should a cluster node react if notified of its own fencing? A cluster node
may receive notification of its own fencing if fencing is misconfigured, or if
fabric fencing is in use that doesn't cut cluster communication. Allowed values
are +stop+ to attempt to immediately stop pacemaker and stay stopped, or
+panic+ to attempt to immediately reboot the local node, falling back to stop
on failure. The default is likely to be changed to +panic+ in a future release.
'(since 2.0.3)'
| cluster-delay | 60s |
indexterm:[cluster-delay,Cluster Option]
indexterm:[Cluster,Option,cluster-delay]
Estimated maximum round-trip delay over the network (excluding action
execution). If the DC requires an action to be executed on another
node, it will consider the action failed if it does not get a response
from the other node in this time (after considering the action's
own timeout). The "correct" value will depend on the speed and load of your
network and cluster nodes.
| dc-deadtime | 20s |
indexterm:[dc-deadtime,Cluster Option]
indexterm:[Cluster,Option,dc-deadtime]
How long to wait for a response from other nodes during startup.
The "correct" value will depend on the speed/load of your network and the type of switches used.
| cluster-ipc-limit | 500 |
indexterm:[cluster-ipc-limit,Cluster Option]
indexterm:[Cluster,Option,cluster-ipc-limit]
The maximum IPC message backlog before one cluster daemon will disconnect
another. This is of use in large clusters, for which a good value is the number
of resources in the cluster multiplied by the number of nodes. The default of
500 is also the minimum. Raise this if you see "Evicting client" messages for
cluster daemon PIDs in the logs.
| pe-error-series-max | -1 |
indexterm:[pe-error-series-max,Cluster Option]
indexterm:[Cluster,Option,pe-error-series-max]
The number of PE inputs resulting in ERRORs to save. Used when reporting problems.
A value of -1 means unlimited (report all).
| pe-warn-series-max | -1 |
indexterm:[pe-warn-series-max,Cluster Option]
indexterm:[Cluster,Option,pe-warn-series-max]
The number of PE inputs resulting in WARNINGs to save. Used when reporting problems.
A value of -1 means unlimited (report all).
| pe-input-series-max | -1 |
indexterm:[pe-input-series-max,Cluster Option]
indexterm:[Cluster,Option,pe-input-series-max]
The number of "normal" PE inputs to save. Used when reporting problems.
A value of -1 means unlimited (report all).
| placement-strategy | default |
indexterm:[placement-strategy,Cluster Option]
indexterm:[Cluster,Option,placement-strategy]
How the cluster should allocate resources to nodes (see <>).
Allowed values are +default+, +utilization+, +balanced+, and +minimal+.
| node-health-strategy | none |
indexterm:[node-health-strategy,Cluster Option]
indexterm:[Cluster,Option,node-health-strategy]
How the cluster should react to node health attributes (see <>).
Allowed values are +none+, +migrate-on-red+, +only-green+, +progressive+, and
+custom+.
| enable-acl | FALSE |
indexterm:[enable-acl,Cluster Option]
indexterm:[Cluster,Option,enable-acl]
Whether access control lists (ACLs) (see <>) can be used to authorize
modifications to the CIB.
| node-health-base | 0 |
indexterm:[node-health-base,Cluster Option]
indexterm:[Cluster,Option,node-health-base]
The base health score assigned to a node. Only used when
+node-health-strategy+ is +progressive+.
| node-health-green | 0 |
indexterm:[node-health-green,Cluster Option]
indexterm:[Cluster,Option,node-health-green]
The score to use for a node health attribute whose value is +green+.
Only used when +node-health-strategy+ is +progressive+ or +custom+.
| node-health-yellow | 0 |
indexterm:[node-health-yellow,Cluster Option]
indexterm:[Cluster,Option,node-health-yellow]
The score to use for a node health attribute whose value is +yellow+.
Only used when +node-health-strategy+ is +progressive+ or +custom+.
| node-health-red | 0 |
indexterm:[node-health-red,Cluster Option]
indexterm:[Cluster,Option,node-health-red]
The score to use for a node health attribute whose value is +red+.
Only used when +node-health-strategy+ is +progressive+ or +custom+.
| cluster-recheck-interval | 15min |
indexterm:[cluster-recheck-interval,Cluster Option]
indexterm:[Cluster,Option,cluster-recheck-interval]
Pacemaker is primarily event-driven, and looks ahead to know when to recheck
the cluster for failure timeouts and most time-based rules. However, it will
also recheck the cluster after this amount of inactivity. This has two goals:
rules with +date_spec+ are only guaranteed to be checked this often, and it
also serves as a fail-safe for certain classes of scheduler bugs. A value of 0
disables this polling; positive values are a time interval.
+| shutdown-lock | false |
+The default of false allows active resources to be recovered elsewhere when
+their node is cleanly shut down, which is what the vast majority of users will
+want. However, some users prefer to make resources highly available only for
+failures, with no recovery for clean shutdowns. If this option is true,
+resources active on a node when it is cleanly shut down are kept "locked" to
+that node (not allowed to run elsewhere) until they start again on that node
+after it rejoins (or for at most shutdown-lock-limit, if set). Stonith
+resources and Pacemaker Remote connections are never locked. Clone and bundle
+instances and the master role of promotable clones are currently never locked,
+though support could be added in a future release. Locks may be manually
+cleared using the `--refresh` option of `crm_resource` (both the resource and
+node must be specified; this works with remote nodes if their connection
+resource's target-role is set to Stopped, but not if Pacemaker Remote is
+stopped on the remote node without disabling the connection resource).
+indexterm:[shutdown-lock,Cluster Option]
+indexterm:[Cluster,Option,shutdown-lock]
+
+| shutdown-lock-limit | 0 |
+If shutdown-lock is true, and this is set to a nonzero time duration, locked
+resources will be allowed to start after this much time has passed since the
+node shutdown was initiated, even if the node has not rejoined. (This works
+with remote nodes only if their connection resource's target-role is set to
+Stopped.)
+indexterm:[shutdown-lock-limit,Cluster Option]
+indexterm:[Cluster,Option,shutdown-lock-limit]
+
| remove-after-stop | FALSE |
indexterm:[remove-after-stop,Cluster Option]
indexterm:[Cluster,Option,remove-after-stop]
_Advanced Use Only:_ Should the cluster remove resources from the LRM after
they are stopped? Values other than the default are, at best, poorly tested and
potentially dangerous.
| startup-fencing | TRUE |
indexterm:[startup-fencing,Cluster Option]
indexterm:[Cluster,Option,startup-fencing]
_Advanced Use Only:_ Should the cluster shoot unseen nodes?
Not using the default is very unsafe!
| election-timeout | 2min |
indexterm:[election-timeout,Cluster Option]
indexterm:[Cluster,Option,election-timeout]
_Advanced Use Only:_ If you need to adjust this value, it probably indicates
the presence of a bug.
| shutdown-escalation | 20min |
indexterm:[shutdown-escalation,Cluster Option]
indexterm:[Cluster,Option,shutdown-escalation]
_Advanced Use Only:_ If you need to adjust this value, it probably indicates
the presence of a bug.
| join-integration-timeout | 3min |
indexterm:[join-integration-timeout,Cluster Option]
indexterm:[Cluster,Option,join-integration-timeout]
_Advanced Use Only:_ If you need to adjust this value, it probably indicates
the presence of a bug.
| join-finalization-timeout | 30min |
indexterm:[join-finalization-timeout,Cluster Option]
indexterm:[Cluster,Option,join-finalization-timeout]
_Advanced Use Only:_ If you need to adjust this value, it probably indicates
the presence of a bug.
| transition-delay | 0s |
indexterm:[transition-delay,Cluster Option]
indexterm:[Cluster,Option,transition-delay]
_Advanced Use Only:_ Delay cluster recovery for the configured interval to
allow for additional/related events to occur. Useful if your configuration is
sensitive to the order in which ping updates arrive.
Enabling this option will slow down cluster recovery under
all conditions.
|=========================================================
diff --git a/include/crm/crm.h b/include/crm/crm.h
index cbf72d3236..d2ffb610df 100644
--- a/include/crm/crm.h
+++ b/include/crm/crm.h
@@ -1,231 +1,231 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef CRM__H
# define CRM__H
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief A dumping ground
* \ingroup core
*/
# include
# include
# include
# include
# include
# include
/*!
* The CRM feature set assists with compatibility in mixed-version clusters.
* The major version number increases when nodes with different versions
* would not work (rolling upgrades are not allowed). The minor version
* number increases when mixed-version clusters are allowed only during
* rolling upgrades (a node with the oldest feature set will be elected DC). The
* minor-minor version number is ignored, but allows resource agents to detect
* cluster support for various features.
*
* The feature set also affects the processing of old saved CIBs (such as for
* many scheduler regression tests).
*
* Particular feature points currently used by pacemaker:
*
* >2.1: Operation updates include timing data
* >=3.0.5: XML v2 digests are created
* >=3.0.8: Peers do not need acks for cancellations
* >=3.0.9: DC will send its own shutdown request to all peers
* XML v2 patchsets are created by default
* >=3.0.13: Fail counts include operation name and interval
* >=3.2.0: DC supports PCMK_LRM_OP_INVALID and PCMK_LRM_OP_NOT_CONNECTED
*/
-# define CRM_FEATURE_SET "3.2.0"
+# define CRM_FEATURE_SET "3.3.0"
# define EOS '\0'
# define DIMOF(a) ((int) (sizeof(a)/sizeof(a[0])) )
# ifndef MAX_NAME
# define MAX_NAME 256
# endif
# ifndef __GNUC__
# define __builtin_expect(expr, result) (expr)
# endif
/* Some handy macros used by the Linux kernel */
# define __likely(expr) __builtin_expect(expr, 1)
# define __unlikely(expr) __builtin_expect(expr, 0)
# define CRM_META "CRM_meta"
extern char *crm_system_name;
/* *INDENT-OFF* */
// Used for some internal IPC timeouts (maybe should be configurable option)
# define MAX_IPC_DELAY 120
// How we represent "infinite" scores
# define CRM_SCORE_INFINITY 1000000
# define CRM_INFINITY_S "INFINITY"
# define CRM_PLUS_INFINITY_S "+" CRM_INFINITY_S
# define CRM_MINUS_INFINITY_S "-" CRM_INFINITY_S
/* @COMPAT API < 2.0.0 Deprecated "infinity" aliases
*
* INFINITY might be defined elsewhere (e.g. math.h), so undefine it first.
* This, of course, complicates any attempt to use the other definition in any
* code that includes this header.
*/
# undef INFINITY
# define INFINITY_S "INFINITY"
# define MINUS_INFINITY_S "-INFINITY"
# define INFINITY 1000000
/* Sub-systems */
# define CRM_SYSTEM_DC "dc"
# define CRM_SYSTEM_DCIB "dcib"
/* The master CIB */
# define CRM_SYSTEM_CIB "cib"
# define CRM_SYSTEM_CRMD "crmd"
# define CRM_SYSTEM_LRMD "lrmd"
# define CRM_SYSTEM_PENGINE "pengine"
# define CRM_SYSTEM_TENGINE "tengine"
# define CRM_SYSTEM_STONITHD "stonithd"
# define CRM_SYSTEM_MCP "pacemakerd"
// Names of internally generated node attributes
# define CRM_ATTR_UNAME "#uname"
# define CRM_ATTR_ID "#id"
# define CRM_ATTR_KIND "#kind"
# define CRM_ATTR_ROLE "#role"
# define CRM_ATTR_IS_DC "#is_dc"
# define CRM_ATTR_CLUSTER_NAME "#cluster-name"
# define CRM_ATTR_SITE_NAME "#site-name"
# define CRM_ATTR_UNFENCED "#node-unfenced"
# define CRM_ATTR_DIGESTS_ALL "#digests-all"
# define CRM_ATTR_DIGESTS_SECURE "#digests-secure"
# define CRM_ATTR_RA_VERSION "#ra-version"
# define CRM_ATTR_PROTOCOL "#attrd-protocol"
/* Valid operations */
# define CRM_OP_NOOP "noop"
# define CRM_OP_JOIN_ANNOUNCE "join_announce"
# define CRM_OP_JOIN_OFFER "join_offer"
# define CRM_OP_JOIN_REQUEST "join_request"
# define CRM_OP_JOIN_ACKNAK "join_ack_nack"
# define CRM_OP_JOIN_CONFIRM "join_confirm"
# define CRM_OP_PING "ping"
# define CRM_OP_NODE_INFO "node-info"
# define CRM_OP_THROTTLE "throttle"
# define CRM_OP_VOTE "vote"
# define CRM_OP_NOVOTE "no-vote"
# define CRM_OP_HELLO "hello"
# define CRM_OP_PECALC "pe_calc"
# define CRM_OP_QUIT "quit"
# define CRM_OP_LOCAL_SHUTDOWN "start_shutdown"
# define CRM_OP_SHUTDOWN_REQ "req_shutdown"
# define CRM_OP_SHUTDOWN "do_shutdown"
# define CRM_OP_FENCE "stonith"
# define CRM_OP_REGISTER "register"
# define CRM_OP_IPC_FWD "ipc_fwd"
# define CRM_OP_INVOKE_LRM "lrm_invoke"
# define CRM_OP_LRM_REFRESH "lrm_refresh" /* Deprecated */
# define CRM_OP_LRM_QUERY "lrm_query"
# define CRM_OP_LRM_DELETE "lrm_delete"
# define CRM_OP_LRM_FAIL "lrm_fail"
# define CRM_OP_PROBED "probe_complete"
# define CRM_OP_REPROBE "probe_again"
# define CRM_OP_CLEAR_FAILCOUNT "clear_failcount"
# define CRM_OP_REMOTE_STATE "remote_state"
# define CRM_OP_RELAXED_SET "one-or-more"
# define CRM_OP_RELAXED_CLONE "clone-one-or-more"
# define CRM_OP_RM_NODE_CACHE "rm_node_cache"
# define CRM_OP_MAINTENANCE_NODES "maintenance_nodes"
/* Possible cluster membership states */
# define CRMD_JOINSTATE_DOWN "down"
# define CRMD_JOINSTATE_PENDING "pending"
# define CRMD_JOINSTATE_MEMBER "member"
# define CRMD_JOINSTATE_NACK "banned"
# define CRMD_ACTION_DELETE "delete"
# define CRMD_ACTION_CANCEL "cancel"
# define CRMD_ACTION_RELOAD "reload"
# define CRMD_ACTION_MIGRATE "migrate_to"
# define CRMD_ACTION_MIGRATED "migrate_from"
# define CRMD_ACTION_START "start"
# define CRMD_ACTION_STARTED "running"
# define CRMD_ACTION_STOP "stop"
# define CRMD_ACTION_STOPPED "stopped"
# define CRMD_ACTION_PROMOTE "promote"
# define CRMD_ACTION_PROMOTED "promoted"
# define CRMD_ACTION_DEMOTE "demote"
# define CRMD_ACTION_DEMOTED "demoted"
# define CRMD_ACTION_NOTIFY "notify"
# define CRMD_ACTION_NOTIFIED "notified"
# define CRMD_ACTION_STATUS "monitor"
# define CRMD_ACTION_METADATA "meta-data"
# define CRMD_METADATA_CALL_TIMEOUT 30000
/* short names */
# define RSC_DELETE CRMD_ACTION_DELETE
# define RSC_CANCEL CRMD_ACTION_CANCEL
# define RSC_MIGRATE CRMD_ACTION_MIGRATE
# define RSC_MIGRATED CRMD_ACTION_MIGRATED
# define RSC_START CRMD_ACTION_START
# define RSC_STARTED CRMD_ACTION_STARTED
# define RSC_STOP CRMD_ACTION_STOP
# define RSC_STOPPED CRMD_ACTION_STOPPED
# define RSC_PROMOTE CRMD_ACTION_PROMOTE
# define RSC_PROMOTED CRMD_ACTION_PROMOTED
# define RSC_DEMOTE CRMD_ACTION_DEMOTE
# define RSC_DEMOTED CRMD_ACTION_DEMOTED
# define RSC_NOTIFY CRMD_ACTION_NOTIFY
# define RSC_NOTIFIED CRMD_ACTION_NOTIFIED
# define RSC_STATUS CRMD_ACTION_STATUS
# define RSC_METADATA CRMD_ACTION_METADATA
/* *INDENT-ON* */
typedef GList *GListPtr;
# include
# include
static inline const char *
crm_action_str(const char *task, guint interval_ms) {
if(safe_str_eq(task, RSC_STATUS) && !interval_ms) {
return "probe";
}
return task;
}
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/lrmd.h b/include/crm/lrmd.h
index cfa2925167..3ad1f05c33 100644
--- a/include/crm/lrmd.h
+++ b/include/crm/lrmd.h
@@ -1,564 +1,566 @@
/*
* Copyright 2012-2019 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef LRMD__H
# define LRMD__H
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief Resource agent executor
* \ingroup lrmd
*/
#include // bool
#include // guint, GList
#include
#include
typedef struct lrmd_s lrmd_t;
typedef struct lrmd_key_value_s {
char *key;
char *value;
struct lrmd_key_value_s *next;
} lrmd_key_value_t;
/* This should be bumped every time there is an incompatible change that
* prevents older clients from connecting to this version of the server.
*/
#define LRMD_PROTOCOL_VERSION "1.1"
/* This is the version that the client version will actually be compared
* against. This should be identical to LRMD_PROTOCOL_VERSION. However, we
* accidentally bumped LRMD_PROTOCOL_VERSION in 6424a647 (1.1.15) when we didn't
* need to, so for now it's different. If we ever have a truly incompatible
* bump, we can drop this and compare against LRMD_PROTOCOL_VERSION.
*/
#define LRMD_MIN_PROTOCOL_VERSION "1.0"
/* *INDENT-OFF* */
#define DEFAULT_REMOTE_KEY_LOCATION PACEMAKER_CONFIG_DIR "/authkey"
#define ALT_REMOTE_KEY_LOCATION "/etc/corosync/authkey"
#define DEFAULT_REMOTE_PORT 3121
#define DEFAULT_REMOTE_USERNAME "lrmd"
#define F_LRMD_OPERATION "lrmd_op"
#define F_LRMD_CLIENTNAME "lrmd_clientname"
#define F_LRMD_IS_IPC_PROVIDER "lrmd_is_ipc_provider"
#define F_LRMD_CLIENTID "lrmd_clientid"
#define F_LRMD_PROTOCOL_VERSION "lrmd_protocol_version"
#define F_LRMD_REMOTE_MSG_TYPE "lrmd_remote_msg_type"
#define F_LRMD_REMOTE_MSG_ID "lrmd_remote_msg_id"
#define F_LRMD_CALLBACK_TOKEN "lrmd_async_id"
#define F_LRMD_CALLID "lrmd_callid"
#define F_LRMD_CALLOPTS "lrmd_callopt"
#define F_LRMD_CALLDATA "lrmd_calldata"
#define F_LRMD_RC "lrmd_rc"
#define F_LRMD_EXEC_RC "lrmd_exec_rc"
#define F_LRMD_OP_STATUS "lrmd_exec_op_status"
#define F_LRMD_TIMEOUT "lrmd_timeout"
#define F_LRMD_WATCHDOG "lrmd_watchdog"
#define F_LRMD_CLASS "lrmd_class"
#define F_LRMD_PROVIDER "lrmd_provider"
#define F_LRMD_TYPE "lrmd_type"
#define F_LRMD_ORIGIN "lrmd_origin"
#define F_LRMD_RSC_RUN_TIME "lrmd_run_time"
#define F_LRMD_RSC_RCCHANGE_TIME "lrmd_rcchange_time"
#define F_LRMD_RSC_EXEC_TIME "lrmd_exec_time"
#define F_LRMD_RSC_QUEUE_TIME "lrmd_queue_time"
#define F_LRMD_RSC_ID "lrmd_rsc_id"
#define F_LRMD_RSC_ACTION "lrmd_rsc_action"
#define F_LRMD_RSC_USERDATA_STR "lrmd_rsc_userdata_str"
#define F_LRMD_RSC_OUTPUT "lrmd_rsc_output"
#define F_LRMD_RSC_EXIT_REASON "lrmd_rsc_exit_reason"
#define F_LRMD_RSC_START_DELAY "lrmd_rsc_start_delay"
#define F_LRMD_RSC_INTERVAL "lrmd_rsc_interval"
#define F_LRMD_RSC_DELETED "lrmd_rsc_deleted"
#define F_LRMD_RSC "lrmd_rsc"
#define F_LRMD_ALERT_ID "lrmd_alert_id"
#define F_LRMD_ALERT_PATH "lrmd_alert_path"
#define F_LRMD_ALERT "lrmd_alert"
#define LRMD_OP_RSC_REG "lrmd_rsc_register"
#define LRMD_OP_RSC_EXEC "lrmd_rsc_exec"
#define LRMD_OP_RSC_CANCEL "lrmd_rsc_cancel"
#define LRMD_OP_RSC_UNREG "lrmd_rsc_unregister"
#define LRMD_OP_RSC_INFO "lrmd_rsc_info"
#define LRMD_OP_RSC_METADATA "lrmd_rsc_metadata"
#define LRMD_OP_POKE "lrmd_rsc_poke"
#define LRMD_OP_NEW_CLIENT "lrmd_rsc_new_client"
#define LRMD_OP_CHECK "lrmd_check"
#define LRMD_OP_ALERT_EXEC "lrmd_alert_exec"
#define LRMD_OP_GET_RECURRING "lrmd_get_recurring"
#define LRMD_IPC_OP_NEW "new"
#define LRMD_IPC_OP_DESTROY "destroy"
#define LRMD_IPC_OP_EVENT "event"
#define LRMD_IPC_OP_REQUEST "request"
#define LRMD_IPC_OP_RESPONSE "response"
#define LRMD_IPC_OP_SHUTDOWN_REQ "shutdown_req"
#define LRMD_IPC_OP_SHUTDOWN_ACK "shutdown_ack"
#define LRMD_IPC_OP_SHUTDOWN_NACK "shutdown_nack"
#define F_LRMD_IPC_OP "lrmd_ipc_op"
#define F_LRMD_IPC_IPC_SERVER "lrmd_ipc_server"
#define F_LRMD_IPC_SESSION "lrmd_ipc_session"
#define F_LRMD_IPC_CLIENT "lrmd_ipc_client"
#define F_LRMD_IPC_USER "lrmd_ipc_user"
#define F_LRMD_IPC_MSG "lrmd_ipc_msg"
#define F_LRMD_IPC_MSG_ID "lrmd_ipc_msg_id"
#define F_LRMD_IPC_MSG_FLAGS "lrmd_ipc_msg_flags"
#define T_LRMD "lrmd"
#define T_LRMD_REPLY "lrmd_reply"
#define T_LRMD_NOTIFY "lrmd_notify"
#define T_LRMD_IPC_PROXY "lrmd_ipc_proxy"
#define T_LRMD_RSC_OP "lrmd_rsc_op"
/* *INDENT-ON* */
/*!
* \brief Create a new connection to the local executor
*/
lrmd_t *lrmd_api_new(void);
/*!
* \brief Create a new TLS connection to a remote executor
*
* \param nodename name of remote node identified with this connection
* \param server name of server to connect to
* \param port port number to connect to
*
* \note nodename and server may be the same value.
*/
lrmd_t *lrmd_remote_api_new(const char *nodename, const char *server, int port);
/*!
* \brief Use after lrmd_poll returns 1 to read and dispatch a message
*
* \param[in,out] lrmd Executor connection object
*
* \return TRUE if connection is still up, FALSE if disconnected
*/
bool lrmd_dispatch(lrmd_t * lrmd);
/*!
* \brief Poll for a specified timeout period to determine if a message
* is ready for dispatch.
* \retval 1 msg is ready
* \retval 0 timeout occurred
* \retval negative error code
*/
int lrmd_poll(lrmd_t * lrmd, int timeout);
/*!
* \brief Destroy executor connection object
*/
void lrmd_api_delete(lrmd_t * lrmd);
lrmd_key_value_t *lrmd_key_value_add(lrmd_key_value_t * kvp, const char *key, const char *value);
/* *INDENT-OFF* */
/* Reserved for future use */
enum lrmd_call_options {
lrmd_opt_none = 0x00000000,
/* lrmd_opt_sync_call = 0x00000001, //Not implemented, patches welcome. */
/*! Only notify the client originating a exec() the results */
lrmd_opt_notify_orig_only = 0x00000002,
/*! Drop recurring operations initiated by a client when client disconnects.
* This call_option is only valid when registering a resource. When used
* remotely with the pacemaker_remote daemon, this option means that recurring
* operations will be dropped once all the remote connections disconnect. */
lrmd_opt_drop_recurring = 0x00000003,
/*! Send notifications for recurring operations only when the result changes */
lrmd_opt_notify_changes_only = 0x00000004,
};
enum lrmd_callback_event {
lrmd_event_register,
lrmd_event_unregister,
lrmd_event_exec_complete,
lrmd_event_disconnect,
lrmd_event_connect,
lrmd_event_poke,
lrmd_event_new_client,
};
/* *INDENT-ON* */
typedef struct lrmd_event_data_s {
/*! Type of event, register, unregister, call_completed... */
enum lrmd_callback_event type;
/*! The resource this event occurred on. */
const char *rsc_id;
/*! The action performed, start, stop, monitor... */
const char *op_type;
/*! The user data passed by caller of exec() API function */
const char *user_data;
/*! The client api call id associated with this event */
int call_id;
/*! The operation's timeout period in ms. */
int timeout;
/*! The operation's recurring interval in ms. */
guint interval_ms;
/*! The operation's start delay value in ms. */
int start_delay;
/*! This operation that just completed is on a deleted rsc. */
int rsc_deleted;
/*! The executed ra return code mapped to OCF */
enum ocf_exitcode rc;
/*! The executor status returned for exec_complete events */
int op_status;
/*! stdout from resource agent operation */
const char *output;
/*! Timestamp of when op ran */
unsigned int t_run;
/*! Timestamp of last rc change */
unsigned int t_rcchange;
/*! Time in length op took to execute */
unsigned int exec_time;
/*! Time in length spent in queue */
unsigned int queue_time;
/*! int connection result. Used for connection and poke events */
int connection_rc;
/* This is a GHashTable containing the
* parameters given to the operation */
void *params;
/*! client node name associated with this connection
* (used to match actions to the proper client when there are multiple)
*/
const char *remote_nodename;
/*! exit failure reason string from resource agent operation */
const char *exit_reason;
} lrmd_event_data_t;
+lrmd_event_data_t *lrmd_new_event(const char *rsc_id, const char *task,
+ guint interval_ms);
lrmd_event_data_t *lrmd_copy_event(lrmd_event_data_t * event);
void lrmd_free_event(lrmd_event_data_t * event);
typedef struct lrmd_rsc_info_s {
char *id;
char *type;
char *standard;
char *provider;
} lrmd_rsc_info_t;
typedef struct lrmd_op_info_s {
char *rsc_id;
char *action;
char *interval_ms_s;
char *timeout_ms_s;
} lrmd_op_info_t;
lrmd_rsc_info_t *lrmd_new_rsc_info(const char *rsc_id, const char *standard,
const char *provider, const char *type);
lrmd_rsc_info_t *lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info);
void lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info);
void lrmd_free_op_info(lrmd_op_info_t *op_info);
typedef void (*lrmd_event_callback) (lrmd_event_data_t * event);
typedef struct lrmd_list_s {
const char *val;
struct lrmd_list_s *next;
} lrmd_list_t;
void lrmd_list_freeall(lrmd_list_t * head);
void lrmd_key_value_freeall(lrmd_key_value_t * head);
typedef struct lrmd_api_operations_s {
/*!
* \brief Connect to an executor
*
* \retval 0, success
* \retval negative error code on failure
*/
int (*connect) (lrmd_t * lrmd, const char *client_name, int *fd);
/*!
* \brief Initiate an executor connection without blocking
* \note this function requires the use of mainloop.
*
* \note The is returned using the event callback.
* \note When this function returns 0, the callback will be invoked
* to report the final result of the connect.
* \retval 0, connect in progress, wait for event callback
* \retval -1, failure.
*/
int (*connect_async) (lrmd_t * lrmd, const char *client_name, int timeout /*ms */ );
/*!
* \brief Is connected to lrmd daemon?
*
* \retval 0, false
* \retval 1, true
*/
int (*is_connected) (lrmd_t * lrmd);
/*!
* \brief Poke executor connection to verify it is still capable of serving requests
* \note The response comes in the form of a poke event to the callback.
*
* \retval 0, wait for response in callback
* \retval -1, connection failure, callback may not be invoked
*/
int (*poke_connection) (lrmd_t * lrmd);
/*!
* \brief Disconnect from the executor.
*
* \retval 0, success
* \retval negative error code on failure
*/
int (*disconnect) (lrmd_t * lrmd);
/*!
* \brief Register a resource with the executor.
*
* \note Synchronous, guaranteed to occur in daemon before function returns.
*
* \retval 0, success
* \retval negative error code on failure
*/
int (*register_rsc) (lrmd_t * lrmd,
const char *rsc_id,
const char *standard,
const char *provider, const char *agent, enum lrmd_call_options options);
/*!
* \brief Retrieve registration info for a rsc
*
* \retval info on success
* \retval NULL on failure
*/
lrmd_rsc_info_t *(*get_rsc_info) (lrmd_t * lrmd,
const char *rsc_id, enum lrmd_call_options options);
/*!
* \brief Retrieve registered recurring operations
*
* \return pcmk_ok on success, -errno otherwise
*/
int (*get_recurring_ops) (lrmd_t *lrmd, const char *rsc_id, int timeout_ms,
enum lrmd_call_options options, GList **output);
/*!
* \brief Unregister a resource from the executor.
*
* \note All pending and recurring operations will be cancelled
* automatically.
*
* \note Synchronous, guaranteed to occur in daemon before function returns.
*
* \retval 0, success
* \retval -1, success, but operations are currently executing on the rsc which will
* return once they are completed.
* \retval negative error code on failure
*
*/
int (*unregister_rsc) (lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options);
/*!
* \brief Set a callback for executor events
*/
void (*set_callback) (lrmd_t * lrmd, lrmd_event_callback callback);
/*!
* \brief Issue a command on a resource
*
* \note Asynchronous, command is queued in daemon on function return, but
* execution of command is not synced.
*
* \note Operations on individual resources are guaranteed to occur
* in the order the client api calls them in.
*
* \note Operations between different resources are not guaranteed
* to occur in any specific order in relation to one another
* regardless of what order the client api is called in.
* \retval call_id to track async event result on success
* \retval negative error code on failure
*/
int (*exec) (lrmd_t * lrmd, const char *rsc_id, const char *action, const char *userdata, /* userdata string given back in event notification */
guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
enum lrmd_call_options options, lrmd_key_value_t * params); /* ownership of params is given up to api here */
/*!
* \brief Cancel a recurring command.
*
* \note Synchronous, guaranteed to occur in daemon before function returns.
*
* \note The cancel is completed async from this call.
* We can be guaranteed the cancel has completed once
* the callback receives an exec_complete event with
* the lrmd_op_status signifying that the operation is
* cancelled.
* \note For each resource, cancel operations and exec operations
* are processed in the order they are received.
* It is safe to assume that for a single resource, a cancel
* will occur in the executor before an exec if the client's cancel
* api call occurs before the exec api call.
*
* It is not however safe to assume any operation on one resource will
* occur before an operation on another resource regardless of
* the order the client api is called in.
*
* \retval 0, cancel command sent.
* \retval negative error code on failure
*/
int (*cancel) (lrmd_t *lrmd, const char *rsc_id, const char *action,
guint interval_ms);
/*!
* \brief Get resource metadata for a specified resource agent
*
* \param[in] lrmd Executor connection (unused)
* \param[in] standard Resource agent class
* \param[in] provider Resource agent provider
* \param[in] agent Resource agent type
* \param[out] output Metadata will be stored here (must not be NULL)
* \param[in] options Options to use with any executor API calls (unused)
*
* \note Caller is responsible for freeing output. This call is currently
* always synchronous (blocking), and always done directly by the
* library (not via the executor connection). This means that it is based
* on the local host environment, even if the executor connection is to a
* remote node, so (for most resource agent classes) this will fail if
* the agent is not installed locally. This also means that, if an
* external agent must be executed, it will be executed by the
* caller's user, not the executor's.
* \todo Add a metadata call to the executor API and let the server handle this.
*
* \retval lrmd_ok success
* \retval negative error code on failure
*/
int (*get_metadata) (lrmd_t * lrmd,
const char *standard,
const char *provider,
const char *agent, char **output, enum lrmd_call_options options);
/*!
* \brief Retrieve a list of installed resource agents.
*
* \note if standard is not provided, all known agents will be returned
* \note list must be freed using lrmd_list_freeall()
*
* \retval num items in list on success
* \retval negative error code on failure
*/
int (*list_agents) (lrmd_t * lrmd, lrmd_list_t ** agents,
const char *standard, const char *provider);
/*!
* \brief Retrieve a list of resource agent providers
*
* \note When the agent is provided, only the agent's provider will be returned
* \note When no agent is supplied, all providers will be returned.
* \note List must be freed using lrmd_list_freeall()
*
* \retval num items in list on success
* \retval negative error code on failure
*/
int (*list_ocf_providers) (lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers);
/*!
* \brief Retrieve a list of standards supported by this machine/installation
*
* \note List must be freed using lrmd_list_freeall()
*
* \retval num items in list on success
* \retval negative error code on failure
*/
int (*list_standards) (lrmd_t * lrmd, lrmd_list_t ** standards);
/*!
* \brief Execute an alert agent
*
* \note Asynchronous, command is queued in daemon on function return, but
* execution of command is not synced.
*
* \note Operations on individual alerts are guaranteed to occur
* in the order the client api calls them in.
*
* \note Operations between different alerts are not guaranteed
* to occur in any specific order in relation to one another
* regardless of what order the client api is called in.
* \retval call_id to track async event result on success
* \retval negative error code on failure
*/
int (*exec_alert) (lrmd_t *lrmd, const char *alert_id,
const char *alert_path, int timeout, /* ms */
lrmd_key_value_t *params); /* ownership of params is given up to api here */
/*!
* \brief Get resource metadata for a resource agent, passing parameters
*
* \param[in] lrmd Executor connection (unused)
* \param[in] standard Resource agent class
* \param[in] provider Resource agent provider
* \param[in] agent Resource agent type
* \param[out] output Metadata will be stored here (must not be NULL)
* \param[in] options Options to use with any executor API calls (unused)
* \param[in] params Parameters to pass to agent via environment
*
* \note This is identical to the get_metadata() API call, except parameters
* will be passed to the resource agent via environment variables.
* \note The API will handle freeing params.
*
* \return lrmd_ok on success, negative error code on failure
*/
int (*get_metadata_params) (lrmd_t *lrmd, const char *standard,
const char *provider, const char *agent,
char **output, enum lrmd_call_options options,
lrmd_key_value_t *params);
} lrmd_api_operations_t;
struct lrmd_s {
lrmd_api_operations_t *cmds;
void *lrmd_private;
};
static inline const char *
lrmd_event_type2str(enum lrmd_callback_event type)
{
switch (type) {
case lrmd_event_register:
return "register";
case lrmd_event_unregister:
return "unregister";
case lrmd_event_exec_complete:
return "exec_complete";
case lrmd_event_disconnect:
return "disconnect";
case lrmd_event_connect:
return "connect";
case lrmd_event_poke:
return "poke";
case lrmd_event_new_client:
return "new_client";
}
return "unknown";
}
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/msg_xml.h b/include/crm/msg_xml.h
index d56e40c637..d0cdf6cd3d 100644
--- a/include/crm/msg_xml.h
+++ b/include/crm/msg_xml.h
@@ -1,422 +1,424 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef XML_TAGS__H
# define XML_TAGS__H
#ifdef __cplusplus
extern "C" {
#endif
# ifndef F_ORIG
# define F_ORIG "src"
# endif
# ifndef F_SEQ
# define F_SEQ "seq"
# endif
# ifndef F_SUBTYPE
# define F_SUBTYPE "subt"
# endif
# ifndef F_TYPE
# define F_TYPE "t"
# endif
# ifndef F_CLIENTNAME
# define F_CLIENTNAME "cn"
# endif
# ifndef F_XML_TAGNAME
# define F_XML_TAGNAME "__name__"
# endif
# ifndef T_CRM
# define T_CRM "crmd"
# endif
# ifndef T_ATTRD
# define T_ATTRD "attrd"
# endif
# define CIB_OPTIONS_FIRST "cib-bootstrap-options"
# define F_CRM_DATA "crm_xml"
# define F_CRM_TASK "crm_task"
# define F_CRM_HOST_TO "crm_host_to"
# define F_CRM_MSG_TYPE F_SUBTYPE
# define F_CRM_SYS_TO "crm_sys_to"
# define F_CRM_SYS_FROM "crm_sys_from"
# define F_CRM_HOST_FROM F_ORIG
# define F_CRM_REFERENCE XML_ATTR_REFERENCE
# define F_CRM_VERSION XML_ATTR_VERSION
# define F_CRM_ORIGIN "origin"
# define F_CRM_USER "crm_user"
# define F_CRM_JOIN_ID "join_id"
# define F_CRM_DC_LEAVING "dc-leaving"
# define F_CRM_ELECTION_ID "election-id"
# define F_CRM_ELECTION_AGE_S "election-age-sec"
# define F_CRM_ELECTION_AGE_US "election-age-nano-sec"
# define F_CRM_ELECTION_OWNER "election-owner"
# define F_CRM_TGRAPH "crm-tgraph-file"
# define F_CRM_TGRAPH_INPUT "crm-tgraph-in"
# define F_CRM_THROTTLE_MODE "crm-limit-mode"
# define F_CRM_THROTTLE_MAX "crm-limit-max"
/*---- Common tags/attrs */
# define XML_DIFF_MARKER "__crm_diff_marker__"
# define XML_TAG_CIB "cib"
# define XML_TAG_FAILED "failed"
# define XML_ATTR_CRM_VERSION "crm_feature_set"
# define XML_ATTR_DIGEST "digest"
# define XML_ATTR_VALIDATION "validate-with"
# define XML_ATTR_RA_VERSION "ra-version"
# define XML_ATTR_QUORUM_PANIC "no-quorum-panic"
# define XML_ATTR_HAVE_QUORUM "have-quorum"
# define XML_ATTR_HAVE_WATCHDOG "have-watchdog"
# define XML_ATTR_GENERATION "epoch"
# define XML_ATTR_GENERATION_ADMIN "admin_epoch"
# define XML_ATTR_NUMUPDATES "num_updates"
# define XML_ATTR_TIMEOUT "timeout"
# define XML_ATTR_ORIGIN "crm-debug-origin"
# define XML_ATTR_TSTAMP "crm-timestamp"
# define XML_CIB_ATTR_WRITTEN "cib-last-written"
# define XML_ATTR_VERSION "version"
# define XML_ATTR_DESC "description"
# define XML_ATTR_ID "id"
# define XML_ATTR_IDREF "id-ref"
# define XML_ATTR_ID_LONG "long-id"
# define XML_ATTR_TYPE "type"
# define XML_ATTR_VERBOSE "verbose"
# define XML_ATTR_OP "op"
# define XML_ATTR_DC_UUID "dc-uuid"
# define XML_ATTR_UPDATE_ORIG "update-origin"
# define XML_ATTR_UPDATE_CLIENT "update-client"
# define XML_ATTR_UPDATE_USER "update-user"
# define XML_BOOLEAN_TRUE "true"
# define XML_BOOLEAN_FALSE "false"
# define XML_BOOLEAN_YES XML_BOOLEAN_TRUE
# define XML_BOOLEAN_NO XML_BOOLEAN_FALSE
# define XML_TAG_OPTIONS "options"
/*---- top level tags/attrs */
# define XML_ATTR_REQUEST "request"
# define XML_ATTR_RESPONSE "response"
# define XML_ATTR_UNAME "uname"
# define XML_ATTR_UUID "id"
# define XML_ATTR_REFERENCE "reference"
# define XML_CRM_TAG_PING "ping_response"
# define XML_PING_ATTR_STATUS "result"
# define XML_PING_ATTR_SYSFROM "crm_subsystem"
# define XML_PING_ATTR_CRMDSTATE "crmd_state"
# define XML_TAG_FRAGMENT "cib_fragment"
# define XML_FAIL_TAG_CIB "failed_update"
# define XML_FAILCIB_ATTR_ID "id"
# define XML_FAILCIB_ATTR_OBJTYPE "object_type"
# define XML_FAILCIB_ATTR_OP "operation"
# define XML_FAILCIB_ATTR_REASON "reason"
/*---- CIB specific tags/attrs */
# define XML_CIB_TAG_SECTION_ALL "all"
# define XML_CIB_TAG_CONFIGURATION "configuration"
# define XML_CIB_TAG_STATUS "status"
# define XML_CIB_TAG_RESOURCES "resources"
# define XML_CIB_TAG_NODES "nodes"
# define XML_CIB_TAG_DOMAINS "domains"
# define XML_CIB_TAG_CONSTRAINTS "constraints"
# define XML_CIB_TAG_CRMCONFIG "crm_config"
# define XML_CIB_TAG_OPCONFIG "op_defaults"
# define XML_CIB_TAG_RSCCONFIG "rsc_defaults"
# define XML_CIB_TAG_ACLS "acls"
# define XML_CIB_TAG_ALERTS "alerts"
# define XML_CIB_TAG_ALERT "alert"
# define XML_CIB_TAG_ALERT_RECIPIENT "recipient"
# define XML_CIB_TAG_ALERT_SELECT "select"
# define XML_CIB_TAG_ALERT_ATTRIBUTES "select_attributes"
# define XML_CIB_TAG_ALERT_FENCING "select_fencing"
# define XML_CIB_TAG_ALERT_NODES "select_nodes"
# define XML_CIB_TAG_ALERT_RESOURCES "select_resources"
# define XML_CIB_TAG_ALERT_ATTR "attribute"
# define XML_CIB_TAG_STATE "node_state"
# define XML_CIB_TAG_NODE "node"
# define XML_CIB_TAG_NVPAIR "nvpair"
# define XML_CIB_TAG_PROPSET "cluster_property_set"
# define XML_TAG_ATTR_SETS "instance_attributes"
# define XML_TAG_META_SETS "meta_attributes"
# define XML_TAG_ATTRS "attributes"
# define XML_TAG_RSC_VER_ATTRS "rsc_versioned_attrs"
# define XML_TAG_OP_VER_ATTRS "op_versioned_attrs"
# define XML_TAG_OP_VER_META "op_versioned_meta"
# define XML_TAG_PARAMS "parameters"
# define XML_TAG_PARAM "param"
# define XML_TAG_UTILIZATION "utilization"
# define XML_TAG_RESOURCE_REF "resource_ref"
# define XML_CIB_TAG_RESOURCE "primitive"
# define XML_CIB_TAG_GROUP "group"
# define XML_CIB_TAG_INCARNATION "clone"
# define XML_CIB_TAG_MASTER "master" // deprecated since 2.0.0
# define XML_CIB_TAG_CONTAINER "bundle"
# define XML_CIB_TAG_RSC_TEMPLATE "template"
# define XML_RSC_ATTR_TARGET "container-attribute-target"
# define XML_RSC_ATTR_RESTART "restart-type"
# define XML_RSC_ATTR_ORDERED "ordered"
# define XML_RSC_ATTR_INTERLEAVE "interleave"
# define XML_RSC_ATTR_INCARNATION "clone"
# define XML_RSC_ATTR_INCARNATION_MAX "clone-max"
# define XML_RSC_ATTR_INCARNATION_MIN "clone-min"
# define XML_RSC_ATTR_INCARNATION_NODEMAX "clone-node-max"
# define XML_RSC_ATTR_PROMOTABLE "promotable"
# define XML_RSC_ATTR_PROMOTED_MAX "promoted-max"
# define XML_RSC_ATTR_PROMOTED_NODEMAX "promoted-node-max"
# define XML_RSC_ATTR_MASTER_MAX "master-max" // deprecated since 2.0.0
# define XML_RSC_ATTR_MASTER_NODEMAX "master-node-max" // deprecated since 2.0.0
# define XML_RSC_ATTR_MANAGED "is-managed"
# define XML_RSC_ATTR_TARGET_ROLE "target-role"
# define XML_RSC_ATTR_UNIQUE "globally-unique"
# define XML_RSC_ATTR_NOTIFY "notify"
# define XML_RSC_ATTR_STICKINESS "resource-stickiness"
# define XML_RSC_ATTR_FAIL_STICKINESS "migration-threshold"
# define XML_RSC_ATTR_FAIL_TIMEOUT "failure-timeout"
# define XML_RSC_ATTR_MULTIPLE "multiple-active"
# define XML_RSC_ATTR_REQUIRES "requires"
# define XML_RSC_ATTR_PROVIDES "provides"
# define XML_RSC_ATTR_CONTAINER "container"
# define XML_RSC_ATTR_INTERNAL_RSC "internal_rsc"
# define XML_RSC_ATTR_MAINTENANCE "maintenance"
# define XML_RSC_ATTR_REMOTE_NODE "remote-node"
# define XML_RSC_ATTR_CLEAR_OP "clear_failure_op"
# define XML_RSC_ATTR_CLEAR_INTERVAL "clear_failure_interval"
# define XML_RSC_ATTR_REMOTE_RA_ADDR "addr"
# define XML_RSC_ATTR_REMOTE_RA_SERVER "server"
# define XML_RSC_ATTR_REMOTE_RA_PORT "port"
# define XML_REMOTE_ATTR_RECONNECT_INTERVAL "reconnect_interval"
# define XML_OP_ATTR_ON_FAIL "on-fail"
# define XML_OP_ATTR_START_DELAY "start-delay"
# define XML_OP_ATTR_ALLOW_MIGRATE "allow-migrate"
# define XML_OP_ATTR_ORIGIN "interval-origin"
# define XML_OP_ATTR_PENDING "record-pending"
# define XML_OP_ATTR_DIGESTS_ALL "digests-all"
# define XML_OP_ATTR_DIGESTS_SECURE "digests-secure"
# define XML_CIB_TAG_LRM "lrm"
# define XML_LRM_TAG_RESOURCES "lrm_resources"
# define XML_LRM_TAG_RESOURCE "lrm_resource"
# define XML_LRM_TAG_RSC_OP "lrm_rsc_op"
# define XML_AGENT_ATTR_CLASS "class"
# define XML_AGENT_ATTR_PROVIDER "provider"
# define XML_CIB_ATTR_REPLACE "replace"
# define XML_CIB_ATTR_SOURCE "source"
# define XML_CIB_ATTR_PRIORITY "priority"
# define XML_CIB_ATTR_SOURCE "source"
# define XML_NODE_JOIN_STATE "join"
# define XML_NODE_EXPECTED "expected"
# define XML_NODE_IN_CLUSTER "in_ccm"
# define XML_NODE_IS_PEER "crmd"
# define XML_NODE_IS_REMOTE "remote_node"
# define XML_NODE_IS_FENCED "node_fenced"
# define XML_NODE_IS_MAINTENANCE "node_in_maintenance"
# define XML_CIB_ATTR_SHUTDOWN "shutdown"
/* Aside from being an old name for the executor, LRM is a misnomer here because
* the controller and scheduler use these to track actions, which are not always
* executor operations.
*/
// XML attribute that takes interval specification (user-facing configuration)
# define XML_LRM_ATTR_INTERVAL "interval"
// XML attribute that takes interval in milliseconds (daemon APIs)
// (identical value as above, but different constant allows clearer code intent)
# define XML_LRM_ATTR_INTERVAL_MS XML_LRM_ATTR_INTERVAL
# define XML_LRM_ATTR_TASK "operation"
# define XML_LRM_ATTR_TASK_KEY "operation_key"
# define XML_LRM_ATTR_TARGET "on_node"
# define XML_LRM_ATTR_TARGET_UUID "on_node_uuid"
/*! Actions to be executed on Pacemaker Remote nodes are routed through the
* controller on the cluster node hosting the remote connection. That cluster
* node is considered the router node for the action.
*/
# define XML_LRM_ATTR_ROUTER_NODE "router_node"
# define XML_LRM_ATTR_RSCID "rsc-id"
# define XML_LRM_ATTR_OPSTATUS "op-status"
# define XML_LRM_ATTR_RC "rc-code"
# define XML_LRM_ATTR_CALLID "call-id"
# define XML_LRM_ATTR_OP_DIGEST "op-digest"
# define XML_LRM_ATTR_OP_RESTART "op-force-restart"
# define XML_LRM_ATTR_OP_SECURE "op-secure-params"
# define XML_LRM_ATTR_RESTART_DIGEST "op-restart-digest"
# define XML_LRM_ATTR_SECURE_DIGEST "op-secure-digest"
# define XML_LRM_ATTR_EXIT_REASON "exit-reason"
# define XML_RSC_OP_LAST_CHANGE "last-rc-change"
# define XML_RSC_OP_LAST_RUN "last-run" // deprecated since 2.0.3
# define XML_RSC_OP_T_EXEC "exec-time"
# define XML_RSC_OP_T_QUEUE "queue-time"
# define XML_LRM_ATTR_MIGRATE_SOURCE "migrate_source"
# define XML_LRM_ATTR_MIGRATE_TARGET "migrate_target"
# define XML_TAG_GRAPH "transition_graph"
# define XML_GRAPH_TAG_RSC_OP "rsc_op"
# define XML_GRAPH_TAG_PSEUDO_EVENT "pseudo_event"
# define XML_GRAPH_TAG_CRM_EVENT "crm_event"
# define XML_GRAPH_TAG_DOWNED "downed"
# define XML_GRAPH_TAG_MAINTENANCE "maintenance"
# define XML_TAG_RULE "rule"
# define XML_RULE_ATTR_SCORE "score"
# define XML_RULE_ATTR_SCORE_ATTRIBUTE "score-attribute"
# define XML_RULE_ATTR_ROLE "role"
# define XML_RULE_ATTR_BOOLEAN_OP "boolean-op"
# define XML_TAG_EXPRESSION "expression"
# define XML_EXPR_ATTR_ATTRIBUTE "attribute"
# define XML_EXPR_ATTR_OPERATION "operation"
# define XML_EXPR_ATTR_VALUE "value"
# define XML_EXPR_ATTR_TYPE "type"
# define XML_EXPR_ATTR_VALUE_SOURCE "value-source"
# define XML_CONS_TAG_RSC_DEPEND "rsc_colocation"
# define XML_CONS_TAG_RSC_ORDER "rsc_order"
# define XML_CONS_TAG_RSC_LOCATION "rsc_location"
# define XML_CONS_TAG_RSC_TICKET "rsc_ticket"
# define XML_CONS_TAG_RSC_SET "resource_set"
# define XML_CONS_ATTR_SYMMETRICAL "symmetrical"
# define XML_LOCATION_ATTR_DISCOVERY "resource-discovery"
# define XML_COLOC_ATTR_SOURCE "rsc"
# define XML_COLOC_ATTR_SOURCE_ROLE "rsc-role"
# define XML_COLOC_ATTR_TARGET "with-rsc"
# define XML_COLOC_ATTR_TARGET_ROLE "with-rsc-role"
# define XML_COLOC_ATTR_NODE_ATTR "node-attribute"
# define XML_COLOC_ATTR_SOURCE_INSTANCE "rsc-instance"
# define XML_COLOC_ATTR_TARGET_INSTANCE "with-rsc-instance"
# define XML_LOC_ATTR_SOURCE "rsc"
# define XML_LOC_ATTR_SOURCE_PATTERN "rsc-pattern"
# define XML_ORDER_ATTR_FIRST "first"
# define XML_ORDER_ATTR_THEN "then"
# define XML_ORDER_ATTR_FIRST_ACTION "first-action"
# define XML_ORDER_ATTR_THEN_ACTION "then-action"
# define XML_ORDER_ATTR_FIRST_INSTANCE "first-instance"
# define XML_ORDER_ATTR_THEN_INSTANCE "then-instance"
# define XML_ORDER_ATTR_KIND "kind"
# define XML_TICKET_ATTR_TICKET "ticket"
# define XML_TICKET_ATTR_LOSS_POLICY "loss-policy"
# define XML_NVPAIR_ATTR_NAME "name"
# define XML_NVPAIR_ATTR_VALUE "value"
# define XML_NODE_ATTR_RSC_DISCOVERY "resource-discovery-enabled"
# define XML_CONFIG_ATTR_DC_DEADTIME "dc-deadtime"
# define XML_CONFIG_ATTR_ELECTION_FAIL "election-timeout"
# define XML_CONFIG_ATTR_FORCE_QUIT "shutdown-escalation"
# define XML_CONFIG_ATTR_RECHECK "cluster-recheck-interval"
# define XML_CONFIG_ATTR_FENCE_REACTION "fence-reaction"
+# define XML_CONFIG_ATTR_SHUTDOWN_LOCK "shutdown-lock"
+# define XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT "shutdown-lock-limit"
# define XML_ALERT_ATTR_PATH "path"
# define XML_ALERT_ATTR_TIMEOUT "timeout"
# define XML_ALERT_ATTR_TSTAMP_FORMAT "timestamp-format"
# define XML_ALERT_ATTR_REC_VALUE "value"
# define XML_CIB_TAG_GENERATION_TUPPLE "generation_tuple"
# define XML_ATTR_TRANSITION_MAGIC "transition-magic"
# define XML_ATTR_TRANSITION_KEY "transition-key"
# define XML_ATTR_TE_NOWAIT "op_no_wait"
# define XML_ATTR_TE_TARGET_RC "op_target_rc"
# define XML_TAG_TRANSIENT_NODEATTRS "transient_attributes"
# define XML_TAG_DIFF_ADDED "diff-added"
# define XML_TAG_DIFF_REMOVED "diff-removed"
# define XML_ACL_TAG_USER "acl_target"
# define XML_ACL_TAG_USERv1 "acl_user"
# define XML_ACL_TAG_GROUP "acl_group"
# define XML_ACL_TAG_ROLE "acl_role"
# define XML_ACL_TAG_PERMISSION "acl_permission"
# define XML_ACL_TAG_ROLE_REF "role"
# define XML_ACL_TAG_ROLE_REFv1 "role_ref"
# define XML_ACL_ATTR_KIND "kind"
# define XML_ACL_TAG_READ "read"
# define XML_ACL_TAG_WRITE "write"
# define XML_ACL_TAG_DENY "deny"
# define XML_ACL_ATTR_REF "reference"
# define XML_ACL_ATTR_REFv1 "ref"
# define XML_ACL_ATTR_TAG "object-type"
# define XML_ACL_ATTR_TAGv1 "tag"
# define XML_ACL_ATTR_XPATH "xpath"
# define XML_ACL_ATTR_ATTRIBUTE "attribute"
# define XML_CIB_TAG_TICKETS "tickets"
# define XML_CIB_TAG_TICKET_STATE "ticket_state"
# define XML_CIB_TAG_TAGS "tags"
# define XML_CIB_TAG_TAG "tag"
# define XML_CIB_TAG_OBJ_REF "obj_ref"
# define XML_TAG_FENCING_TOPOLOGY "fencing-topology"
# define XML_TAG_FENCING_LEVEL "fencing-level"
# define XML_ATTR_STONITH_INDEX "index"
# define XML_ATTR_STONITH_TARGET "target"
# define XML_ATTR_STONITH_TARGET_VALUE "target-value"
# define XML_ATTR_STONITH_TARGET_PATTERN "target-pattern"
# define XML_ATTR_STONITH_TARGET_ATTRIBUTE "target-attribute"
# define XML_ATTR_STONITH_DEVICES "devices"
# define XML_TAG_DIFF "diff"
# define XML_DIFF_VERSION "version"
# define XML_DIFF_VSOURCE "source"
# define XML_DIFF_VTARGET "target"
# define XML_DIFF_CHANGE "change"
# define XML_DIFF_LIST "change-list"
# define XML_DIFF_ATTR "change-attr"
# define XML_DIFF_RESULT "change-result"
# define XML_DIFF_OP "operation"
# define XML_DIFF_PATH "path"
# define XML_DIFF_POSITION "position"
# include
# define ID(x) crm_element_value(x, XML_ATTR_ID)
# define TYPE(x) crm_element_name(x)
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/pengine/internal.h b/include/crm/pengine/internal.h
index 119624d17d..bc2c70e8b7 100644
--- a/include/crm/pengine/internal.h
+++ b/include/crm/pengine/internal.h
@@ -1,439 +1,441 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PE_INTERNAL__H
# define PE_INTERNAL__H
# include
# include
# include
# include
# define pe_rsc_info(rsc, fmt, args...) crm_log_tag(LOG_INFO, rsc ? rsc->id : "", fmt, ##args)
# define pe_rsc_debug(rsc, fmt, args...) crm_log_tag(LOG_DEBUG, rsc ? rsc->id : "", fmt, ##args)
# define pe_rsc_trace(rsc, fmt, args...) crm_log_tag(LOG_TRACE, rsc ? rsc->id : "", fmt, ##args)
# define pe_err(fmt...) { was_processing_error = TRUE; crm_config_error = TRUE; crm_err(fmt); }
# define pe_warn(fmt...) { was_processing_warning = TRUE; crm_config_warning = TRUE; crm_warn(fmt); }
# define pe_proc_err(fmt...) { was_processing_error = TRUE; crm_err(fmt); }
# define pe_proc_warn(fmt...) { was_processing_warning = TRUE; crm_warn(fmt); }
# define pe_set_action_bit(action, bit) action->flags = crm_set_bit(__FUNCTION__, __LINE__, action->uuid, action->flags, bit)
# define pe_clear_action_bit(action, bit) action->flags = crm_clear_bit(__FUNCTION__, __LINE__, action->uuid, action->flags, bit)
typedef struct pe__location_constraint_s {
char *id; // Constraint XML ID
pe_resource_t *rsc_lh; // Resource being located
enum rsc_role_e role_filter; // Role to locate
enum pe_discover_e discover_mode; // Resource discovery
GListPtr node_list_rh; // List of pe_node_t*
} pe__location_t;
typedef struct pe__order_constraint_s {
int id;
enum pe_ordering type;
void *lh_opaque;
resource_t *lh_rsc;
action_t *lh_action;
char *lh_action_task;
void *rh_opaque;
resource_t *rh_rsc;
action_t *rh_action;
char *rh_action_task;
} pe__ordering_t;
typedef struct notify_data_s {
GSList *keys; // Environment variable name/value pairs
const char *action;
action_t *pre;
action_t *post;
action_t *pre_done;
action_t *post_done;
GListPtr active; /* notify_entry_t* */
GListPtr inactive; /* notify_entry_t* */
GListPtr start; /* notify_entry_t* */
GListPtr stop; /* notify_entry_t* */
GListPtr demote; /* notify_entry_t* */
GListPtr promote; /* notify_entry_t* */
GListPtr master; /* notify_entry_t* */
GListPtr slave; /* notify_entry_t* */
GHashTable *allowed_nodes;
} notify_data_t;
bool pe_can_fence(pe_working_set_t *data_set, node_t *node);
int merge_weights(int w1, int w2);
void add_hash_param(GHashTable * hash, const char *name, const char *value);
char *native_parameter(resource_t * rsc, node_t * node, gboolean create, const char *name,
pe_working_set_t * data_set);
pe_node_t *native_location(const pe_resource_t *rsc, GList **list, int current);
void pe_metadata(void);
void verify_pe_options(GHashTable * options);
void common_update_score(resource_t * rsc, const char *id, int score);
void native_add_running(resource_t * rsc, node_t * node, pe_working_set_t * data_set);
gboolean native_unpack(resource_t * rsc, pe_working_set_t * data_set);
gboolean group_unpack(resource_t * rsc, pe_working_set_t * data_set);
gboolean clone_unpack(resource_t * rsc, pe_working_set_t * data_set);
gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set);
resource_t *native_find_rsc(resource_t *rsc, const char *id, const node_t *node,
int flags);
gboolean native_active(resource_t * rsc, gboolean all);
gboolean group_active(resource_t * rsc, gboolean all);
gboolean clone_active(resource_t * rsc, gboolean all);
gboolean pe__bundle_active(pe_resource_t *rsc, gboolean all);
void native_print(resource_t * rsc, const char *pre_text, long options, void *print_data);
void group_print(resource_t * rsc, const char *pre_text, long options, void *print_data);
void clone_print(resource_t * rsc, const char *pre_text, long options, void *print_data);
void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options,
void *print_data);
int pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name
, size_t pairs_count, ...);
char *pe__node_display_name(node_t *node, bool print_detail);
int pe__ban_html(pcmk__output_t *out, va_list args);
int pe__ban_text(pcmk__output_t *out, va_list args);
int pe__ban_xml(pcmk__output_t *out, va_list args);
int pe__clone_xml(pcmk__output_t *out, va_list args);
int pe__clone_html(pcmk__output_t *out, va_list args);
int pe__clone_text(pcmk__output_t *out, va_list args);
int pe__cluster_counts_html(pcmk__output_t *out, va_list args);
int pe__cluster_counts_text(pcmk__output_t *out, va_list args);
int pe__cluster_counts_xml(pcmk__output_t *out, va_list args);
int pe__cluster_dc_html(pcmk__output_t *out, va_list args);
int pe__cluster_dc_text(pcmk__output_t *out, va_list args);
int pe__cluster_dc_xml(pcmk__output_t *out, va_list args);
int pe__cluster_options_html(pcmk__output_t *out, va_list args);
int pe__cluster_options_log(pcmk__output_t *out, va_list args);
int pe__cluster_options_text(pcmk__output_t *out, va_list args);
int pe__cluster_options_xml(pcmk__output_t *out, va_list args);
int pe__cluster_stack_html(pcmk__output_t *out, va_list args);
int pe__cluster_stack_text(pcmk__output_t *out, va_list args);
int pe__cluster_stack_xml(pcmk__output_t *out, va_list args);
int pe__cluster_times_html(pcmk__output_t *out, va_list args);
int pe__cluster_times_xml(pcmk__output_t *out, va_list args);
int pe__cluster_times_text(pcmk__output_t *out, va_list args);
int pe__failed_action_text(pcmk__output_t *out, va_list args);
int pe__failed_action_xml(pcmk__output_t *out, va_list args);
int pe__group_xml(pcmk__output_t *out, va_list args);
int pe__group_html(pcmk__output_t *out, va_list args);
int pe__group_text(pcmk__output_t *out, va_list args);
int pe__bundle_xml(pcmk__output_t *out, va_list args);
int pe__bundle_html(pcmk__output_t *out, va_list args);
int pe__bundle_text(pcmk__output_t *out, va_list args);
int pe__node_html(pcmk__output_t *out, va_list args);
int pe__node_text(pcmk__output_t *out, va_list args);
int pe__node_xml(pcmk__output_t *out, va_list args);
int pe__node_attribute_html(pcmk__output_t *out, va_list args);
int pe__node_attribute_text(pcmk__output_t *out, va_list args);
int pe__node_attribute_xml(pcmk__output_t *out, va_list args);
int pe__op_history_text(pcmk__output_t *out, va_list args);
int pe__op_history_xml(pcmk__output_t *out, va_list args);
int pe__resource_header_text(pcmk__output_t *out, va_list args);
int pe__resource_history_text(pcmk__output_t *out, va_list args);
int pe__resource_history_xml(pcmk__output_t *out, va_list args);
int pe__resource_xml(pcmk__output_t *out, va_list args);
int pe__resource_html(pcmk__output_t *out, va_list args);
int pe__resource_text(pcmk__output_t *out, va_list args);
int pe__ticket_html(pcmk__output_t *out, va_list args);
int pe__ticket_text(pcmk__output_t *out, va_list args);
int pe__ticket_xml(pcmk__output_t *out, va_list args);
void native_free(resource_t * rsc);
void group_free(resource_t * rsc);
void clone_free(resource_t * rsc);
void pe__free_bundle(pe_resource_t *rsc);
enum rsc_role_e native_resource_state(const resource_t * rsc, gboolean current);
enum rsc_role_e group_resource_state(const resource_t * rsc, gboolean current);
enum rsc_role_e clone_resource_state(const resource_t * rsc, gboolean current);
enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc,
gboolean current);
void pe__count_common(pe_resource_t *rsc);
void pe__count_bundle(pe_resource_t *rsc);
gboolean common_unpack(xmlNode * xml_obj, resource_t ** rsc, resource_t * parent,
pe_working_set_t * data_set);
void common_free(resource_t * rsc);
extern node_t *node_copy(const node_t *this_node);
extern time_t get_effective_time(pe_working_set_t * data_set);
/* Failure handling utilities (from failcounts.c) */
// bit flags for fail count handling options
enum pe_fc_flags_e {
pe_fc_default = 0x00,
pe_fc_effective = 0x01, // don't count expired failures
pe_fc_fillers = 0x02, // if container, include filler failures in count
};
int pe_get_failcount(node_t *node, resource_t *rsc, time_t *last_failure,
uint32_t flags, xmlNode *xml_op,
pe_working_set_t *data_set);
pe_action_t *pe__clear_failcount(pe_resource_t *rsc, pe_node_t *node,
const char *reason,
pe_working_set_t *data_set);
/* Functions for finding/counting a resource's active nodes */
pe_node_t *pe__find_active_on(const pe_resource_t *rsc,
unsigned int *count_all,
unsigned int *count_clean);
pe_node_t *pe__find_active_requires(const pe_resource_t *rsc,
unsigned int *count);
static inline pe_node_t *
pe__current_node(const pe_resource_t *rsc)
{
return pe__find_active_on(rsc, NULL, NULL);
}
/* Binary like operators for lists of nodes */
extern void node_list_exclude(GHashTable * list, GListPtr list2, gboolean merge_scores);
extern GListPtr node_list_dup(GListPtr list, gboolean reset, gboolean filter);
extern GHashTable *node_hash_from_list(GListPtr list);
static inline gpointer
pe_hash_table_lookup(GHashTable * hash, gconstpointer key)
{
if (hash) {
return g_hash_table_lookup(hash, key);
}
return NULL;
}
extern action_t *get_pseudo_op(const char *name, pe_working_set_t * data_set);
extern gboolean order_actions(action_t * lh_action, action_t * rh_action, enum pe_ordering order);
GHashTable *node_hash_dup(GHashTable * hash);
/* Printing functions for debug */
extern void print_node(const char *pre_text, node_t * node, gboolean details);
extern void print_str_str(gpointer key, gpointer value, gpointer user_data);
extern void pe__output_node(node_t * node, gboolean details, pcmk__output_t *out);
extern void dump_node_scores_worker(int level, const char *file, const char *function, int line,
resource_t * rsc, const char *comment, GHashTable * nodes);
extern void dump_node_capacity(int level, const char *comment, node_t * node);
extern void dump_rsc_utilization(int level, const char *comment, resource_t * rsc, node_t * node);
# define dump_node_scores(level, rsc, text, nodes) do { \
dump_node_scores_worker(level, __FILE__, __FUNCTION__, __LINE__, rsc, text, nodes); \
} while(0)
/* Sorting functions */
extern gint sort_rsc_priority(gconstpointer a, gconstpointer b);
extern gint sort_rsc_index(gconstpointer a, gconstpointer b);
extern xmlNode *find_rsc_op_entry(resource_t * rsc, const char *key);
extern action_t *custom_action(resource_t * rsc, char *key, const char *task, node_t * on_node,
gboolean optional, gboolean foo, pe_working_set_t * data_set);
# define delete_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_DELETE, 0)
# define delete_action(rsc, node, optional) custom_action( \
rsc, delete_key(rsc), CRMD_ACTION_DELETE, node, \
optional, TRUE, data_set);
# define stopped_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_STOPPED, 0)
# define stopped_action(rsc, node, optional) custom_action( \
rsc, stopped_key(rsc), CRMD_ACTION_STOPPED, node, \
optional, TRUE, data_set);
# define stop_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_STOP, 0)
# define stop_action(rsc, node, optional) custom_action( \
rsc, stop_key(rsc), CRMD_ACTION_STOP, node, \
optional, TRUE, data_set);
# define reload_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_RELOAD, 0)
# define start_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_START, 0)
# define start_action(rsc, node, optional) custom_action( \
rsc, start_key(rsc), CRMD_ACTION_START, node, \
optional, TRUE, data_set)
# define started_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_STARTED, 0)
# define started_action(rsc, node, optional) custom_action( \
rsc, started_key(rsc), CRMD_ACTION_STARTED, node, \
optional, TRUE, data_set)
# define promote_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_PROMOTE, 0)
# define promote_action(rsc, node, optional) custom_action( \
rsc, promote_key(rsc), CRMD_ACTION_PROMOTE, node, \
optional, TRUE, data_set)
# define promoted_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_PROMOTED, 0)
# define promoted_action(rsc, node, optional) custom_action( \
rsc, promoted_key(rsc), CRMD_ACTION_PROMOTED, node, \
optional, TRUE, data_set)
# define demote_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_DEMOTE, 0)
# define demote_action(rsc, node, optional) custom_action( \
rsc, demote_key(rsc), CRMD_ACTION_DEMOTE, node, \
optional, TRUE, data_set)
# define demoted_key(rsc) generate_op_key(rsc->id, CRMD_ACTION_DEMOTED, 0)
# define demoted_action(rsc, node, optional) custom_action( \
rsc, demoted_key(rsc), CRMD_ACTION_DEMOTED, node, \
optional, TRUE, data_set)
extern int pe_get_configured_timeout(resource_t *rsc, const char *action,
pe_working_set_t *data_set);
extern action_t *find_first_action(GListPtr input, const char *uuid, const char *task,
node_t * on_node);
extern enum action_tasks get_complex_task(resource_t * rsc, const char *name,
gboolean allow_non_atomic);
extern GListPtr find_actions(GListPtr input, const char *key, const node_t *on_node);
GList *find_actions_exact(GList *input, const char *key,
const pe_node_t *on_node);
extern GListPtr find_recurring_actions(GListPtr input, node_t * not_on_node);
GList *pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node,
const char *task, bool require_node);
extern void pe_free_action(action_t * action);
extern void resource_location(resource_t * rsc, node_t * node, int score, const char *tag,
pe_working_set_t * data_set);
extern gint sort_op_by_callid(gconstpointer a, gconstpointer b);
extern gboolean get_target_role(resource_t * rsc, enum rsc_role_e *role);
extern resource_t *find_clone_instance(resource_t * rsc, const char *sub_id,
pe_working_set_t * data_set);
extern void destroy_ticket(gpointer data);
extern ticket_t *ticket_new(const char *ticket_id, pe_working_set_t * data_set);
// Resources for manipulating resource names
const char *pe_base_name_end(const char *id);
char *clone_strip(const char *last_rsc_id);
char *clone_zero(const char *last_rsc_id);
static inline bool
pe_base_name_eq(resource_t *rsc, const char *id)
{
if (id && rsc && rsc->id) {
// Number of characters in rsc->id before any clone suffix
size_t base_len = pe_base_name_end(rsc->id) - rsc->id + 1;
return (strlen(id) == base_len) && !strncmp(id, rsc->id, base_len);
}
return FALSE;
}
int pe__target_rc_from_xml(xmlNode *xml_op);
gint sort_node_uname(gconstpointer a, gconstpointer b);
bool is_set_recursive(resource_t * rsc, long long flag, bool any);
enum rsc_digest_cmp_val {
/*! Digests are the same */
RSC_DIGEST_MATCH = 0,
/*! Params that require a restart changed */
RSC_DIGEST_RESTART,
/*! Some parameter changed. */
RSC_DIGEST_ALL,
/*! rsc op didn't have a digest associated with it, so
* it is unknown if parameters changed or not. */
RSC_DIGEST_UNKNOWN,
};
typedef struct op_digest_cache_s {
enum rsc_digest_cmp_val rc;
xmlNode *params_all;
xmlNode *params_secure;
xmlNode *params_restart;
char *digest_all_calc;
char *digest_secure_calc;
char *digest_restart_calc;
} op_digest_cache_t;
op_digest_cache_t *rsc_action_digest_cmp(resource_t * rsc, xmlNode * xml_op, node_t * node,
pe_working_set_t * data_set);
action_t *pe_fence_op(node_t * node, const char *op, bool optional, const char *reason, pe_working_set_t * data_set);
void trigger_unfencing(
resource_t * rsc, node_t *node, const char *reason, action_t *dependency, pe_working_set_t * data_set);
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite);
void pe_action_set_flag_reason(const char *function, long line, pe_action_t *action, pe_action_t *reason, const char *text, enum pe_action_flags flags, bool overwrite);
#define pe_action_required(action, reason, text) pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, reason, text, pe_action_optional, FALSE)
#define pe_action_implies(action, reason, flag) pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, reason, NULL, flag, FALSE)
void set_bit_recursive(resource_t * rsc, unsigned long long flag);
void clear_bit_recursive(resource_t * rsc, unsigned long long flag);
gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref);
void print_rscs_brief(GListPtr rsc_list, const char * pre_text, long options,
void * print_data, gboolean print_all);
void pe__rscs_brief_output(pcmk__output_t *out, GListPtr rsc_list, long options, gboolean print_all);
void pe_fence_node(pe_working_set_t * data_set, node_t * node, const char *reason);
node_t *pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pe_working_set_t * data_set);
bool remote_id_conflict(const char *remote_name, pe_working_set_t *data);
void common_print(resource_t * rsc, const char *pre_text, const char *name, node_t *node, long options, void *print_data);
void pe__common_output_text(pcmk__output_t *out, resource_t * rsc, const char *name, node_t *node, long options);
void pe__common_output_html(pcmk__output_t *out, resource_t * rsc, const char *name, node_t *node, long options);
pe_resource_t *pe__find_bundle_replica(const pe_resource_t *bundle,
const pe_node_t *node);
bool pe__bundle_needs_remote_name(pe_resource_t *rsc);
const char *pe__add_bundle_remote_name(pe_resource_t *rsc, xmlNode *xml,
const char *field);
const char *pe_node_attribute_calculated(const pe_node_t *node,
const char *name,
const resource_t *rsc);
const char *pe_node_attribute_raw(pe_node_t *node, const char *name);
bool pe__is_universal_clone(pe_resource_t *rsc,
pe_working_set_t *data_set);
void pe__add_param_check(xmlNode *rsc_op, pe_resource_t *rsc, pe_node_t *node,
enum pe_check_parameters, pe_working_set_t *data_set);
void pe__foreach_param_check(pe_working_set_t *data_set,
void (*cb)(pe_resource_t*, pe_node_t*, xmlNode*,
enum pe_check_parameters,
pe_working_set_t*));
void pe__free_param_checks(pe_working_set_t *data_set);
bool pe__shutdown_requested(pe_node_t *node);
void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set);
#define BOOL2STR(x) ((x) ? "true" : "false")
/*!
* \internal
* \brief Register xml formatting message functions.
*/
void pe__register_messages(pcmk__output_t *out);
void pe__unpack_dataset_nvpairs(xmlNode *xml_obj, const char *set_name,
GHashTable *node_hash, GHashTable *hash,
const char *always_first, gboolean overwrite,
pe_working_set_t *data_set);
bool pe__resource_is_disabled(pe_resource_t *rsc);
+pe_action_t *pe__clear_resource_history(pe_resource_t *rsc, pe_node_t *node,
+ pe_working_set_t *data_set);
#endif
diff --git a/include/crm/pengine/pe_types.h b/include/crm/pengine/pe_types.h
index 23e1c46b9c..572787b71c 100644
--- a/include/crm/pengine/pe_types.h
+++ b/include/crm/pengine/pe_types.h
@@ -1,508 +1,514 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PE_TYPES__H
# define PE_TYPES__H
#ifdef __cplusplus
extern "C" {
#endif
/*!
* \file
* \brief Data types for cluster status
* \ingroup pengine
*/
# include // bool
# include // time_t
# include // gboolean, guint, GList, GHashTable
# include // GListPtr
# include
# include
typedef struct pe_node_s pe_node_t;
typedef struct pe_action_s pe_action_t;
typedef struct pe_resource_s pe_resource_t;
typedef struct pe_working_set_s pe_working_set_t;
enum pe_obj_types {
pe_unknown = -1,
pe_native = 0,
pe_group = 1,
pe_clone = 2,
pe_container = 3,
};
typedef struct resource_object_functions_s {
gboolean (*unpack) (pe_resource_t*, pe_working_set_t*);
pe_resource_t *(*find_rsc) (pe_resource_t *parent, const char *search,
const pe_node_t *node, int flags);
/* parameter result must be free'd */
char *(*parameter) (pe_resource_t*, pe_node_t*, gboolean, const char*,
pe_working_set_t*);
void (*print) (pe_resource_t*, const char*, long, void*);
gboolean (*active) (pe_resource_t*, gboolean);
enum rsc_role_e (*state) (const pe_resource_t*, gboolean);
pe_node_t *(*location) (const pe_resource_t*, GList**, int);
void (*free) (pe_resource_t*);
void (*count) (pe_resource_t*);
} resource_object_functions_t;
typedef struct resource_alloc_functions_s resource_alloc_functions_t;
enum pe_quorum_policy {
no_quorum_freeze,
no_quorum_stop,
no_quorum_ignore,
no_quorum_suicide
};
enum node_type {
node_ping,
node_member,
node_remote
};
//! \deprecated will be removed in a future release
enum pe_restart {
pe_restart_restart,
pe_restart_ignore
};
//! Determine behavior of pe_find_resource_with_flags()
enum pe_find {
pe_find_renamed = 0x001, //!< match resource ID or LRM history ID
pe_find_anon = 0x002, //!< match base name of anonymous clone instances
pe_find_clone = 0x004, //!< match only clone instances
pe_find_current = 0x008, //!< match resource active on specified node
pe_find_inactive = 0x010, //!< match resource not running anywhere
pe_find_any = 0x020, //!< match base name of any clone instance
};
# define pe_flag_have_quorum 0x00000001ULL
# define pe_flag_symmetric_cluster 0x00000002ULL
# define pe_flag_maintenance_mode 0x00000008ULL
# define pe_flag_stonith_enabled 0x00000010ULL
# define pe_flag_have_stonith_resource 0x00000020ULL
# define pe_flag_enable_unfencing 0x00000040ULL
# define pe_flag_concurrent_fencing 0x00000080ULL
# define pe_flag_stop_rsc_orphans 0x00000100ULL
# define pe_flag_stop_action_orphans 0x00000200ULL
# define pe_flag_stop_everything 0x00000400ULL
# define pe_flag_start_failure_fatal 0x00001000ULL
# define pe_flag_remove_after_stop 0x00002000ULL
# define pe_flag_startup_fencing 0x00004000ULL
+# define pe_flag_shutdown_lock 0x00008000ULL
# define pe_flag_startup_probes 0x00010000ULL
# define pe_flag_have_status 0x00020000ULL
# define pe_flag_have_remote_nodes 0x00040000ULL
# define pe_flag_quick_location 0x00100000ULL
# define pe_flag_sanitized 0x00200000ULL
# define pe_flag_stdout 0x00400000ULL
//! Don't count total, disabled and blocked resource instances
# define pe_flag_no_counts 0x00800000ULL
struct pe_working_set_s {
xmlNode *input;
crm_time_t *now;
/* options extracted from the input */
char *dc_uuid;
pe_node_t *dc_node;
const char *stonith_action;
const char *placement_strategy;
unsigned long long flags;
int stonith_timeout;
enum pe_quorum_policy no_quorum_policy;
GHashTable *config_hash;
GHashTable *tickets;
// Actions for which there can be only one (e.g. fence nodeX)
GHashTable *singletons;
GListPtr nodes;
GListPtr resources;
GListPtr placement_constraints;
GListPtr ordering_constraints;
GListPtr colocation_constraints;
GListPtr ticket_constraints;
GListPtr actions;
xmlNode *failed;
xmlNode *op_defaults;
xmlNode *rsc_defaults;
/* stats */
int num_synapse;
int max_valid_nodes;
int order_id;
int action_id;
/* final output */
xmlNode *graph;
GHashTable *template_rsc_sets;
const char *localhost;
GHashTable *tags;
int blocked_resources;
int disabled_resources;
GList *param_check; // History entries that need to be checked
GList *stop_needed; // Containers that need stop actions
time_t recheck_by; // Hint to controller to re-run scheduler by this time
int ninstances; // Total number of resource instances
+ guint shutdown_lock;// How long (seconds) to lock resources to shutdown node
};
enum pe_check_parameters {
/* Clear fail count if parameters changed for un-expired start or monitor
* last_failure.
*/
pe_check_last_failure,
/* Clear fail count if parameters changed for start, monitor, promote, or
* migrate_from actions for active resources.
*/
pe_check_active,
};
struct pe_node_shared_s {
const char *id;
const char *uname;
enum node_type type;
/* @TODO convert these flags into a bitfield */
gboolean online;
gboolean standby;
gboolean standby_onfail;
gboolean pending;
gboolean unclean;
gboolean unseen;
gboolean shutdown;
gboolean expected_up;
gboolean is_dc;
gboolean maintenance;
gboolean rsc_discovery_enabled;
gboolean remote_requires_reset;
gboolean remote_was_fenced;
gboolean remote_maintenance; /* what the remote-rsc is thinking */
gboolean unpacked;
int num_resources;
pe_resource_t *remote_rsc;
GListPtr running_rsc; /* pe_resource_t* */
GListPtr allocated_rsc; /* pe_resource_t* */
GHashTable *attrs; /* char* => char* */
GHashTable *utilization;
GHashTable *digest_cache; //!< cache of calculated resource digests
};
struct pe_node_s {
int weight;
gboolean fixed;
int count;
struct pe_node_shared_s *details;
int rsc_discover_mode;
};
# define pe_rsc_orphan 0x00000001ULL
# define pe_rsc_managed 0x00000002ULL
# define pe_rsc_block 0x00000004ULL
# define pe_rsc_orphan_container_filler 0x00000008ULL
# define pe_rsc_notify 0x00000010ULL
# define pe_rsc_unique 0x00000020ULL
# define pe_rsc_fence_device 0x00000040ULL
# define pe_rsc_promotable 0x00000080ULL
# define pe_rsc_provisional 0x00000100ULL
# define pe_rsc_allocating 0x00000200ULL
# define pe_rsc_merging 0x00000400ULL
# define pe_rsc_reload 0x00002000ULL
# define pe_rsc_allow_remote_remotes 0x00004000ULL
# define pe_rsc_failed 0x00010000ULL
# define pe_rsc_runnable 0x00040000ULL
# define pe_rsc_start_pending 0x00080000ULL
# define pe_rsc_starting 0x00100000ULL
# define pe_rsc_stopping 0x00200000ULL
# define pe_rsc_allow_migrate 0x00800000ULL
# define pe_rsc_failure_ignored 0x01000000ULL
# define pe_rsc_maintenance 0x04000000ULL
# define pe_rsc_is_container 0x08000000ULL
# define pe_rsc_needs_quorum 0x10000000ULL
# define pe_rsc_needs_fencing 0x20000000ULL
# define pe_rsc_needs_unfencing 0x40000000ULL
enum pe_graph_flags {
pe_graph_none = 0x00000,
pe_graph_updated_first = 0x00001,
pe_graph_updated_then = 0x00002,
pe_graph_disable = 0x00004,
};
/* *INDENT-OFF* */
enum pe_action_flags {
pe_action_pseudo = 0x00001,
pe_action_runnable = 0x00002,
pe_action_optional = 0x00004,
pe_action_print_always = 0x00008,
pe_action_have_node_attrs = 0x00010,
pe_action_implied_by_stonith = 0x00040,
pe_action_migrate_runnable = 0x00080,
pe_action_dumped = 0x00100,
pe_action_processed = 0x00200,
pe_action_clear = 0x00400,
pe_action_dangle = 0x00800,
/* This action requires one or more of its dependencies to be runnable.
* We use this to clear the runnable flag before checking dependencies.
*/
pe_action_requires_any = 0x01000,
pe_action_reschedule = 0x02000,
pe_action_tracking = 0x04000,
pe_action_dedup = 0x08000, //! Internal state tracking when creating graph
+
+ pe_action_dc = 0x10000, //! Action may run on DC instead of target
};
/* *INDENT-ON* */
struct pe_resource_s {
char *id;
char *clone_name;
xmlNode *xml;
xmlNode *orig_xml;
xmlNode *ops_xml;
pe_working_set_t *cluster;
pe_resource_t *parent;
enum pe_obj_types variant;
void *variant_opaque;
resource_object_functions_t *fns;
resource_alloc_functions_t *cmds;
enum rsc_recovery_type recovery_type;
// @TODO only pe_restart_restart is of interest, so merge into flags
enum pe_restart restart_type; //!< \deprecated will be removed in future release
int priority;
int stickiness;
int sort_index;
int failure_timeout;
int migration_threshold;
guint remote_reconnect_ms;
char *pending_task;
unsigned long long flags;
// @TODO merge these into flags
gboolean is_remote_node;
gboolean exclusive_discover;
//!@{
//! This field should be treated as internal to Pacemaker
GListPtr rsc_cons_lhs; // List of rsc_colocation_t*
GListPtr rsc_cons; // List of rsc_colocation_t*
GListPtr rsc_location; // List of pe__location_t*
GListPtr actions; // List of pe_action_t*
GListPtr rsc_tickets; // List of rsc_ticket*
//!@}
pe_node_t *allocated_to;
pe_node_t *partial_migration_target;
pe_node_t *partial_migration_source;
GListPtr running_on; /* pe_node_t* */
GHashTable *known_on; /* pe_node_t* */
GHashTable *allowed_nodes; /* pe_node_t* */
enum rsc_role_e role;
enum rsc_role_e next_role;
GHashTable *meta;
GHashTable *parameters;
GHashTable *utilization;
GListPtr children; /* pe_resource_t* */
GListPtr dangling_migrations; /* pe_node_t* */
pe_resource_t *container;
GListPtr fillers;
pe_node_t *pending_node; // Node on which pending_task is happening
+ pe_node_t *lock_node; // Resource is shutdown-locked to this node
+ time_t lock_time; // When shutdown lock started
#if ENABLE_VERSIONED_ATTRS
xmlNode *versioned_parameters;
#endif
};
#if ENABLE_VERSIONED_ATTRS
// Used as action->action_details if action->rsc is not NULL
typedef struct pe_rsc_action_details_s {
xmlNode *versioned_parameters;
xmlNode *versioned_meta;
} pe_rsc_action_details_t;
#endif
struct pe_action_s {
int id;
int priority;
pe_resource_t *rsc;
pe_node_t *node;
xmlNode *op_entry;
char *task;
char *uuid;
char *cancel_task;
char *reason;
enum pe_action_flags flags;
enum rsc_start_requirement needs;
enum action_fail_response on_fail;
enum rsc_role_e fail_role;
GHashTable *meta;
GHashTable *extra;
/*
* These two varables are associated with the constraint logic
* that involves first having one or more actions runnable before
* then allowing this action to execute.
*
* These varables are used with features such as 'clone-min' which
* requires at minimum X number of cloned instances to be running
* before an order dependency can run. Another option that uses
* this is 'require-all=false' in ordering constrants. This option
* says "only require one instance of a resource to start before
* allowing dependencies to start" -- basically, require-all=false is
* the same as clone-min=1.
*/
/* current number of known runnable actions in the before list. */
int runnable_before;
/* the number of "before" runnable actions required for this action
* to be considered runnable */
int required_runnable_before;
GListPtr actions_before; /* pe_action_wrapper_t* */
GListPtr actions_after; /* pe_action_wrapper_t* */
/* Some of the above fields could be moved to the details,
* except for API backward compatibility.
*/
void *action_details; // varies by type of action
};
typedef struct pe_ticket_s {
char *id;
gboolean granted;
time_t last_granted;
gboolean standby;
GHashTable *state;
} pe_ticket_t;
typedef struct pe_tag_s {
char *id;
GListPtr refs;
} pe_tag_t;
//! Internal tracking for transition graph creation
enum pe_link_state {
pe_link_not_dumped, //! Internal tracking for transition graph creation
pe_link_dumped, //! Internal tracking for transition graph creation
pe_link_dup, //! \deprecated No longer used by Pacemaker
};
enum pe_discover_e {
pe_discover_always = 0,
pe_discover_never,
pe_discover_exclusive,
};
/* *INDENT-OFF* */
enum pe_ordering {
pe_order_none = 0x0, /* deleted */
pe_order_optional = 0x1, /* pure ordering, nothing implied */
pe_order_apply_first_non_migratable = 0x2, /* Only apply this constraint's ordering if first is not migratable. */
pe_order_implies_first = 0x10, /* If 'then' is required, ensure 'first' is too */
pe_order_implies_then = 0x20, /* If 'first' is required, ensure 'then' is too */
pe_order_implies_first_master = 0x40, /* Imply 'first' is required when 'then' is required and then's rsc holds Master role. */
/* first requires then to be both runnable and migrate runnable. */
pe_order_implies_first_migratable = 0x80,
pe_order_runnable_left = 0x100, /* 'then' requires 'first' to be runnable */
pe_order_pseudo_left = 0x200, /* 'then' can only be pseudo if 'first' is runnable */
pe_order_implies_then_on_node = 0x400, /* If 'first' is required on 'nodeX',
* ensure instances of 'then' on 'nodeX' are too.
* Only really useful if 'then' is a clone and 'first' is not
*/
pe_order_probe = 0x800, /* If 'first->rsc' is
* - running but about to stop, ignore the constraint
* - otherwise, behave as runnable_left
*/
pe_order_restart = 0x1000, /* 'then' is runnable if 'first' is optional or runnable */
pe_order_stonith_stop = 0x2000, /* only applies if the action is non-pseudo */
pe_order_serialize_only = 0x4000, /* serialize */
pe_order_same_node = 0x8000, /* applies only if 'first' and 'then' are on same node */
pe_order_implies_first_printed = 0x10000, /* Like ..implies_first but only ensures 'first' is printed, not mandatory */
pe_order_implies_then_printed = 0x20000, /* Like ..implies_then but only ensures 'then' is printed, not mandatory */
pe_order_asymmetrical = 0x100000, /* Indicates asymmetrical one way ordering constraint. */
pe_order_load = 0x200000, /* Only relevant if... */
pe_order_one_or_more = 0x400000, /* 'then' is runnable only if one or more of its dependencies are too */
pe_order_anti_colocation = 0x800000,
pe_order_preserve = 0x1000000, /* Hack for breaking user ordering constraints with container resources */
pe_order_then_cancels_first = 0x2000000, // if 'then' becomes required, 'first' becomes optional
pe_order_trace = 0x4000000, /* test marker */
};
/* *INDENT-ON* */
typedef struct pe_action_wrapper_s {
enum pe_ordering type;
enum pe_link_state state;
pe_action_t *action;
} pe_action_wrapper_t;
// Deprecated type aliases
typedef struct pe_action_s action_t; //!< \deprecated Use pe_action_t instead
typedef struct pe_action_wrapper_s action_wrapper_t; //!< \deprecated Use pe_action_wrapper_t instead
typedef struct pe_node_s node_t; //!< \deprecated Use pe_node_t instead
typedef struct pe_resource_s resource_t; //!< \deprecated Use pe_resource_t instead
typedef struct pe_tag_s tag_t; //!< \deprecated Use pe_tag_t instead
typedef struct pe_ticket_s ticket_t; //!< \deprecated Use pe_ticket_t instead
typedef enum pe_quorum_policy no_quorum_policy_t; //!< \deprecated Use enum pe_quorum_policy instead
#ifdef __cplusplus
}
#endif
#endif // PE_TYPES__H
diff --git a/include/crm_internal.h b/include/crm_internal.h
index 1f25686353..2fa53dd253 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -1,310 +1,312 @@
/*
* Copyright 2006-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef CRM_INTERNAL__H
# define CRM_INTERNAL__H
# include
# include
# include
# include
# include
# include
# include
# include
# include
/* This symbol allows us to deprecate public API and prevent internal code from
* using it while still keeping it for backward compatibility.
*/
#define PCMK__NO_COMPAT
/* Dynamic loading of libraries */
void *find_library_function(void **handle, const char *lib, const char *fn, int fatal);
/* For ACLs */
char *pcmk__uid2username(uid_t uid);
const char *crm_acl_get_set_user(xmlNode * request, const char *field, const char *peer_user);
# if ENABLE_ACL
# include
static inline gboolean
is_privileged(const char *user)
{
if (user == NULL) {
return FALSE;
} else if (strcmp(user, CRM_DAEMON_USER) == 0) {
return TRUE;
} else if (strcmp(user, "root") == 0) {
return TRUE;
}
return FALSE;
}
# endif
/* CLI option processing*/
# ifdef HAVE_GETOPT_H
# include
# else
# define no_argument 0
# define required_argument 1
# endif
# define pcmk_option_default 0x00000
# define pcmk_option_hidden 0x00001
# define pcmk_option_paragraph 0x00002
# define pcmk_option_example 0x00004
struct crm_option {
/* Fields from 'struct option' in getopt.h */
/* name of long option */
const char *name;
/*
* one of no_argument, required_argument, and optional_argument:
* whether option takes an argument
*/
int has_arg;
/* if not NULL, set *flag to val when option found */
int *flag;
/* if flag not NULL, value to set *flag to; else return value */
int val;
/* Custom fields */
const char *desc;
long flags;
};
void crm_set_options(const char *short_options, const char *usage, struct crm_option *long_options,
const char *app_desc);
int crm_get_option(int argc, char **argv, int *index);
int crm_get_option_long(int argc, char **argv, int *index, const char **longname);
_Noreturn void crm_help(char cmd, crm_exit_t exit_code);
/* Cluster Option Processing */
typedef struct pe_cluster_option_s {
const char *name;
const char *alt_name;
const char *type;
const char *values;
const char *default_value;
gboolean(*is_valid) (const char *);
const char *description_short;
const char *description_long;
} pe_cluster_option;
const char *cluster_option(GHashTable * options, gboolean(*validate) (const char *),
const char *name, const char *old_name, const char *def_value);
const char *get_cluster_pref(GHashTable * options, pe_cluster_option * option_list, int len,
const char *name);
void config_metadata(const char *name, const char *version, const char *desc_short,
const char *desc_long, pe_cluster_option * option_list, int len);
void verify_all_options(GHashTable * options, pe_cluster_option * option_list, int len);
gboolean check_time(const char *value);
gboolean check_timer(const char *value);
gboolean check_boolean(const char *value);
gboolean check_number(const char *value);
gboolean check_positive_number(const char *value);
gboolean check_quorum(const char *value);
gboolean check_script(const char *value);
gboolean check_utilization(const char *value);
long crm_get_sbd_timeout(void);
long crm_auto_watchdog_timeout(void);
gboolean check_sbd_timeout(const char *value);
void crm_args_fini(void);
/* char2score */
extern int node_score_red;
extern int node_score_green;
extern int node_score_yellow;
/* Assorted convenience functions */
void crm_make_daemon(const char *name, gboolean daemonize, const char *pidfile);
// printf-style format to create operation ID from resource, action, interval
#define CRM_OP_FMT "%s_%s_%u"
static inline long long
crm_clear_bit(const char *function, int line, const char *target, long long word, long long bit)
{
long long rc = (word & ~bit);
if (rc == word) {
/* Unchanged */
} else if (target) {
crm_trace("Bit 0x%.8llx for %s cleared by %s:%d", bit, target, function, line);
} else {
crm_trace("Bit 0x%.8llx cleared by %s:%d", bit, function, line);
}
return rc;
}
static inline long long
crm_set_bit(const char *function, int line, const char *target, long long word, long long bit)
{
long long rc = (word | bit);
if (rc == word) {
/* Unchanged */
} else if (target) {
crm_trace("Bit 0x%.8llx for %s set by %s:%d", bit, target, function, line);
} else {
crm_trace("Bit 0x%.8llx set by %s:%d", bit, function, line);
}
return rc;
}
# define set_bit(word, bit) word = crm_set_bit(__FUNCTION__, __LINE__, NULL, word, bit)
# define clear_bit(word, bit) word = crm_clear_bit(__FUNCTION__, __LINE__, NULL, word, bit)
char *generate_hash_key(const char *crm_msg_reference, const char *sys);
const char *daemon_option(const char *option);
void set_daemon_option(const char *option, const char *value);
gboolean daemon_option_enabled(const char *daemon, const char *option);
void strip_text_nodes(xmlNode * xml);
void pcmk_panic(const char *origin);
pid_t pcmk_locate_sbd(void);
# define crm_config_err(fmt...) { crm_config_error = TRUE; crm_err(fmt); }
# define crm_config_warn(fmt...) { crm_config_warning = TRUE; crm_warn(fmt); }
# define F_ATTRD_KEY "attr_key"
# define F_ATTRD_ATTRIBUTE "attr_name"
# define F_ATTRD_REGEX "attr_regex"
# define F_ATTRD_TASK "task"
# define F_ATTRD_VALUE "attr_value"
# define F_ATTRD_SET "attr_set"
# define F_ATTRD_IS_REMOTE "attr_is_remote"
# define F_ATTRD_IS_PRIVATE "attr_is_private"
# define F_ATTRD_SECTION "attr_section"
# define F_ATTRD_DAMPEN "attr_dampening"
# define F_ATTRD_HOST "attr_host"
# define F_ATTRD_HOST_ID "attr_host_id"
# define F_ATTRD_USER "attr_user"
# define F_ATTRD_WRITER "attr_writer"
# define F_ATTRD_VERSION "attr_version"
# define F_ATTRD_RESOURCE "attr_resource"
# define F_ATTRD_OPERATION "attr_clear_operation"
# define F_ATTRD_INTERVAL "attr_clear_interval"
# define F_ATTRD_IS_FORCE_WRITE "attrd_is_force_write"
/* attrd operations */
# define ATTRD_OP_PEER_REMOVE "peer-remove"
# define ATTRD_OP_UPDATE "update"
# define ATTRD_OP_UPDATE_BOTH "update-both"
# define ATTRD_OP_UPDATE_DELAY "update-delay"
# define ATTRD_OP_QUERY "query"
# define ATTRD_OP_REFRESH "refresh"
# define ATTRD_OP_FLUSH "flush"
# define ATTRD_OP_SYNC "sync"
# define ATTRD_OP_SYNC_RESPONSE "sync-response"
# define ATTRD_OP_CLEAR_FAILURE "clear-failure"
+# define PCMK__XA_MODE "mode"
+
# define PCMK_ENV_PHYSICAL_HOST "physical_host"
# if SUPPORT_COROSYNC
# include
# include
typedef struct qb_ipc_request_header cs_ipc_header_request_t;
typedef struct qb_ipc_response_header cs_ipc_header_response_t;
# else
typedef struct {
int size __attribute__ ((aligned(8)));
int id __attribute__ ((aligned(8)));
} __attribute__ ((aligned(8))) cs_ipc_header_request_t;
typedef struct {
int size __attribute__ ((aligned(8)));
int id __attribute__ ((aligned(8)));
int error __attribute__ ((aligned(8)));
} __attribute__ ((aligned(8))) cs_ipc_header_response_t;
# endif
void
attrd_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb);
void
stonith_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb);
qb_ipcs_service_t *
crmd_ipc_server_init(struct qb_ipcs_service_handlers *cb);
void cib_ipc_servers_init(qb_ipcs_service_t **ipcs_ro,
qb_ipcs_service_t **ipcs_rw,
qb_ipcs_service_t **ipcs_shm,
struct qb_ipcs_service_handlers *ro_cb,
struct qb_ipcs_service_handlers *rw_cb);
void cib_ipc_servers_destroy(qb_ipcs_service_t *ipcs_ro,
qb_ipcs_service_t *ipcs_rw,
qb_ipcs_service_t *ipcs_shm);
static inline void *realloc_safe(void *ptr, size_t size)
{
void *ret = realloc(ptr, size);
if (ret == NULL) {
free(ptr); /* make coverity happy */
abort();
}
return ret;
}
const char *crm_xml_add_last_written(xmlNode *xml_node);
void crm_xml_dump(xmlNode * data, int options, char **buffer, int *offset, int *max, int depth);
void crm_buffer_add_char(char **buffer, int *offset, int *max, char c);
gboolean crm_digest_verify(xmlNode *input, const char *expected);
/* cross-platform compatibility functions */
char *crm_compat_realpath(const char *path);
/* IPC Proxy Backend Shared Functions */
typedef struct remote_proxy_s {
char *node_name;
char *session_id;
gboolean is_local;
crm_ipc_t *ipc;
mainloop_io_t *source;
uint32_t last_request_id;
lrmd_t *lrm;
} remote_proxy_t;
remote_proxy_t *remote_proxy_new(
lrmd_t *lrmd, struct ipc_client_callbacks *proxy_callbacks,
const char *node_name, const char *session_id, const char *channel);
int remote_proxy_check(lrmd_t *lrmd, GHashTable *hash);
void remote_proxy_cb(lrmd_t *lrmd, const char *node_name, xmlNode *msg);
void remote_proxy_ack_shutdown(lrmd_t *lrmd);
void remote_proxy_nack_shutdown(lrmd_t *lrmd);
int remote_proxy_dispatch(const char *buffer, ssize_t length, gpointer userdata);
void remote_proxy_disconnected(gpointer data);
void remote_proxy_free(gpointer data);
void remote_proxy_relay_event(remote_proxy_t *proxy, xmlNode *msg);
void remote_proxy_relay_response(remote_proxy_t *proxy, xmlNode *msg, int msg_id);
#endif /* CRM_INTERNAL__H */
diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c
index 2469c52c54..d16743d9e9 100644
--- a/lib/lrmd/lrmd_client.c
+++ b/lib/lrmd/lrmd_client.c
@@ -1,2040 +1,2072 @@
/*
- * Copyright 2012-2018 David Vossel
+ * Copyright 2012-2020 the Pacemaker project contributors
+ *
+ * The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#ifdef HAVE_GNUTLS_GNUTLS_H
# undef KEYFILE
# include
#endif
#include
#include
#include
#include
#include
#define MAX_TLS_RECV_WAIT 10000
CRM_TRACE_INIT_DATA(lrmd);
static int lrmd_api_disconnect(lrmd_t * lrmd);
static int lrmd_api_is_connected(lrmd_t * lrmd);
/* IPC proxy functions */
int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg);
static void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg);
void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg));
#ifdef HAVE_GNUTLS_GNUTLS_H
# define LRMD_CLIENT_HANDSHAKE_TIMEOUT 5000 /* 5 seconds */
gnutls_psk_client_credentials_t psk_cred_s;
int lrmd_tls_set_key(gnutls_datum_t * key);
static void lrmd_tls_disconnect(lrmd_t * lrmd);
static int global_remote_msg_id = 0;
int lrmd_tls_send_msg(crm_remote_t * session, xmlNode * msg, uint32_t id, const char *msg_type);
static void lrmd_tls_connection_destroy(gpointer userdata);
#endif
typedef struct lrmd_private_s {
enum client_type type;
char *token;
mainloop_io_t *source;
/* IPC parameters */
crm_ipc_t *ipc;
crm_remote_t *remote;
/* Extra TLS parameters */
char *remote_nodename;
#ifdef HAVE_GNUTLS_GNUTLS_H
char *server;
int port;
gnutls_psk_client_credentials_t psk_cred_c;
/* while the async connection is occurring, this is the id
* of the connection timeout timer. */
int async_timer;
int sock;
/* since tls requires a round trip across the network for a
* request/reply, there are times where we just want to be able
* to send a request from the client and not wait around (or even care
* about) what the reply is. */
int expected_late_replies;
GList *pending_notify;
crm_trigger_t *process_notify;
#endif
lrmd_event_callback callback;
/* Internal IPC proxy msg passing for remote guests */
void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg);
void *proxy_callback_userdata;
char *peer_version;
} lrmd_private_t;
static lrmd_list_t *
lrmd_list_add(lrmd_list_t * head, const char *value)
{
lrmd_list_t *p, *end;
p = calloc(1, sizeof(lrmd_list_t));
p->val = strdup(value);
end = head;
while (end && end->next) {
end = end->next;
}
if (end) {
end->next = p;
} else {
head = p;
}
return head;
}
void
lrmd_list_freeall(lrmd_list_t * head)
{
lrmd_list_t *p;
while (head) {
char *val = (char *)head->val;
p = head->next;
free(val);
free(head);
head = p;
}
}
lrmd_key_value_t *
lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value)
{
lrmd_key_value_t *p, *end;
p = calloc(1, sizeof(lrmd_key_value_t));
p->key = strdup(key);
p->value = strdup(value);
end = head;
while (end && end->next) {
end = end->next;
}
if (end) {
end->next = p;
} else {
head = p;
}
return head;
}
void
lrmd_key_value_freeall(lrmd_key_value_t * head)
{
lrmd_key_value_t *p;
while (head) {
p = head->next;
free(head->key);
free(head->value);
free(head);
head = p;
}
}
+/*!
+ * Create a new lrmd_event_data_t object
+ *
+ * \param[in] rsc_id ID of resource involved in event
+ * \param[in] task Action name
+ * \param[in] interval_ms Action interval
+ *
+ * \return Newly allocated and initialized lrmd_event_data_t
+ * \note This functions asserts on memory errors, so the return value is
+ * guaranteed to be non-NULL. The caller is responsible for freeing the
+ * result with lrmd_free_event().
+ */
+lrmd_event_data_t *
+lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms)
+{
+ lrmd_event_data_t *event = calloc(1, sizeof(lrmd_event_data_t));
+
+ CRM_ASSERT(event != NULL);
+ if (rsc_id != NULL) {
+ event->rsc_id = strdup(rsc_id);
+ CRM_ASSERT(event->rsc_id != NULL);
+ }
+ if (task != NULL) {
+ event->op_type = strdup(task);
+ CRM_ASSERT(event->op_type != NULL);
+ }
+ event->interval_ms = interval_ms;
+ return event;
+}
+
lrmd_event_data_t *
lrmd_copy_event(lrmd_event_data_t * event)
{
lrmd_event_data_t *copy = NULL;
copy = calloc(1, sizeof(lrmd_event_data_t));
/* This will get all the int values.
* we just have to be careful not to leave any
* dangling pointers to strings. */
memcpy(copy, event, sizeof(lrmd_event_data_t));
copy->rsc_id = event->rsc_id ? strdup(event->rsc_id) : NULL;
copy->op_type = event->op_type ? strdup(event->op_type) : NULL;
copy->user_data = event->user_data ? strdup(event->user_data) : NULL;
copy->output = event->output ? strdup(event->output) : NULL;
copy->exit_reason = event->exit_reason ? strdup(event->exit_reason) : NULL;
copy->remote_nodename = event->remote_nodename ? strdup(event->remote_nodename) : NULL;
copy->params = crm_str_table_dup(event->params);
return copy;
}
void
lrmd_free_event(lrmd_event_data_t * event)
{
if (!event) {
return;
}
/* free gives me grief if i try to cast */
free((char *)event->rsc_id);
free((char *)event->op_type);
free((char *)event->user_data);
free((char *)event->output);
free((char *)event->exit_reason);
free((char *)event->remote_nodename);
if (event->params) {
g_hash_table_destroy(event->params);
}
free(event);
}
static int
lrmd_dispatch_internal(lrmd_t * lrmd, xmlNode * msg)
{
const char *type;
const char *proxy_session = crm_element_value(msg, F_LRMD_IPC_SESSION);
lrmd_private_t *native = lrmd->lrmd_private;
lrmd_event_data_t event = { 0, };
if (proxy_session != NULL) {
/* this is proxy business */
lrmd_internal_proxy_dispatch(lrmd, msg);
return 1;
} else if (!native->callback) {
/* no callback set */
crm_trace("notify event received but client has not set callback");
return 1;
}
event.remote_nodename = native->remote_nodename;
type = crm_element_value(msg, F_LRMD_OPERATION);
crm_element_value_int(msg, F_LRMD_CALLID, &event.call_id);
event.rsc_id = crm_element_value(msg, F_LRMD_RSC_ID);
if (crm_str_eq(type, LRMD_OP_RSC_REG, TRUE)) {
event.type = lrmd_event_register;
} else if (crm_str_eq(type, LRMD_OP_RSC_UNREG, TRUE)) {
event.type = lrmd_event_unregister;
} else if (crm_str_eq(type, LRMD_OP_RSC_EXEC, TRUE)) {
time_t epoch = 0;
crm_element_value_int(msg, F_LRMD_TIMEOUT, &event.timeout);
crm_element_value_ms(msg, F_LRMD_RSC_INTERVAL, &event.interval_ms);
crm_element_value_int(msg, F_LRMD_RSC_START_DELAY, &event.start_delay);
crm_element_value_int(msg, F_LRMD_EXEC_RC, (int *)&event.rc);
crm_element_value_int(msg, F_LRMD_OP_STATUS, &event.op_status);
crm_element_value_int(msg, F_LRMD_RSC_DELETED, &event.rsc_deleted);
crm_element_value_epoch(msg, F_LRMD_RSC_RUN_TIME, &epoch);
event.t_run = (unsigned int) epoch;
crm_element_value_epoch(msg, F_LRMD_RSC_RCCHANGE_TIME, &epoch);
event.t_rcchange = (unsigned int) epoch;
crm_element_value_int(msg, F_LRMD_RSC_EXEC_TIME, (int *)&event.exec_time);
crm_element_value_int(msg, F_LRMD_RSC_QUEUE_TIME, (int *)&event.queue_time);
event.op_type = crm_element_value(msg, F_LRMD_RSC_ACTION);
event.user_data = crm_element_value(msg, F_LRMD_RSC_USERDATA_STR);
event.output = crm_element_value(msg, F_LRMD_RSC_OUTPUT);
event.exit_reason = crm_element_value(msg, F_LRMD_RSC_EXIT_REASON);
event.type = lrmd_event_exec_complete;
event.params = xml2list(msg);
} else if (crm_str_eq(type, LRMD_OP_NEW_CLIENT, TRUE)) {
event.type = lrmd_event_new_client;
} else if (crm_str_eq(type, LRMD_OP_POKE, TRUE)) {
event.type = lrmd_event_poke;
} else {
return 1;
}
crm_trace("op %s notify event received", type);
native->callback(&event);
if (event.params) {
g_hash_table_destroy(event.params);
}
return 1;
}
static int
lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *msg;
int rc;
if (!native->callback) {
/* no callback set */
return 1;
}
msg = string2xml(buffer);
rc = lrmd_dispatch_internal(lrmd, msg);
free_xml(msg);
return rc;
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_free_xml(gpointer userdata)
{
free_xml((xmlNode *) userdata);
}
static int
lrmd_tls_connected(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->remote->tls_session) {
return TRUE;
}
return FALSE;
}
static int
lrmd_tls_dispatch(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *xml = NULL;
int rc = 0;
int disconnected = 0;
if (lrmd_tls_connected(lrmd) == FALSE) {
crm_trace("TLS dispatch triggered after disconnect");
return 0;
}
crm_trace("TLS dispatch triggered");
/* First check if there are any pending notifies to process that came
* while we were waiting for replies earlier. */
if (native->pending_notify) {
GList *iter = NULL;
crm_trace("Processing pending notifies");
for (iter = native->pending_notify; iter; iter = iter->next) {
lrmd_dispatch_internal(lrmd, iter->data);
}
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
/* Next read the current buffer and see if there are any messages to handle. */
rc = crm_remote_ready(native->remote, 0);
if (rc == 0) {
/* nothing to read, see if any full messages are already in buffer. */
xml = crm_remote_parse_buffer(native->remote);
} else if (rc < 0) {
disconnected = 1;
} else {
crm_remote_recv(native->remote, -1, &disconnected);
xml = crm_remote_parse_buffer(native->remote);
}
while (xml) {
const char *msg_type = crm_element_value(xml, F_LRMD_REMOTE_MSG_TYPE);
if (safe_str_eq(msg_type, "notify")) {
lrmd_dispatch_internal(lrmd, xml);
} else if (safe_str_eq(msg_type, "reply")) {
if (native->expected_late_replies > 0) {
native->expected_late_replies--;
} else {
int reply_id = 0;
crm_element_value_int(xml, F_LRMD_CALLID, &reply_id);
/* if this happens, we want to know about it */
crm_err("Got outdated Pacemaker Remote reply %d", reply_id);
}
}
free_xml(xml);
xml = crm_remote_parse_buffer(native->remote);
}
if (disconnected) {
crm_info("Lost %s executor connection while reading data",
(native->remote_nodename? native->remote_nodename : "local"));
lrmd_tls_disconnect(lrmd);
return 0;
}
return 1;
}
#endif
/* Not used with mainloop */
int
lrmd_poll(lrmd_t * lrmd, int timeout)
{
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case CRM_CLIENT_IPC:
return crm_ipc_ready(native->ipc);
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
if (native->pending_notify) {
return 1;
}
return crm_remote_ready(native->remote, 0);
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
return 0;
}
/* Not used with mainloop */
bool
lrmd_dispatch(lrmd_t * lrmd)
{
lrmd_private_t *private = NULL;
CRM_ASSERT(lrmd != NULL);
private = lrmd->lrmd_private;
switch (private->type) {
case CRM_CLIENT_IPC:
while (crm_ipc_ready(private->ipc)) {
if (crm_ipc_read(private->ipc) > 0) {
const char *msg = crm_ipc_buffer(private->ipc);
lrmd_ipc_dispatch(msg, strlen(msg), lrmd);
}
}
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
lrmd_tls_dispatch(lrmd);
break;
#endif
default:
crm_err("Unsupported connection type: %d", private->type);
}
if (lrmd_api_is_connected(lrmd) == FALSE) {
crm_err("Connection closed");
return FALSE;
}
return TRUE;
}
static xmlNode *
lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout,
enum lrmd_call_options options)
{
xmlNode *op_msg = create_xml_node(NULL, "lrmd_command");
CRM_CHECK(op_msg != NULL, return NULL);
CRM_CHECK(token != NULL, return NULL);
crm_xml_add(op_msg, F_XML_TAGNAME, "lrmd_command");
crm_xml_add(op_msg, F_TYPE, T_LRMD);
crm_xml_add(op_msg, F_LRMD_CALLBACK_TOKEN, token);
crm_xml_add(op_msg, F_LRMD_OPERATION, op);
crm_xml_add_int(op_msg, F_LRMD_TIMEOUT, timeout);
crm_xml_add_int(op_msg, F_LRMD_CALLOPTS, options);
if (data != NULL) {
add_message_xml(op_msg, F_LRMD_CALLDATA, data);
}
crm_trace("Created executor %s command with call options %.8lx (%d)",
op, (long)options, options);
return op_msg;
}
static void
lrmd_ipc_connection_destroy(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
crm_info("IPC connection destroyed");
/* Prevent these from being cleaned up in lrmd_api_disconnect() */
native->ipc = NULL;
native->source = NULL;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.type = lrmd_event_disconnect;
event.remote_nodename = native->remote_nodename;
native->callback(&event);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_tls_connection_destroy(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
crm_info("TLS connection destroyed");
if (native->remote->tls_session) {
gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
}
if (native->psk_cred_c) {
gnutls_psk_free_client_credentials(native->psk_cred_c);
}
if (native->sock) {
close(native->sock);
}
if (native->process_notify) {
mainloop_destroy_trigger(native->process_notify);
native->process_notify = NULL;
}
if (native->pending_notify) {
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
free(native->remote->buffer);
native->remote->buffer = NULL;
native->source = 0;
native->sock = 0;
native->psk_cred_c = NULL;
native->remote->tls_session = NULL;
native->sock = 0;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.remote_nodename = native->remote_nodename;
event.type = lrmd_event_disconnect;
native->callback(&event);
}
return;
}
int
lrmd_tls_send_msg(crm_remote_t * session, xmlNode * msg, uint32_t id, const char *msg_type)
{
crm_xml_add_int(msg, F_LRMD_REMOTE_MSG_ID, id);
crm_xml_add(msg, F_LRMD_REMOTE_MSG_TYPE, msg_type);
return crm_remote_send(session, msg);
}
static xmlNode *
lrmd_tls_recv_reply(lrmd_t * lrmd, int total_timeout, int expected_reply_id, int *disconnected)
{
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *xml = NULL;
time_t start = time(NULL);
const char *msg_type = NULL;
int reply_id = 0;
int remaining_timeout = 0;
/* A timeout of 0 here makes no sense. We have to wait a period of time
* for the response to come back. If -1 or 0, default to 10 seconds. */
if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) {
total_timeout = MAX_TLS_RECV_WAIT;
}
while (!xml) {
xml = crm_remote_parse_buffer(native->remote);
if (!xml) {
/* read some more off the tls buffer if we still have time left. */
if (remaining_timeout) {
remaining_timeout = total_timeout - ((time(NULL) - start) * 1000);
} else {
remaining_timeout = total_timeout;
}
if (remaining_timeout <= 0) {
crm_err("Never received the expected reply during the timeout period, disconnecting.");
*disconnected = TRUE;
return NULL;
}
crm_remote_recv(native->remote, remaining_timeout, disconnected);
xml = crm_remote_parse_buffer(native->remote);
if (!xml) {
crm_err("Unable to receive expected reply, disconnecting.");
*disconnected = TRUE;
return NULL;
} else if (*disconnected) {
return NULL;
}
}
CRM_ASSERT(xml != NULL);
crm_element_value_int(xml, F_LRMD_REMOTE_MSG_ID, &reply_id);
msg_type = crm_element_value(xml, F_LRMD_REMOTE_MSG_TYPE);
if (!msg_type) {
crm_err("Empty msg type received while waiting for reply");
free_xml(xml);
xml = NULL;
} else if (safe_str_eq(msg_type, "notify")) {
/* got a notify while waiting for reply, trigger the notify to be processed later */
crm_info("queueing notify");
native->pending_notify = g_list_append(native->pending_notify, xml);
if (native->process_notify) {
crm_info("notify trigger set.");
mainloop_set_trigger(native->process_notify);
}
xml = NULL;
} else if (safe_str_neq(msg_type, "reply")) {
/* msg isn't a reply, make some noise */
crm_err("Expected a reply, got %s", msg_type);
free_xml(xml);
xml = NULL;
} else if (reply_id != expected_reply_id) {
if (native->expected_late_replies > 0) {
native->expected_late_replies--;
} else {
crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id);
}
free_xml(xml);
xml = NULL;
}
}
if (native->remote->buffer && native->process_notify) {
mainloop_set_trigger(native->process_notify);
}
return xml;
}
static int
lrmd_tls_send(lrmd_t * lrmd, xmlNode * msg)
{
int rc = 0;
lrmd_private_t *native = lrmd->lrmd_private;
global_remote_msg_id++;
if (global_remote_msg_id <= 0) {
global_remote_msg_id = 1;
}
rc = lrmd_tls_send_msg(native->remote, msg, global_remote_msg_id, "request");
if (rc <= 0) {
crm_err("Disconnecting because TLS message could not be sent to Pacemaker Remote");
lrmd_tls_disconnect(lrmd);
return -ENOTCONN;
}
return pcmk_ok;
}
static int
lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply)
{
int rc = 0;
int disconnected = 0;
xmlNode *xml = NULL;
if (lrmd_tls_connected(lrmd) == FALSE) {
return -1;
}
rc = lrmd_tls_send(lrmd, msg);
if (rc < 0) {
return rc;
}
xml = lrmd_tls_recv_reply(lrmd, timeout, global_remote_msg_id, &disconnected);
if (disconnected) {
crm_err("Pacemaker Remote disconnected while waiting for reply to request id %d",
global_remote_msg_id);
lrmd_tls_disconnect(lrmd);
rc = -ENOTCONN;
} else if (!xml) {
crm_err("Did not receive reply from Pacemaker Remote for request id %d (timeout %dms)",
global_remote_msg_id, timeout);
rc = -ECOMM;
}
if (reply) {
*reply = xml;
} else {
free_xml(xml);
}
return rc;
}
#endif
static int
lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply)
{
int rc = -1;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case CRM_CLIENT_IPC:
rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply);
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
return rc;
}
static int
lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg)
{
int rc = -1;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case CRM_CLIENT_IPC:
rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
rc = lrmd_tls_send(lrmd, msg);
if (rc == pcmk_ok) {
/* we don't want to wait around for the reply, but
* since the request/reply protocol needs to behave the same
* as libqb, a reply will eventually come later anyway. */
native->expected_late_replies++;
}
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
return rc;
}
static int
lrmd_api_is_connected(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case CRM_CLIENT_IPC:
return crm_ipc_connected(native->ipc);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
return lrmd_tls_connected(lrmd);
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
return 0;
}
/*!
* \internal
* \brief Send a prepared API command to the executor
*
* \param[in] lrmd Existing connection to the executor
* \param[in] op Name of API command to send
* \param[in] data Command data XML to add to the sent command
* \param[out] output_data If expecting a reply, it will be stored here
* \param[in] timeout Timeout in milliseconds (if 0, defaults to
* a sensible value per the type of connection,
* standard vs. pacemaker remote);
* also propagated to the command XML
* \param[in] call_options Call options to pass to server when sending
* \param[in] expect_reply If TRUE, wait for a reply from the server;
* must be TRUE for IPC (as opposed to TLS) clients
*
* \return pcmk_ok on success, -errno on error
*/
static int
lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data,
xmlNode **output_data, int timeout,
enum lrmd_call_options options, gboolean expect_reply)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *op_msg = NULL;
xmlNode *op_reply = NULL;
if (!lrmd_api_is_connected(lrmd)) {
return -ENOTCONN;
}
if (op == NULL) {
crm_err("No operation specified");
return -EINVAL;
}
CRM_CHECK(native->token != NULL,;
);
crm_trace("Sending %s op to executor", op);
op_msg = lrmd_create_op(native->token, op, data, timeout, options);
if (op_msg == NULL) {
return -EINVAL;
}
if (expect_reply) {
rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply);
} else {
rc = lrmd_send_xml_no_reply(lrmd, op_msg);
goto done;
}
if (rc < 0) {
crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc);
rc = -ECOMM;
goto done;
} else if(op_reply == NULL) {
rc = -ENOMSG;
goto done;
}
rc = pcmk_ok;
crm_trace("%s op reply received", op);
if (crm_element_value_int(op_reply, F_LRMD_RC, &rc) != 0) {
rc = -ENOMSG;
goto done;
}
crm_log_xml_trace(op_reply, "Reply");
if (output_data) {
*output_data = op_reply;
op_reply = NULL; /* Prevent subsequent free */
}
done:
if (lrmd_api_is_connected(lrmd) == FALSE) {
crm_err("Executor disconnected");
}
free_xml(op_msg);
free_xml(op_reply);
return rc;
}
static int
lrmd_api_poke_connection(lrmd_t * lrmd)
{
int rc;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0, native->type == CRM_CLIENT_IPC ? TRUE : FALSE);
free_xml(data);
return rc < 0 ? rc : pcmk_ok;
}
int
remote_proxy_check(lrmd_t * lrmd, GHashTable *hash)
{
int rc;
const char *value;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *data = create_xml_node(NULL, F_LRMD_OPERATION);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
value = g_hash_table_lookup(hash, "stonith-watchdog-timeout");
crm_xml_add(data, F_LRMD_WATCHDOG, value);
rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0, native->type == CRM_CLIENT_IPC ? TRUE : FALSE);
free_xml(data);
return rc < 0 ? rc : pcmk_ok;
}
static int
lrmd_handshake(lrmd_t * lrmd, const char *name)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *reply = NULL;
xmlNode *hello = create_xml_node(NULL, "lrmd_command");
crm_xml_add(hello, F_TYPE, T_LRMD);
crm_xml_add(hello, F_LRMD_OPERATION, CRM_OP_REGISTER);
crm_xml_add(hello, F_LRMD_CLIENTNAME, name);
crm_xml_add(hello, F_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION);
/* advertise that we are a proxy provider */
if (native->proxy_callback) {
crm_xml_add(hello, F_LRMD_IS_IPC_PROVIDER, "true");
}
rc = lrmd_send_xml(lrmd, hello, -1, &reply);
if (rc < 0) {
crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc);
rc = -ECOMM;
} else if (reply == NULL) {
crm_err("Did not receive registration reply");
rc = -EPROTO;
} else {
const char *version = crm_element_value(reply, F_LRMD_PROTOCOL_VERSION);
const char *msg_type = crm_element_value(reply, F_LRMD_OPERATION);
const char *tmp_ticket = crm_element_value(reply, F_LRMD_CLIENTID);
crm_element_value_int(reply, F_LRMD_RC, &rc);
if (rc == -EPROTO) {
crm_err("Executor protocol version mismatch between client (%s) and server (%s)",
LRMD_PROTOCOL_VERSION, version);
crm_log_xml_err(reply, "Protocol Error");
} else if (safe_str_neq(msg_type, CRM_OP_REGISTER)) {
crm_err("Invalid registration message: %s", msg_type);
crm_log_xml_err(reply, "Bad reply");
rc = -EPROTO;
} else if (tmp_ticket == NULL) {
crm_err("No registration token provided");
crm_log_xml_err(reply, "Bad reply");
rc = -EPROTO;
} else {
crm_trace("Obtained registration token: %s", tmp_ticket);
native->token = strdup(tmp_ticket);
native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */
rc = pcmk_ok;
}
}
free_xml(reply);
free_xml(hello);
if (rc != pcmk_ok) {
lrmd_api_disconnect(lrmd);
}
return rc;
}
static int
lrmd_ipc_connect(lrmd_t * lrmd, int *fd)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
static struct ipc_client_callbacks lrmd_callbacks = {
.dispatch = lrmd_ipc_dispatch,
.destroy = lrmd_ipc_connection_destroy
};
crm_info("Connecting to executor");
if (fd) {
/* No mainloop */
native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0);
if (native->ipc && crm_ipc_connect(native->ipc)) {
*fd = crm_ipc_get_fd(native->ipc);
} else if (native->ipc) {
crm_perror(LOG_ERR, "Connection to executor failed");
rc = -ENOTCONN;
}
} else {
native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks);
native->ipc = mainloop_get_ipc_client(native->source);
}
if (native->ipc == NULL) {
crm_debug("Could not connect to the executor API");
rc = -ENOTCONN;
}
return rc;
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source)
{
dest->data = gnutls_malloc(source->size);
CRM_ASSERT(dest->data);
memcpy(dest->data, source->data, source->size);
dest->size = source->size;
}
static void
clear_gnutls_datum(gnutls_datum_t *datum)
{
gnutls_free(datum->data);
datum->data = NULL;
datum->size = 0;
}
#define KEY_READ_LEN 256
static int
set_key(gnutls_datum_t * key, const char *location)
{
FILE *stream;
size_t buf_len = KEY_READ_LEN;
static gnutls_datum_t key_cache = { 0, };
static time_t key_cache_updated = 0;
if (location == NULL) {
return -1;
}
if (key_cache.data != NULL) {
if ((time(NULL) - key_cache_updated) < 60) {
copy_gnutls_datum(key, &key_cache);
crm_debug("Using cached Pacemaker Remote key");
return 0;
} else {
clear_gnutls_datum(&key_cache);
key_cache_updated = 0;
crm_debug("Cleared Pacemaker Remote key cache");
}
}
stream = fopen(location, "r");
if (!stream) {
return -1;
}
key->data = gnutls_malloc(buf_len);
key->size = 0;
while (!feof(stream)) {
int next = fgetc(stream);
if (next == EOF) {
if (!feof(stream)) {
crm_err("Error reading Pacemaker Remote key; copy in memory may be corrupted");
}
break;
}
if (key->size == buf_len) {
buf_len = key->size + KEY_READ_LEN;
key->data = gnutls_realloc(key->data, buf_len);
CRM_ASSERT(key->data);
}
key->data[key->size++] = (unsigned char) next;
}
fclose(stream);
if (key->size == 0) {
clear_gnutls_datum(key);
return -1;
}
if (key_cache.data == NULL) {
copy_gnutls_datum(&key_cache, key);
key_cache_updated = time(NULL);
crm_debug("Cached Pacemaker Remote key");
}
return 0;
}
int
lrmd_tls_set_key(gnutls_datum_t * key)
{
const char *specific_location = getenv("PCMK_authkey_location");
if (set_key(key, specific_location) == 0) {
crm_debug("Using custom authkey location %s", specific_location);
return pcmk_ok;
} else if (specific_location) {
crm_err("No valid Pacemaker Remote key found at %s, trying default location", specific_location);
}
if ((set_key(key, DEFAULT_REMOTE_KEY_LOCATION) != 0)
&& (set_key(key, ALT_REMOTE_KEY_LOCATION) != 0)) {
crm_err("No valid Pacemaker Remote key found at %s", DEFAULT_REMOTE_KEY_LOCATION);
return -ENOKEY;
}
return pcmk_ok;
}
static void
lrmd_gnutls_global_init(void)
{
static int gnutls_init = 0;
if (!gnutls_init) {
crm_gnutls_global_init();
}
gnutls_init = 1;
}
#endif
static void
report_async_connection_result(lrmd_t * lrmd, int rc)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.type = lrmd_event_connect;
event.remote_nodename = native->remote_nodename;
event.connection_rc = rc;
native->callback(&event);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_tcp_connect_cb(void *userdata, int sock)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
char *name;
static struct mainloop_fd_callbacks lrmd_tls_callbacks = {
.dispatch = lrmd_tls_dispatch,
.destroy = lrmd_tls_connection_destroy,
};
int rc = sock;
gnutls_datum_t psk_key = { NULL, 0 };
native->async_timer = 0;
if (rc < 0) {
lrmd_tls_connection_destroy(lrmd);
crm_info("Could not connect to Pacemaker Remote at %s:%d",
native->server, native->port);
report_async_connection_result(lrmd, rc);
return;
}
/* The TCP connection was successful, so establish the TLS connection.
* @TODO make this async to avoid blocking code in client
*/
native->sock = sock;
rc = lrmd_tls_set_key(&psk_key);
if (rc != 0) {
crm_warn("Could not set key for Pacemaker Remote at %s:%d " CRM_XS " rc=%d",
native->server, native->port, rc);
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, rc);
return;
}
gnutls_psk_allocate_client_credentials(&native->psk_cred_c);
gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW);
gnutls_free(psk_key.data);
native->remote->tls_session = pcmk__new_tls_session(sock, GNUTLS_CLIENT,
GNUTLS_CRD_PSK,
native->psk_cred_c);
if (native->remote->tls_session == NULL) {
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, -EPROTO);
return;
}
if (crm_initiate_client_tls_handshake(native->remote, LRMD_CLIENT_HANDSHAKE_TIMEOUT) != 0) {
crm_warn("Disconnecting after TLS handshake with Pacemaker Remote server %s:%d failed",
native->server, native->port);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = NULL;
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, -EKEYREJECTED);
return;
}
crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded",
native->server, native->port);
name = crm_strdup_printf("pacemaker-remote-%s:%d",
native->server, native->port);
native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_tls_dispatch, lrmd);
native->source =
mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &lrmd_tls_callbacks);
rc = lrmd_handshake(lrmd, name);
free(name);
report_async_connection_result(lrmd, rc);
return;
}
static int
lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ )
{
int sock = 0;
int timer_id = 0;
lrmd_private_t *native = lrmd->lrmd_private;
lrmd_gnutls_global_init();
sock = crm_remote_tcp_connect_async(native->server, native->port, timeout,
&timer_id, lrmd, lrmd_tcp_connect_cb);
if (sock < 0) {
return sock;
}
native->sock = sock;
native->async_timer = timer_id;
return pcmk_ok;
}
static int
lrmd_tls_connect(lrmd_t * lrmd, int *fd)
{
static struct mainloop_fd_callbacks lrmd_tls_callbacks = {
.dispatch = lrmd_tls_dispatch,
.destroy = lrmd_tls_connection_destroy,
};
int rc;
lrmd_private_t *native = lrmd->lrmd_private;
int sock;
gnutls_datum_t psk_key = { NULL, 0 };
lrmd_gnutls_global_init();
sock = crm_remote_tcp_connect(native->server, native->port);
if (sock < 0) {
crm_warn("Could not establish Pacemaker Remote connection to %s", native->server);
lrmd_tls_connection_destroy(lrmd);
return -ENOTCONN;
}
native->sock = sock;
rc = lrmd_tls_set_key(&psk_key);
if (rc < 0) {
lrmd_tls_connection_destroy(lrmd);
return rc;
}
gnutls_psk_allocate_client_credentials(&native->psk_cred_c);
gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW);
gnutls_free(psk_key.data);
native->remote->tls_session = pcmk__new_tls_session(sock, GNUTLS_CLIENT,
GNUTLS_CRD_PSK,
native->psk_cred_c);
if (native->remote->tls_session == NULL) {
lrmd_tls_connection_destroy(lrmd);
return -EPROTO;
}
if (crm_initiate_client_tls_handshake(native->remote, LRMD_CLIENT_HANDSHAKE_TIMEOUT) != 0) {
crm_err("Session creation for %s:%d failed", native->server, native->port);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = NULL;
lrmd_tls_connection_destroy(lrmd);
return -EKEYREJECTED;
}
crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server,
native->port);
if (fd) {
*fd = sock;
} else {
char *name = crm_strdup_printf("pacemaker-remote-%s:%d",
native->server, native->port);
native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, lrmd_tls_dispatch, lrmd);
native->source =
mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &lrmd_tls_callbacks);
free(name);
}
return pcmk_ok;
}
#endif
static int
lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd)
{
int rc = -ENOTCONN;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case CRM_CLIENT_IPC:
rc = lrmd_ipc_connect(lrmd, fd);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
rc = lrmd_tls_connect(lrmd, fd);
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
if (rc == pcmk_ok) {
rc = lrmd_handshake(lrmd, name);
}
return rc;
}
static int
lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout)
{
int rc = 0;
lrmd_private_t *native = lrmd->lrmd_private;
CRM_CHECK(native && native->callback, return -1);
switch (native->type) {
case CRM_CLIENT_IPC:
/* fake async connection with ipc. it should be fast
* enough that we gain very little from async */
rc = lrmd_api_connect(lrmd, name, NULL);
if (!rc) {
report_async_connection_result(lrmd, rc);
}
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
rc = lrmd_tls_connect_async(lrmd, timeout);
if (rc) {
/* connection failed, report rc now */
report_async_connection_result(lrmd, rc);
}
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
return rc;
}
static void
lrmd_ipc_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->source != NULL) {
/* Attached to mainloop */
mainloop_del_ipc_client(native->source);
native->source = NULL;
native->ipc = NULL;
} else if (native->ipc) {
/* Not attached to mainloop */
crm_ipc_t *ipc = native->ipc;
native->ipc = NULL;
crm_ipc_close(ipc);
crm_ipc_destroy(ipc);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_tls_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->remote->tls_session) {
gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = 0;
}
if (native->async_timer) {
g_source_remove(native->async_timer);
native->async_timer = 0;
}
if (native->source != NULL) {
/* Attached to mainloop */
mainloop_del_ipc_client(native->source);
native->source = NULL;
} else if (native->sock) {
close(native->sock);
native->sock = 0;
}
if (native->pending_notify) {
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
}
#endif
static int
lrmd_api_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
crm_info("Disconnecting %s %s executor connection",
crm_client_type_text(native->type),
(native->remote_nodename? native->remote_nodename : "local"));
switch (native->type) {
case CRM_CLIENT_IPC:
lrmd_ipc_disconnect(lrmd);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case CRM_CLIENT_TLS:
lrmd_tls_disconnect(lrmd);
break;
#endif
default:
crm_err("Unsupported connection type: %d", native->type);
}
free(native->token);
native->token = NULL;
free(native->peer_version);
native->peer_version = NULL;
return 0;
}
static int
lrmd_api_register_rsc(lrmd_t * lrmd,
const char *rsc_id,
const char *class,
const char *provider, const char *type, enum lrmd_call_options options)
{
int rc = pcmk_ok;
xmlNode *data = NULL;
if (!class || !type || !rsc_id) {
return -EINVAL;
}
if (is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && !provider) {
return -EINVAL;
}
data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
crm_xml_add(data, F_LRMD_CLASS, class);
crm_xml_add(data, F_LRMD_PROVIDER, provider);
crm_xml_add(data, F_LRMD_TYPE, type);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, TRUE);
free_xml(data);
return rc;
}
static int
lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, TRUE);
free_xml(data);
return rc;
}
lrmd_rsc_info_t *
lrmd_new_rsc_info(const char *rsc_id, const char *standard,
const char *provider, const char *type)
{
lrmd_rsc_info_t *rsc_info = calloc(1, sizeof(lrmd_rsc_info_t));
CRM_ASSERT(rsc_info);
if (rsc_id) {
rsc_info->id = strdup(rsc_id);
CRM_ASSERT(rsc_info->id);
}
if (standard) {
rsc_info->standard = strdup(standard);
CRM_ASSERT(rsc_info->standard);
}
if (provider) {
rsc_info->provider = strdup(provider);
CRM_ASSERT(rsc_info->provider);
}
if (type) {
rsc_info->type = strdup(type);
CRM_ASSERT(rsc_info->type);
}
return rsc_info;
}
lrmd_rsc_info_t *
lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info)
{
return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard,
rsc_info->provider, rsc_info->type);
}
void
lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info)
{
if (!rsc_info) {
return;
}
free(rsc_info->id);
free(rsc_info->type);
free(rsc_info->standard);
free(rsc_info->provider);
free(rsc_info);
}
static lrmd_rsc_info_t *
lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options)
{
lrmd_rsc_info_t *rsc_info = NULL;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
xmlNode *output = NULL;
const char *class = NULL;
const char *provider = NULL;
const char *type = NULL;
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, TRUE);
free_xml(data);
if (!output) {
return NULL;
}
class = crm_element_value(output, F_LRMD_CLASS);
provider = crm_element_value(output, F_LRMD_PROVIDER);
type = crm_element_value(output, F_LRMD_TYPE);
if (!class || !type) {
free_xml(output);
return NULL;
} else if (is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)
&& !provider) {
free_xml(output);
return NULL;
}
rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type);
free_xml(output);
return rsc_info;
}
void
lrmd_free_op_info(lrmd_op_info_t *op_info)
{
if (op_info) {
free(op_info->rsc_id);
free(op_info->action);
free(op_info->interval_ms_s);
free(op_info->timeout_ms_s);
free(op_info);
}
}
static int
lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms,
enum lrmd_call_options options, GList **output)
{
xmlNode *data = NULL;
xmlNode *output_xml = NULL;
int rc = pcmk_ok;
if (output == NULL) {
return -EINVAL;
}
*output = NULL;
// Send request
if (rsc_id) {
data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
}
rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml,
timeout_ms, options, TRUE);
if (data) {
free_xml(data);
}
// Process reply
if ((rc != pcmk_ok) || (output_xml == NULL)) {
return rc;
}
for (xmlNode *rsc_xml = first_named_child(output_xml, F_LRMD_RSC);
rsc_xml != NULL; rsc_xml = crm_next_same_xml(rsc_xml)) {
rsc_id = crm_element_value(rsc_xml, F_LRMD_RSC_ID);
if (rsc_id == NULL) {
crm_err("Could not parse recurring operation information from executor");
continue;
}
for (xmlNode *op_xml = first_named_child(rsc_xml, T_LRMD_RSC_OP);
op_xml != NULL; op_xml = crm_next_same_xml(op_xml)) {
lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t));
CRM_CHECK(op_info != NULL, break);
op_info->rsc_id = strdup(rsc_id);
op_info->action = crm_element_value_copy(op_xml, F_LRMD_RSC_ACTION);
op_info->interval_ms_s = crm_element_value_copy(op_xml,
F_LRMD_RSC_INTERVAL);
op_info->timeout_ms_s = crm_element_value_copy(op_xml,
F_LRMD_TIMEOUT);
*output = g_list_prepend(*output, op_info);
}
}
free_xml(output_xml);
return rc;
}
static void
lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback)
{
lrmd_private_t *native = lrmd->lrmd_private;
native->callback = callback;
}
void
lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg))
{
lrmd_private_t *native = lrmd->lrmd_private;
native->proxy_callback = callback;
native->proxy_callback_userdata = userdata;
}
void
lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->proxy_callback) {
crm_log_xml_trace(msg, "PROXY_INBOUND");
native->proxy_callback(lrmd, native->proxy_callback_userdata, msg);
}
}
int
lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg)
{
if (lrmd == NULL) {
return -ENOTCONN;
}
crm_xml_add(msg, F_LRMD_OPERATION, CRM_OP_IPC_FWD);
crm_log_xml_trace(msg, "PROXY_OUTBOUND");
return lrmd_send_xml_no_reply(lrmd, msg);
}
static int
stonith_get_metadata(const char *provider, const char *type, char **output)
{
int rc = pcmk_ok;
stonith_t *stonith_api = stonith_api_new();
if (stonith_api == NULL) {
crm_err("Could not get fence agent meta-data: API memory allocation failed");
return -ENOMEM;
}
rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type,
provider, output, 0);
if ((rc == pcmk_ok) && (*output == NULL)) {
rc = -EIO;
}
stonith_api->cmds->free(stonith_api);
return rc;
}
static int
lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider,
const char *type, char **output,
enum lrmd_call_options options)
{
return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type,
output, options, NULL);
}
static int
lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard,
const char *provider, const char *type,
char **output, enum lrmd_call_options options,
lrmd_key_value_t *params)
{
svc_action_t *action = NULL;
GHashTable *params_table = NULL;
if (!standard || !type) {
lrmd_key_value_freeall(params);
return -EINVAL;
}
if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_STONITH)) {
lrmd_key_value_freeall(params);
return stonith_get_metadata(provider, type, output);
}
params_table = crm_str_table_new();
for (const lrmd_key_value_t *param = params; param; param = param->next) {
g_hash_table_insert(params_table, strdup(param->key), strdup(param->value));
}
action = resources_action_create(type, standard, provider, type,
CRMD_ACTION_METADATA, 0,
CRMD_METADATA_CALL_TIMEOUT, params_table,
0);
lrmd_key_value_freeall(params);
if (action == NULL) {
crm_err("Unable to retrieve meta-data for %s:%s:%s",
standard, provider, type);
return -EINVAL;
}
if (!services_action_sync(action)) {
crm_err("Failed to retrieve meta-data for %s:%s:%s",
standard, provider, type);
services_action_free(action);
return -EIO;
}
if (!action->stdout_data) {
crm_err("Failed to receive meta-data for %s:%s:%s",
standard, provider, type);
services_action_free(action);
return -EIO;
}
*output = strdup(action->stdout_data);
services_action_free(action);
return pcmk_ok;
}
static int
lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action,
const char *userdata, guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
enum lrmd_call_options options, lrmd_key_value_t * params)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
xmlNode *args = create_xml_node(data, XML_TAG_ATTRS);
lrmd_key_value_t *tmp = NULL;
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
crm_xml_add(data, F_LRMD_RSC_ACTION, action);
crm_xml_add(data, F_LRMD_RSC_USERDATA_STR, userdata);
crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms);
crm_xml_add_int(data, F_LRMD_TIMEOUT, timeout);
crm_xml_add_int(data, F_LRMD_RSC_START_DELAY, start_delay);
for (tmp = params; tmp; tmp = tmp->next) {
hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args);
}
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, TRUE);
free_xml(data);
lrmd_key_value_freeall(params);
return rc;
}
/* timeout is in ms */
static int
lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path,
int timeout, lrmd_key_value_t *params)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_ALERT);
xmlNode *args = create_xml_node(data, XML_TAG_ATTRS);
lrmd_key_value_t *tmp = NULL;
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_ALERT_ID, alert_id);
crm_xml_add(data, F_LRMD_ALERT_PATH, alert_path);
crm_xml_add_int(data, F_LRMD_TIMEOUT, timeout);
for (tmp = params; tmp; tmp = tmp->next) {
hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args);
}
rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout,
lrmd_opt_notify_orig_only, TRUE);
free_xml(data);
lrmd_key_value_freeall(params);
return rc;
}
static int
lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action,
guint interval_ms)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, F_LRMD_ORIGIN, __FUNCTION__);
crm_xml_add(data, F_LRMD_RSC_ACTION, action);
crm_xml_add(data, F_LRMD_RSC_ID, rsc_id);
crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, TRUE);
free_xml(data);
return rc;
}
static int
list_stonith_agents(lrmd_list_t ** resources)
{
int rc = 0;
stonith_t *stonith_api = stonith_api_new();
stonith_key_value_t *stonith_resources = NULL;
stonith_key_value_t *dIter = NULL;
if (stonith_api == NULL) {
crm_err("Could not list fence agents: API memory allocation failed");
return -ENOMEM;
}
stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL,
&stonith_resources, 0);
stonith_api->cmds->free(stonith_api);
for (dIter = stonith_resources; dIter; dIter = dIter->next) {
rc++;
if (resources) {
*resources = lrmd_list_add(*resources, dIter->value);
}
}
stonith_key_value_freeall(stonith_resources, 1, 0);
return rc;
}
static int
lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class,
const char *provider)
{
int rc = 0;
int stonith_count = 0; // Initially, whether to include stonith devices
if (safe_str_eq(class, PCMK_RESOURCE_CLASS_STONITH)) {
stonith_count = 1;
} else {
GListPtr gIter = NULL;
GList *agents = resources_list_agents(class, provider);
for (gIter = agents; gIter != NULL; gIter = gIter->next) {
*resources = lrmd_list_add(*resources, (const char *)gIter->data);
rc++;
}
g_list_free_full(agents, free);
if (!class) {
stonith_count = 1;
}
}
if (stonith_count) {
// Now, if stonith devices are included, how many there are
stonith_count = list_stonith_agents(resources);
if (stonith_count > 0) {
rc += stonith_count;
}
}
if (rc == 0) {
crm_notice("No agents found for class %s", class);
rc = -EPROTONOSUPPORT;
}
return rc;
}
static int
does_provider_have_agent(const char *agent, const char *provider, const char *class)
{
int found = 0;
GList *agents = NULL;
GListPtr gIter2 = NULL;
agents = resources_list_agents(class, provider);
for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) {
if (safe_str_eq(agent, gIter2->data)) {
found = 1;
}
}
g_list_free_full(agents, free);
return found;
}
static int
lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers)
{
int rc = pcmk_ok;
char *provider = NULL;
GList *ocf_providers = NULL;
GListPtr gIter = NULL;
ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF);
for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) {
provider = gIter->data;
if (!agent || does_provider_have_agent(agent, provider,
PCMK_RESOURCE_CLASS_OCF)) {
*providers = lrmd_list_add(*providers, (const char *)gIter->data);
rc++;
}
}
g_list_free_full(ocf_providers, free);
return rc;
}
static int
lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported)
{
int rc = 0;
GList *standards = NULL;
GListPtr gIter = NULL;
standards = resources_list_standards();
for (gIter = standards; gIter != NULL; gIter = gIter->next) {
*supported = lrmd_list_add(*supported, (const char *)gIter->data);
rc++;
}
if (list_stonith_agents(NULL) > 0) {
*supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH);
rc++;
}
g_list_free_full(standards, free);
return rc;
}
lrmd_t *
lrmd_api_new(void)
{
lrmd_t *new_lrmd = NULL;
lrmd_private_t *pvt = NULL;
new_lrmd = calloc(1, sizeof(lrmd_t));
pvt = calloc(1, sizeof(lrmd_private_t));
pvt->remote = calloc(1, sizeof(crm_remote_t));
new_lrmd->cmds = calloc(1, sizeof(lrmd_api_operations_t));
pvt->type = CRM_CLIENT_IPC;
new_lrmd->lrmd_private = pvt;
new_lrmd->cmds->connect = lrmd_api_connect;
new_lrmd->cmds->connect_async = lrmd_api_connect_async;
new_lrmd->cmds->is_connected = lrmd_api_is_connected;
new_lrmd->cmds->poke_connection = lrmd_api_poke_connection;
new_lrmd->cmds->disconnect = lrmd_api_disconnect;
new_lrmd->cmds->register_rsc = lrmd_api_register_rsc;
new_lrmd->cmds->unregister_rsc = lrmd_api_unregister_rsc;
new_lrmd->cmds->get_rsc_info = lrmd_api_get_rsc_info;
new_lrmd->cmds->get_recurring_ops = lrmd_api_get_recurring_ops;
new_lrmd->cmds->set_callback = lrmd_api_set_callback;
new_lrmd->cmds->get_metadata = lrmd_api_get_metadata;
new_lrmd->cmds->exec = lrmd_api_exec;
new_lrmd->cmds->cancel = lrmd_api_cancel;
new_lrmd->cmds->list_agents = lrmd_api_list_agents;
new_lrmd->cmds->list_ocf_providers = lrmd_api_list_ocf_providers;
new_lrmd->cmds->list_standards = lrmd_api_list_standards;
new_lrmd->cmds->exec_alert = lrmd_api_exec_alert;
new_lrmd->cmds->get_metadata_params = lrmd_api_get_metadata_params;
return new_lrmd;
}
lrmd_t *
lrmd_remote_api_new(const char *nodename, const char *server, int port)
{
#ifdef HAVE_GNUTLS_GNUTLS_H
lrmd_t *new_lrmd = lrmd_api_new();
lrmd_private_t *native = new_lrmd->lrmd_private;
if (!nodename && !server) {
lrmd_api_delete(new_lrmd);
return NULL;
}
native->type = CRM_CLIENT_TLS;
native->remote_nodename = nodename ? strdup(nodename) : strdup(server);
native->server = server ? strdup(server) : strdup(nodename);
native->port = port;
if (native->port == 0) {
native->port = crm_default_remote_port();
}
return new_lrmd;
#else
crm_err("Cannot communicate with Pacemaker Remote because GnuTLS is not enabled for this build");
return NULL;
#endif
}
void
lrmd_api_delete(lrmd_t * lrmd)
{
if (!lrmd) {
return;
}
lrmd->cmds->disconnect(lrmd); /* no-op if already disconnected */
free(lrmd->cmds);
if (lrmd->lrmd_private) {
lrmd_private_t *native = lrmd->lrmd_private;
#ifdef HAVE_GNUTLS_GNUTLS_H
free(native->server);
#endif
free(native->remote_nodename);
free(native->remote);
free(native->token);
free(native->peer_version);
}
free(lrmd->lrmd_private);
free(lrmd);
}
diff --git a/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c
index fc2f4cfc56..195d055fbf 100644
--- a/lib/pacemaker/pcmk_sched_allocate.c
+++ b/lib/pacemaker/pcmk_sched_allocate.c
@@ -1,2969 +1,3072 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
CRM_TRACE_INIT_DATA(pacemaker);
void set_alloc_actions(pe_working_set_t * data_set);
extern void ReloadRsc(resource_t * rsc, node_t *node, pe_working_set_t * data_set);
extern gboolean DeleteRsc(resource_t * rsc, node_t * node, gboolean optional, pe_working_set_t * data_set);
static void apply_remote_node_ordering(pe_working_set_t *data_set);
static enum remote_connection_state get_remote_node_state(pe_node_t *node);
enum remote_connection_state {
remote_state_unknown = 0,
remote_state_alive = 1,
remote_state_resting = 2,
remote_state_failed = 3,
remote_state_stopped = 4
};
static const char *
state2text(enum remote_connection_state state)
{
switch (state) {
case remote_state_unknown:
return "unknown";
case remote_state_alive:
return "alive";
case remote_state_resting:
return "resting";
case remote_state_failed:
return "failed";
case remote_state_stopped:
return "stopped";
}
return "impossible";
}
resource_alloc_functions_t resource_class_alloc_functions[] = {
{
native_merge_weights,
native_color,
native_create_actions,
native_create_probe,
native_internal_constraints,
native_rsc_colocation_lh,
native_rsc_colocation_rh,
native_rsc_location,
native_action_flags,
native_update_actions,
native_expand,
native_append_meta,
},
{
group_merge_weights,
group_color,
group_create_actions,
native_create_probe,
group_internal_constraints,
group_rsc_colocation_lh,
group_rsc_colocation_rh,
group_rsc_location,
group_action_flags,
group_update_actions,
group_expand,
group_append_meta,
},
{
clone_merge_weights,
clone_color,
clone_create_actions,
clone_create_probe,
clone_internal_constraints,
clone_rsc_colocation_lh,
clone_rsc_colocation_rh,
clone_rsc_location,
clone_action_flags,
pcmk__multi_update_actions,
clone_expand,
clone_append_meta,
},
{
pcmk__bundle_merge_weights,
pcmk__bundle_color,
pcmk__bundle_create_actions,
pcmk__bundle_create_probe,
pcmk__bundle_internal_constraints,
pcmk__bundle_rsc_colocation_lh,
pcmk__bundle_rsc_colocation_rh,
pcmk__bundle_rsc_location,
pcmk__bundle_action_flags,
pcmk__multi_update_actions,
pcmk__bundle_expand,
pcmk__bundle_append_meta,
}
};
gboolean
update_action_flags(action_t * action, enum pe_action_flags flags, const char *source, int line)
{
static unsigned long calls = 0;
gboolean changed = FALSE;
gboolean clear = is_set(flags, pe_action_clear);
enum pe_action_flags last = action->flags;
if (clear) {
action->flags = crm_clear_bit(source, line, action->uuid, action->flags, flags);
} else {
action->flags = crm_set_bit(source, line, action->uuid, action->flags, flags);
}
if (last != action->flags) {
calls++;
changed = TRUE;
/* Useful for tracking down _who_ changed a specific flag */
/* CRM_ASSERT(calls != 534); */
clear_bit(flags, pe_action_clear);
crm_trace("%s on %s: %sset flags 0x%.6x (was 0x%.6x, now 0x%.6x, %lu, %s)",
action->uuid, action->node ? action->node->details->uname : "[none]",
clear ? "un-" : "", flags, last, action->flags, calls, source);
}
return changed;
}
static gboolean
check_rsc_parameters(resource_t * rsc, node_t * node, xmlNode * rsc_entry,
gboolean active_here, pe_working_set_t * data_set)
{
int attr_lpc = 0;
gboolean force_restart = FALSE;
gboolean delete_resource = FALSE;
gboolean changed = FALSE;
const char *value = NULL;
const char *old_value = NULL;
const char *attr_list[] = {
XML_ATTR_TYPE,
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER
};
for (; attr_lpc < DIMOF(attr_list); attr_lpc++) {
value = crm_element_value(rsc->xml, attr_list[attr_lpc]);
old_value = crm_element_value(rsc_entry, attr_list[attr_lpc]);
if (value == old_value /* i.e. NULL */
|| crm_str_eq(value, old_value, TRUE)) {
continue;
}
changed = TRUE;
trigger_unfencing(rsc, node, "Device definition changed", NULL, data_set);
if (active_here) {
force_restart = TRUE;
crm_notice("Forcing restart of %s on %s, %s changed: %s -> %s",
rsc->id, node->details->uname, attr_list[attr_lpc],
crm_str(old_value), crm_str(value));
}
}
if (force_restart) {
/* make sure the restart happens */
stop_action(rsc, node, FALSE);
set_bit(rsc->flags, pe_rsc_start_pending);
delete_resource = TRUE;
} else if (changed) {
delete_resource = TRUE;
}
return delete_resource;
}
static void
CancelXmlOp(resource_t * rsc, xmlNode * xml_op, node_t * active_node,
const char *reason, pe_working_set_t * data_set)
{
guint interval_ms = 0;
action_t *cancel = NULL;
const char *task = NULL;
const char *call_id = NULL;
const char *interval_ms_s = NULL;
CRM_CHECK(xml_op != NULL, return);
CRM_CHECK(active_node != NULL, return);
task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
call_id = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
interval_ms_s = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS);
interval_ms = crm_parse_ms(interval_ms_s);
crm_info("Action " CRM_OP_FMT " on %s will be stopped: %s",
rsc->id, task, interval_ms,
active_node->details->uname, (reason? reason : "unknown"));
cancel = pe_cancel_op(rsc, task, interval_ms, active_node, data_set);
add_hash_param(cancel->meta, XML_LRM_ATTR_CALLID, call_id);
custom_action_order(rsc, stop_key(rsc), NULL, rsc, NULL, cancel, pe_order_optional, data_set);
}
static gboolean
check_action_definition(resource_t * rsc, node_t * active_node, xmlNode * xml_op,
pe_working_set_t * data_set)
{
char *key = NULL;
guint interval_ms = 0;
const char *interval_ms_s = NULL;
const op_digest_cache_t *digest_data = NULL;
gboolean did_change = FALSE;
const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
const char *digest_secure = NULL;
CRM_CHECK(active_node != NULL, return FALSE);
interval_ms_s = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS);
interval_ms = crm_parse_ms(interval_ms_s);
if (interval_ms > 0) {
xmlNode *op_match = NULL;
/* we need to reconstruct the key because of the way we used to construct resource IDs */
key = generate_op_key(rsc->id, task, interval_ms);
pe_rsc_trace(rsc, "Checking parameters for %s", key);
op_match = find_rsc_op_entry(rsc, key);
if (op_match == NULL && is_set(data_set->flags, pe_flag_stop_action_orphans)) {
CancelXmlOp(rsc, xml_op, active_node, "orphan", data_set);
free(key);
return TRUE;
} else if (op_match == NULL) {
pe_rsc_debug(rsc, "Orphan action detected: %s on %s", key, active_node->details->uname);
free(key);
return TRUE;
}
free(key);
key = NULL;
}
crm_trace("Testing " CRM_OP_FMT " on %s",
rsc->id, task, interval_ms, active_node->details->uname);
if ((interval_ms == 0) && safe_str_eq(task, RSC_STATUS)) {
/* Reload based on the start action not a probe */
task = RSC_START;
} else if ((interval_ms == 0) && safe_str_eq(task, RSC_MIGRATED)) {
/* Reload based on the start action not a migrate */
task = RSC_START;
} else if ((interval_ms == 0) && safe_str_eq(task, RSC_PROMOTE)) {
/* Reload based on the start action not a promote */
task = RSC_START;
}
digest_data = rsc_action_digest_cmp(rsc, xml_op, active_node, data_set);
if(is_set(data_set->flags, pe_flag_sanitized)) {
digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST);
}
if(digest_data->rc != RSC_DIGEST_MATCH
&& digest_secure
&& digest_data->digest_secure_calc
&& strcmp(digest_data->digest_secure_calc, digest_secure) == 0) {
if (is_set(data_set->flags, pe_flag_stdout)) {
printf("Only 'private' parameters to " CRM_OP_FMT " on %s changed: %s\n",
rsc->id, task, interval_ms, active_node->details->uname,
crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC));
}
} else if (digest_data->rc == RSC_DIGEST_RESTART) {
/* Changes that force a restart */
pe_action_t *required = NULL;
did_change = TRUE;
key = generate_op_key(rsc->id, task, interval_ms);
crm_log_xml_info(digest_data->params_restart, "params:restart");
required = custom_action(rsc, key, task, NULL, TRUE, TRUE, data_set);
pe_action_set_flag_reason(__FUNCTION__, __LINE__, required, NULL,
"resource definition change", pe_action_optional, TRUE);
trigger_unfencing(rsc, active_node, "Device parameters changed", NULL, data_set);
} else if ((digest_data->rc == RSC_DIGEST_ALL) || (digest_data->rc == RSC_DIGEST_UNKNOWN)) {
/* Changes that can potentially be handled by a reload */
const char *digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST);
did_change = TRUE;
trigger_unfencing(rsc, active_node, "Device parameters changed (reload)", NULL, data_set);
crm_log_xml_info(digest_data->params_all, "params:reload");
key = generate_op_key(rsc->id, task, interval_ms);
if (interval_ms > 0) {
action_t *op = NULL;
#if 0
/* Always reload/restart the entire resource */
ReloadRsc(rsc, active_node, data_set);
#else
/* Re-sending the recurring op is sufficient - the old one will be cancelled automatically */
op = custom_action(rsc, key, task, active_node, TRUE, TRUE, data_set);
set_bit(op->flags, pe_action_reschedule);
#endif
} else if (digest_restart) {
pe_rsc_trace(rsc, "Reloading '%s' action for resource %s", task, rsc->id);
/* Reload this resource */
ReloadRsc(rsc, active_node, data_set);
free(key);
} else {
pe_action_t *required = NULL;
pe_rsc_trace(rsc, "Resource %s doesn't know how to reload", rsc->id);
/* Re-send the start/demote/promote op
* Recurring ops will be detected independently
*/
required = custom_action(rsc, key, task, NULL, TRUE, TRUE, data_set);
pe_action_set_flag_reason(__FUNCTION__, __LINE__, required, NULL,
"resource definition change", pe_action_optional, TRUE);
}
}
return did_change;
}
/*!
* \internal
* \brief Do deferred action checks after allocation
*
* \param[in] data_set Working set for cluster
*/
static void
check_params(pe_resource_t *rsc, pe_node_t *node, xmlNode *rsc_op,
enum pe_check_parameters check, pe_working_set_t *data_set)
{
const char *reason = NULL;
op_digest_cache_t *digest_data = NULL;
switch (check) {
case pe_check_active:
if (check_action_definition(rsc, node, rsc_op, data_set)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
reason = "action definition changed";
}
break;
case pe_check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, data_set);
switch (digest_data->rc) {
case RSC_DIGEST_UNKNOWN:
crm_trace("Resource %s history entry %s on %s has no digest to compare",
rsc->id, ID(rsc_op), node->details->id);
break;
case RSC_DIGEST_MATCH:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason) {
pe__clear_failcount(rsc, node, reason, data_set);
}
}
static void
check_actions_for(xmlNode * rsc_entry, resource_t * rsc, node_t * node, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
int offset = -1;
guint interval_ms = 0;
int stop_index = 0;
int start_index = 0;
const char *task = NULL;
const char *interval_ms_s = NULL;
xmlNode *rsc_op = NULL;
GListPtr op_list = NULL;
GListPtr sorted_op_list = NULL;
CRM_CHECK(node != NULL, return);
if (is_set(rsc->flags, pe_rsc_orphan)) {
resource_t *parent = uber_parent(rsc);
if(parent == NULL
|| pe_rsc_is_clone(parent) == FALSE
|| is_set(parent->flags, pe_rsc_unique)) {
pe_rsc_trace(rsc, "Skipping param check for %s and deleting: orphan", rsc->id);
DeleteRsc(rsc, node, FALSE, data_set);
} else {
pe_rsc_trace(rsc, "Skipping param check for %s (orphan clone)", rsc->id);
}
return;
} else if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) {
if (check_rsc_parameters(rsc, node, rsc_entry, FALSE, data_set)) {
DeleteRsc(rsc, node, FALSE, data_set);
}
pe_rsc_trace(rsc, "Skipping param check for %s: no longer active on %s",
rsc->id, node->details->uname);
return;
}
pe_rsc_trace(rsc, "Processing %s on %s", rsc->id, node->details->uname);
if (check_rsc_parameters(rsc, node, rsc_entry, TRUE, data_set)) {
DeleteRsc(rsc, node, FALSE, data_set);
}
for (rsc_op = __xml_first_child_element(rsc_entry); rsc_op != NULL;
rsc_op = __xml_next_element(rsc_op)) {
if (crm_str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, TRUE)) {
op_list = g_list_prepend(op_list, rsc_op);
}
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
offset++;
if (start_index < stop_index) {
/* stopped */
continue;
} else if (offset < start_index) {
/* action occurred prior to a start */
continue;
}
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
interval_ms_s = crm_element_value(rsc_op, XML_LRM_ATTR_INTERVAL_MS);
interval_ms = crm_parse_ms(interval_ms_s);
if ((interval_ms > 0) &&
(is_set(rsc->flags, pe_rsc_maintenance) || node->details->maintenance)) {
// Maintenance mode cancels recurring operations
CancelXmlOp(rsc, rsc_op, node, "maintenance mode", data_set);
} else if ((interval_ms > 0)
|| safe_str_eq(task, RSC_STATUS)
|| safe_str_eq(task, RSC_START)
|| safe_str_eq(task, RSC_PROMOTE)
|| safe_str_eq(task, RSC_MIGRATED)) {
/* If a resource operation failed, and the operation's definition
* has changed, clear any fail count so they can be retried fresh.
*/
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources to nodes yet, so if the
* REMOTE_CONTAINER_HACK is used, we may calculate the digest
* based on the literal "#uname" value rather than the properly
* substituted value. That would mistakenly make the action
* definition appear to have been changed. Defer the check until
* later in this case.
*/
pe__add_param_check(rsc_op, rsc, node, pe_check_active,
data_set);
} else if (check_action_definition(rsc, node, rsc_op, data_set)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
pe__clear_failcount(rsc, node, "action definition changed",
data_set);
}
}
}
g_list_free(sorted_op_list);
}
static GListPtr
find_rsc_list(GListPtr result, resource_t * rsc, const char *id, gboolean renamed_clones,
gboolean partial, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
gboolean match = FALSE;
if (id == NULL) {
return NULL;
}
if (rsc == NULL) {
if (data_set == NULL) {
return NULL;
}
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
result = find_rsc_list(result, child, id, renamed_clones, partial,
NULL);
}
return result;
}
if (partial) {
if (strstr(rsc->id, id)) {
match = TRUE;
} else if (renamed_clones && rsc->clone_name && strstr(rsc->clone_name, id)) {
match = TRUE;
}
} else {
if (strcmp(rsc->id, id) == 0) {
match = TRUE;
} else if (renamed_clones && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) {
match = TRUE;
}
}
if (match) {
result = g_list_prepend(result, rsc);
}
if (rsc->children) {
gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
result = find_rsc_list(result, child, id, renamed_clones, partial, NULL);
}
}
return result;
}
static void
check_actions(pe_working_set_t * data_set)
{
const char *id = NULL;
node_t *node = NULL;
xmlNode *lrm_rscs = NULL;
xmlNode *status = get_object_root(XML_CIB_TAG_STATUS, data_set->input);
xmlNode *node_state = NULL;
for (node_state = __xml_first_child_element(status); node_state != NULL;
node_state = __xml_next_element(node_state)) {
if (crm_str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, TRUE)) {
id = crm_element_value(node_state, XML_ATTR_ID);
lrm_rscs = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
lrm_rscs = find_xml_node(lrm_rscs, XML_LRM_TAG_RESOURCES, FALSE);
node = pe_find_node_id(data_set->nodes, id);
if (node == NULL) {
continue;
/* Still need to check actions for a maintenance node to cancel existing monitor operations */
} else if (can_run_resources(node) == FALSE && node->details->maintenance == FALSE) {
crm_trace("Skipping param check for %s: can't run resources",
node->details->uname);
continue;
}
crm_trace("Processing node %s", node->details->uname);
if (node->details->online || is_set(data_set->flags, pe_flag_stonith_enabled)) {
xmlNode *rsc_entry = NULL;
for (rsc_entry = __xml_first_child_element(lrm_rscs);
rsc_entry != NULL;
rsc_entry = __xml_next_element(rsc_entry)) {
if (crm_str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, TRUE)) {
if (xml_has_children(rsc_entry)) {
GListPtr gIter = NULL;
GListPtr result = NULL;
const char *rsc_id = ID(rsc_entry);
CRM_CHECK(rsc_id != NULL, return);
result = find_rsc_list(NULL, NULL, rsc_id, TRUE, FALSE, data_set);
for (gIter = result; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
if (rsc->variant != pe_native) {
continue;
}
check_actions_for(rsc_entry, rsc, node, data_set);
}
g_list_free(result);
}
}
}
}
}
}
}
static void
apply_placement_constraints(pe_working_set_t * data_set)
{
for (GList *gIter = data_set->placement_constraints;
gIter != NULL; gIter = gIter->next) {
pe__location_t *cons = gIter->data;
cons->rsc_lh->cmds->rsc_location(cons->rsc_lh, cons);
}
}
static gboolean
failcount_clear_action_exists(node_t * node, resource_t * rsc)
{
gboolean rc = FALSE;
GList *list = pe__resource_actions(rsc, node, CRM_OP_CLEAR_FAILCOUNT, TRUE);
if (list) {
rc = TRUE;
}
g_list_free(list);
return rc;
}
/*!
* \internal
* \brief Force resource away if failures hit migration threshold
*
* \param[in,out] rsc Resource to check for failures
* \param[in,out] node Node to check for failures
* \param[in,out] data_set Cluster working set to update
*/
static void
check_migration_threshold(resource_t *rsc, node_t *node,
pe_working_set_t *data_set)
{
int fail_count, countdown;
resource_t *failed;
/* Migration threshold of 0 means never force away */
if (rsc->migration_threshold == 0) {
return;
}
// If we're ignoring failures, also ignore the migration threshold
if (is_set(rsc->flags, pe_rsc_failure_ignored)) {
return;
}
/* If there are no failures, there's no need to force away */
fail_count = pe_get_failcount(node, rsc, NULL,
pe_fc_effective|pe_fc_fillers, NULL,
data_set);
if (fail_count <= 0) {
return;
}
/* How many more times recovery will be tried on this node */
countdown = QB_MAX(rsc->migration_threshold - fail_count, 0);
/* If failed resource has a parent, we'll force the parent away */
failed = rsc;
if (is_not_set(rsc->flags, pe_rsc_unique)) {
failed = uber_parent(rsc);
}
if (countdown == 0) {
resource_location(failed, node, -INFINITY, "__fail_limit__", data_set);
crm_warn("Forcing %s away from %s after %d failures (max=%d)",
failed->id, node->details->uname, fail_count,
rsc->migration_threshold);
} else {
crm_info("%s can fail %d more times on %s before being forced off",
failed->id, countdown, node->details->uname);
}
}
static void
common_apply_stickiness(resource_t * rsc, node_t * node, pe_working_set_t * data_set)
{
if (rsc->children) {
GListPtr gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
common_apply_stickiness(child_rsc, node, data_set);
}
return;
}
if (is_set(rsc->flags, pe_rsc_managed)
&& rsc->stickiness != 0 && pcmk__list_of_1(rsc->running_on)) {
node_t *current = pe_find_node_id(rsc->running_on, node->details->id);
node_t *match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (current == NULL) {
} else if (match != NULL || is_set(data_set->flags, pe_flag_symmetric_cluster)) {
resource_t *sticky_rsc = rsc;
resource_location(sticky_rsc, node, rsc->stickiness, "stickiness", data_set);
pe_rsc_debug(sticky_rsc, "Resource %s: preferring current location"
" (node=%s, weight=%d)", sticky_rsc->id,
node->details->uname, rsc->stickiness);
} else {
GHashTableIter iter;
node_t *nIter = NULL;
pe_rsc_debug(rsc, "Ignoring stickiness for %s: the cluster is asymmetric"
" and node %s is not explicitly allowed", rsc->id, node->details->uname);
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&nIter)) {
crm_err("%s[%s] = %d", rsc->id, nIter->details->uname, nIter->weight);
}
}
}
/* Check the migration threshold only if a failcount clear action
* has not already been placed for this resource on the node.
* There is no sense in potentially forcing the resource from this
* node if the failcount is being reset anyway.
*
* @TODO A clear_failcount operation can be scheduled in stage4() via
* check_actions_for(), or in stage5() via check_params(). This runs in
* stage2(), so it cannot detect those, meaning we might check the migration
* threshold when we shouldn't -- worst case, we stop or move the resource,
* then move it back next transition.
*/
if (failcount_clear_action_exists(node, rsc) == FALSE) {
check_migration_threshold(rsc, node, data_set);
}
}
void
complex_set_cmds(resource_t * rsc)
{
GListPtr gIter = rsc->children;
rsc->cmds = &resource_class_alloc_functions[rsc->variant];
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
complex_set_cmds(child_rsc);
}
}
void
set_alloc_actions(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
complex_set_cmds(rsc);
}
}
static void
calculate_system_health(gpointer gKey, gpointer gValue, gpointer user_data)
{
const char *key = (const char *)gKey;
const char *value = (const char *)gValue;
int *system_health = (int *)user_data;
if (!gKey || !gValue || !user_data) {
return;
}
if (crm_starts_with(key, "#health")) {
int score;
/* Convert the value into an integer */
score = char2score(value);
/* Add it to the running total */
*system_health = merge_weights(score, *system_health);
}
}
static gboolean
apply_system_health(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
const char *health_strategy = pe_pref(data_set->config_hash, "node-health-strategy");
int base_health = 0;
if (health_strategy == NULL || safe_str_eq(health_strategy, "none")) {
/* Prevent any accidental health -> score translation */
node_score_red = 0;
node_score_yellow = 0;
node_score_green = 0;
return TRUE;
} else if (safe_str_eq(health_strategy, "migrate-on-red")) {
/* Resources on nodes which have health values of red are
* weighted away from that node.
*/
node_score_red = -INFINITY;
node_score_yellow = 0;
node_score_green = 0;
} else if (safe_str_eq(health_strategy, "only-green")) {
/* Resources on nodes which have health values of red or yellow
* are forced away from that node.
*/
node_score_red = -INFINITY;
node_score_yellow = -INFINITY;
node_score_green = 0;
} else if (safe_str_eq(health_strategy, "progressive")) {
/* Same as the above, but use the r/y/g scores provided by the user
* Defaults are provided by the pe_prefs table
* Also, custom health "base score" can be used
*/
base_health = crm_parse_int(pe_pref(data_set->config_hash, "node-health-base"), "0");
} else if (safe_str_eq(health_strategy, "custom")) {
/* Requires the admin to configure the rsc_location constaints for
* processing the stored health scores
*/
/* TODO: Check for the existence of appropriate node health constraints */
return TRUE;
} else {
crm_err("Unknown node health strategy: %s", health_strategy);
return FALSE;
}
crm_info("Applying automated node health strategy: %s", health_strategy);
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
int system_health = base_health;
node_t *node = (node_t *) gIter->data;
/* Search through the node hash table for system health entries. */
g_hash_table_foreach(node->details->attrs, calculate_system_health, &system_health);
crm_info(" Node %s has an combined system health of %d",
node->details->uname, system_health);
/* If the health is non-zero, then create a new rsc2node so that the
* weight will be added later on.
*/
if (system_health != 0) {
GListPtr gIter2 = data_set->resources;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
resource_t *rsc = (resource_t *) gIter2->data;
rsc2node_new(health_strategy, rsc, system_health, NULL, node, data_set);
}
}
}
return TRUE;
}
gboolean
stage0(pe_working_set_t * data_set)
{
xmlNode *cib_constraints = get_object_root(XML_CIB_TAG_CONSTRAINTS, data_set->input);
if (data_set->input == NULL) {
return FALSE;
}
if (is_set(data_set->flags, pe_flag_have_status) == FALSE) {
crm_trace("Calculating status");
cluster_status(data_set);
}
set_alloc_actions(data_set);
apply_system_health(data_set);
unpack_constraints(cib_constraints, data_set);
return TRUE;
}
/*
* Check nodes for resources started outside of the LRM
*/
gboolean
probe_resources(pe_working_set_t * data_set)
{
action_t *probe_node_complete = NULL;
for (GListPtr gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
const char *probed = pe_node_attribute_raw(node, CRM_OP_PROBED);
if (node->details->online == FALSE) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& (get_remote_node_state(node) == remote_state_failed)) {
pe_fence_node(data_set, node, "the connection is unrecoverable");
}
continue;
} else if (node->details->unclean) {
continue;
} else if (node->details->rsc_discovery_enabled == FALSE) {
/* resource discovery is disabled for this node */
continue;
}
if (probed != NULL && crm_is_true(probed) == FALSE) {
action_t *probe_op = custom_action(NULL, crm_strdup_printf("%s-%s", CRM_OP_REPROBE, node->details->uname),
CRM_OP_REPROBE, node, FALSE, TRUE, data_set);
add_hash_param(probe_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE);
continue;
}
for (GListPtr gIter2 = data_set->resources; gIter2 != NULL; gIter2 = gIter2->next) {
resource_t *rsc = (resource_t *) gIter2->data;
rsc->cmds->create_probe(rsc, node, probe_node_complete, FALSE, data_set);
}
}
return TRUE;
}
static void
rsc_discover_filter(resource_t *rsc, node_t *node)
{
GListPtr gIter = rsc->children;
resource_t *top = uber_parent(rsc);
node_t *match;
if (rsc->exclusive_discover == FALSE && top->exclusive_discover == FALSE) {
return;
}
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
rsc_discover_filter(child_rsc, node);
}
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (match && match->rsc_discover_mode != pe_discover_exclusive) {
match->weight = -INFINITY;
}
}
+static time_t
+shutdown_time(pe_node_t *node, pe_working_set_t *data_set)
+{
+ const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
+ time_t result = 0;
+
+ if (shutdown) {
+ errno = 0;
+ result = (time_t) crm_int_helper(shutdown, NULL);
+ if (errno != 0) {
+ result = 0;
+ }
+ }
+ return result? result : get_effective_time(data_set);
+}
+
+static void
+apply_shutdown_lock(pe_resource_t *rsc, pe_working_set_t *data_set)
+{
+ const char *class;
+
+ // Only primitives and (uncloned) groups may be locked
+ if (rsc->variant == pe_group) {
+ for (GList *item = rsc->children; item != NULL;
+ item = item->next) {
+ apply_shutdown_lock((pe_resource_t *) item->data, data_set);
+ }
+ } else if (rsc->variant != pe_native) {
+ return;
+ }
+
+ // Fence devices and remote connections can't be locked
+ class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
+ if ((class == NULL) || !strcmp(class, PCMK_RESOURCE_CLASS_STONITH)
+ || pe__resource_is_remote_conn(rsc, data_set)) {
+ return;
+ }
+
+ if (rsc->lock_node != NULL) {
+ // The lock was obtained from resource history
+
+ if (rsc->running_on != NULL) {
+ /* The resource was started elsewhere even though it is now
+ * considered locked. This shouldn't be possible, but as a
+ * failsafe, we don't want to disturb the resource now.
+ */
+ pe_rsc_info(rsc,
+ "Cancelling shutdown lock because %s is already active",
+ rsc->id);
+ pe__clear_resource_history(rsc, rsc->lock_node, data_set);
+ rsc->lock_node = NULL;
+ rsc->lock_time = 0;
+ }
+
+ // Only a resource active on exactly one node can be locked
+ } else if (pcmk__list_of_1(rsc->running_on)) {
+ pe_node_t *node = rsc->running_on->data;
+
+ if (node->details->shutdown) {
+ if (node->details->unclean) {
+ pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown",
+ rsc->id, node->details->uname);
+ } else {
+ rsc->lock_node = node;
+ rsc->lock_time = shutdown_time(node, data_set);
+ }
+ }
+ }
+
+ if (rsc->lock_node == NULL) {
+ // No lock needed
+ return;
+ }
+
+ if (data_set->shutdown_lock > 0) {
+ time_t lock_expiration = rsc->lock_time + data_set->shutdown_lock;
+
+ pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
+ rsc->id, rsc->lock_node->details->uname,
+ (long long) lock_expiration);
+ pe__update_recheck_time(++lock_expiration, data_set);
+ } else {
+ pe_rsc_info(rsc, "Locking %s to %s due to shutdown",
+ rsc->id, rsc->lock_node->details->uname);
+ }
+
+ // If resource is locked to one node, ban it from all other nodes
+ for (GList *item = data_set->nodes; item != NULL; item = item->next) {
+ pe_node_t *node = item->data;
+
+ if (strcmp(node->details->uname, rsc->lock_node->details->uname)) {
+ resource_location(rsc, node, -CRM_SCORE_INFINITY,
+ XML_CONFIG_ATTR_SHUTDOWN_LOCK, data_set);
+ }
+ }
+}
+
/*
* Count how many valid nodes we have (so we know the maximum number of
* colors we can resolve).
*
* Apply node constraints (i.e. filter the "allowed_nodes" part of resources)
*/
gboolean
stage2(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
+ if (is_set(data_set->flags, pe_flag_shutdown_lock)) {
+ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
+ apply_shutdown_lock((pe_resource_t *) gIter->data, data_set);
+ }
+ }
+
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
if (node == NULL) {
/* error */
} else if (node->weight >= 0.0 /* global weight */
&& node->details->online && node->details->type != node_ping) {
data_set->max_valid_nodes++;
}
}
apply_placement_constraints(data_set);
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
GListPtr gIter2 = NULL;
node_t *node = (node_t *) gIter->data;
gIter2 = data_set->resources;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
resource_t *rsc = (resource_t *) gIter2->data;
common_apply_stickiness(rsc, node, data_set);
rsc_discover_filter(rsc, node);
}
}
return TRUE;
}
/*
* Create internal resource constraints before allocation
*/
gboolean
stage3(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
rsc->cmds->internal_constraints(rsc, data_set);
}
return TRUE;
}
/*
* Check for orphaned or redefined actions
*/
gboolean
stage4(pe_working_set_t * data_set)
{
check_actions(data_set);
return TRUE;
}
static void *
convert_const_pointer(const void *ptr)
{
/* Worst function ever */
return (void *)ptr;
}
static gint
sort_rsc_process_order(gconstpointer a, gconstpointer b, gpointer data)
{
int rc = 0;
int r1_weight = -INFINITY;
int r2_weight = -INFINITY;
const char *reason = "existence";
const GListPtr nodes = (GListPtr) data;
const resource_t *resource1 = a;
const resource_t *resource2 = b;
node_t *r1_node = NULL;
node_t *r2_node = NULL;
GListPtr gIter = NULL;
GHashTable *r1_nodes = NULL;
GHashTable *r2_nodes = NULL;
if (a == NULL && b == NULL) {
goto done;
}
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
reason = "priority";
r1_weight = resource1->priority;
r2_weight = resource2->priority;
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
reason = "no node list";
if (nodes == NULL) {
goto done;
}
r1_nodes = rsc_merge_weights(convert_const_pointer(resource1),
resource1->id, NULL, NULL, 1,
pe_weights_forward | pe_weights_init);
dump_node_scores(LOG_TRACE, NULL, resource1->id, r1_nodes);
r2_nodes = rsc_merge_weights(convert_const_pointer(resource2),
resource2->id, NULL, NULL, 1,
pe_weights_forward | pe_weights_init);
dump_node_scores(LOG_TRACE, NULL, resource2->id, r2_nodes);
/* Current location score */
reason = "current location";
r1_weight = -INFINITY;
r2_weight = -INFINITY;
if (resource1->running_on) {
r1_node = pe__current_node(resource1);
r1_node = g_hash_table_lookup(r1_nodes, r1_node->details->id);
if (r1_node != NULL) {
r1_weight = r1_node->weight;
}
}
if (resource2->running_on) {
r2_node = pe__current_node(resource2);
r2_node = g_hash_table_lookup(r2_nodes, r2_node->details->id);
if (r2_node != NULL) {
r2_weight = r2_node->weight;
}
}
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
reason = "score";
for (gIter = nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
r1_node = NULL;
r2_node = NULL;
r1_weight = -INFINITY;
if (r1_nodes) {
r1_node = g_hash_table_lookup(r1_nodes, node->details->id);
}
if (r1_node) {
r1_weight = r1_node->weight;
}
r2_weight = -INFINITY;
if (r2_nodes) {
r2_node = g_hash_table_lookup(r2_nodes, node->details->id);
}
if (r2_node) {
r2_weight = r2_node->weight;
}
if (r1_weight > r2_weight) {
rc = -1;
goto done;
}
if (r1_weight < r2_weight) {
rc = 1;
goto done;
}
}
done:
crm_trace("%s (%d) on %s %c %s (%d) on %s: %s",
resource1->id, r1_weight, r1_node ? r1_node->details->id : "n/a",
rc < 0 ? '>' : rc > 0 ? '<' : '=',
resource2->id, r2_weight, r2_node ? r2_node->details->id : "n/a", reason);
if (r1_nodes) {
g_hash_table_destroy(r1_nodes);
}
if (r2_nodes) {
g_hash_table_destroy(r2_nodes);
}
return rc;
}
static void
allocate_resources(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
if (is_set(data_set->flags, pe_flag_have_remote_nodes)) {
/* Force remote connection resources to be allocated first. This
* also forces any colocation dependencies to be allocated as well */
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
if (rsc->is_remote_node == FALSE) {
continue;
}
pe_rsc_trace(rsc, "Allocating: %s", rsc->id);
/* For remote node connection resources, always prefer the partial
* migration target during resource allocation, if the rsc is in the
* middle of a migration.
*/
rsc->cmds->allocate(rsc, rsc->partial_migration_target, data_set);
}
}
/* now do the rest of the resources */
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
if (rsc->is_remote_node == TRUE) {
continue;
}
pe_rsc_trace(rsc, "Allocating: %s", rsc->id);
rsc->cmds->allocate(rsc, NULL, data_set);
}
}
/* We always use pe_order_preserve with these convenience functions to exempt
* internally generated constraints from the prohibition of user constraints
* involving remote connection resources.
*
* The start ordering additionally uses pe_order_runnable_left so that the
* specified action is not runnable if the start is not runnable.
*/
static inline void
order_start_then_action(resource_t *lh_rsc, action_t *rh_action,
enum pe_ordering extra, pe_working_set_t *data_set)
{
if (lh_rsc && rh_action && data_set) {
custom_action_order(lh_rsc, start_key(lh_rsc), NULL,
rh_action->rsc, NULL, rh_action,
pe_order_preserve | pe_order_runnable_left | extra,
data_set);
}
}
static inline void
order_action_then_stop(action_t *lh_action, resource_t *rh_rsc,
enum pe_ordering extra, pe_working_set_t *data_set)
{
if (lh_action && rh_rsc && data_set) {
custom_action_order(lh_action->rsc, NULL, lh_action,
rh_rsc, stop_key(rh_rsc), NULL,
pe_order_preserve | extra, data_set);
}
}
// Clear fail counts for orphaned rsc on all online nodes
static void
cleanup_orphans(resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
if (node->details->online
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL,
data_set)) {
pe_action_t *clear_op = NULL;
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
data_set);
/* We can't use order_action_then_stop() here because its
* pe_order_preserve breaks things
*/
custom_action_order(clear_op->rsc, NULL, clear_op,
rsc, stop_key(rsc), NULL,
pe_order_optional, data_set);
}
}
}
gboolean
stage5(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
int log_prio = show_utilization? LOG_STDOUT : utilization_log_level;
if (safe_str_neq(data_set->placement_strategy, "default")) {
GListPtr nodes = g_list_copy(data_set->nodes);
nodes = sort_nodes_by_weight(nodes, NULL, data_set);
data_set->resources =
g_list_sort_with_data(data_set->resources, sort_rsc_process_order, nodes);
g_list_free(nodes);
}
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
dump_node_capacity(log_prio, "Original", node);
}
crm_trace("Allocating services");
/* Take (next) highest resource, assign it and create its actions */
allocate_resources(data_set);
gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
dump_node_capacity(log_prio, "Remaining", node);
}
// Process deferred action checks
pe__foreach_param_check(data_set, check_params);
pe__free_param_checks(data_set);
if (is_set(data_set->flags, pe_flag_startup_probes)) {
crm_trace("Calculating needed probes");
/* This code probably needs optimization
* ptest -x with 100 nodes, 100 clones and clone-max=100:
With probes:
ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status
ptest[14781]: 2010/09/27_17:56:46 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints
ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints
ptest[14781]: 2010/09/27_17:56:47 notice: TRACE: do_calculations: pengine.c:292 Check actions
ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: do_calculations: pengine.c:299 Allocate resources
ptest[14781]: 2010/09/27_17:56:48 notice: TRACE: stage5: allocate.c:881 Allocating services
ptest[14781]: 2010/09/27_17:56:49 notice: TRACE: stage5: allocate.c:894 Calculating needed probes
ptest[14781]: 2010/09/27_17:56:51 notice: TRACE: stage5: allocate.c:899 Creating actions
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: stage5: allocate.c:905 Creating done
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases
ptest[14781]: 2010/09/27_17:56:52 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints
36s
ptest[14781]: 2010/09/27_17:57:28 notice: TRACE: do_calculations: pengine.c:320 Create transition graph
Without probes:
ptest[14637]: 2010/09/27_17:56:21 notice: TRACE: do_calculations: pengine.c:258 Calculate cluster status
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:278 Applying placement constraints
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:285 Create internal constraints
ptest[14637]: 2010/09/27_17:56:22 notice: TRACE: do_calculations: pengine.c:292 Check actions
ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: do_calculations: pengine.c:299 Allocate resources
ptest[14637]: 2010/09/27_17:56:23 notice: TRACE: stage5: allocate.c:881 Allocating services
ptest[14637]: 2010/09/27_17:56:24 notice: TRACE: stage5: allocate.c:899 Creating actions
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: stage5: allocate.c:905 Creating done
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:306 Processing fencing and shutdown cases
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:313 Applying ordering constraints
ptest[14637]: 2010/09/27_17:56:25 notice: TRACE: do_calculations: pengine.c:320 Create transition graph
*/
probe_resources(data_set);
}
crm_trace("Handle orphans");
if (is_set(data_set->flags, pe_flag_stop_rsc_orphans)) {
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
/* There's no need to recurse into rsc->children because those
* should just be unallocated clone instances.
*/
if (is_set(rsc->flags, pe_rsc_orphan)) {
cleanup_orphans(rsc, data_set);
}
}
}
crm_trace("Creating actions");
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
rsc->cmds->create_actions(rsc, data_set);
}
crm_trace("Creating done");
return TRUE;
}
static gboolean
is_managed(const resource_t * rsc)
{
GListPtr gIter = rsc->children;
if (is_set(rsc->flags, pe_rsc_managed)) {
return TRUE;
}
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
if (is_managed(child_rsc)) {
return TRUE;
}
}
return FALSE;
}
static gboolean
any_managed_resources(pe_working_set_t * data_set)
{
GListPtr gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
if (is_managed(rsc)) {
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Create pseudo-op for guest node fence, and order relative to it
*
* \param[in] node Guest node to fence
* \param[in] data_set Working set of CIB state
*/
static void
fence_guest(pe_node_t *node, pe_working_set_t *data_set)
{
resource_t *container = node->details->remote_rsc->container;
pe_action_t *stop = NULL;
pe_action_t *stonith_op = NULL;
/* The fence action is just a label; we don't do anything differently for
* off vs. reboot. We specify it explicitly, rather than let it default to
* cluster's default action, because we are not _initiating_ fencing -- we
* are creating a pseudo-event to describe fencing that is already occurring
* by other means (container recovery).
*/
const char *fence_action = "off";
/* Check whether guest's container resource has any explicit stop or
* start (the stop may be implied by fencing of the guest's host).
*/
if (container) {
stop = find_first_action(container->actions, NULL, CRMD_ACTION_STOP, NULL);
if (find_first_action(container->actions, NULL, CRMD_ACTION_START, NULL)) {
fence_action = "reboot";
}
}
/* Create a fence pseudo-event, so we have an event to order actions
* against, and the controller can always detect it.
*/
stonith_op = pe_fence_op(node, fence_action, FALSE, "guest is unclean", data_set);
update_action_flags(stonith_op, pe_action_pseudo | pe_action_runnable,
__FUNCTION__, __LINE__);
/* We want to imply stops/demotes after the guest is stopped, not wait until
* it is restarted, so we always order pseudo-fencing after stop, not start
* (even though start might be closer to what is done for a real reboot).
*/
if(stop && is_set(stop->flags, pe_action_pseudo)) {
pe_action_t *parent_stonith_op = pe_fence_op(stop->node, NULL, FALSE, NULL, data_set);
crm_info("Implying guest node %s is down (action %d) after %s fencing",
node->details->uname, stonith_op->id, stop->node->details->uname);
order_actions(parent_stonith_op, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
} else if (stop) {
order_actions(stop, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
crm_info("Implying guest node %s is down (action %d) "
"after container %s is stopped (action %d)",
node->details->uname, stonith_op->id,
container->id, stop->id);
} else {
/* If we're fencing the guest node but there's no stop for the guest
* resource, we must think the guest is already stopped. However, we may
* think so because its resource history was just cleaned. To avoid
* unnecessarily considering the guest node down if it's really up,
* order the pseudo-fencing after any stop of the connection resource,
* which will be ordered after any container (re-)probe.
*/
stop = find_first_action(node->details->remote_rsc->actions, NULL,
RSC_STOP, NULL);
if (stop) {
order_actions(stop, stonith_op, pe_order_optional);
crm_info("Implying guest node %s is down (action %d) "
"after connection is stopped (action %d)",
node->details->uname, stonith_op->id, stop->id);
} else {
/* Not sure why we're fencing, but everything must already be
* cleanly stopped.
*/
crm_info("Implying guest node %s is down (action %d) ",
node->details->uname, stonith_op->id);
}
}
/* Order/imply other actions relative to pseudo-fence as with real fence */
pcmk__order_vs_fence(stonith_op, data_set);
}
/*
* Create dependencies for stonith and shutdown operations
*/
gboolean
stage6(pe_working_set_t * data_set)
{
action_t *dc_down = NULL;
action_t *stonith_op = NULL;
gboolean integrity_lost = FALSE;
gboolean need_stonith = TRUE;
GListPtr gIter;
GListPtr stonith_ops = NULL;
GList *shutdown_ops = NULL;
/* Remote ordering constraints need to happen prior to calculating fencing
* because it is one more place we will mark the node as dirty.
*
* A nice side effect of doing them early is that apply_*_ordering() can be
* simpler because pe_fence_node() has already done some of the work.
*/
crm_trace("Creating remote ordering constraints");
apply_remote_node_ordering(data_set);
crm_trace("Processing fencing and shutdown cases");
if (any_managed_resources(data_set) == FALSE) {
crm_notice("Delaying fencing operations until there are resources to manage");
need_stonith = FALSE;
}
/* Check each node for stonith/shutdown */
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pe__is_guest_node(node)) {
if (node->details->remote_requires_reset && need_stonith
&& pe_can_fence(data_set, node)) {
fence_guest(node, data_set);
}
continue;
}
stonith_op = NULL;
if (node->details->unclean
&& need_stonith && pe_can_fence(data_set, node)) {
stonith_op = pe_fence_op(node, NULL, FALSE, "node is unclean", data_set);
pe_warn("Scheduling Node %s for STONITH", node->details->uname);
pcmk__order_vs_fence(stonith_op, data_set);
if (node->details->is_dc) {
// Remember if the DC is being fenced
dc_down = stonith_op;
} else {
if (is_not_set(data_set->flags, pe_flag_concurrent_fencing)
&& (stonith_ops != NULL)) {
/* Concurrent fencing is disabled, so order each non-DC
* fencing in a chain. If there is any DC fencing or
* shutdown, it will be ordered after the last action in the
* chain later.
*/
order_actions((pe_action_t *) stonith_ops->data,
stonith_op, pe_order_optional);
}
// Remember all non-DC fencing actions in a separate list
stonith_ops = g_list_prepend(stonith_ops, stonith_op);
}
} else if (node->details->online && node->details->shutdown &&
/* TODO define what a shutdown op means for a remote node.
* For now we do not send shutdown operations for remote nodes, but
* if we can come up with a good use for this in the future, we will. */
pe__is_guest_or_remote_node(node) == FALSE) {
action_t *down_op = sched_shutdown_op(node, data_set);
if (node->details->is_dc) {
// Remember if the DC is being shut down
dc_down = down_op;
} else {
// Remember non-DC shutdowns for later ordering
shutdown_ops = g_list_prepend(shutdown_ops, down_op);
}
}
if (node->details->unclean && stonith_op == NULL) {
integrity_lost = TRUE;
pe_warn("Node %s is unclean!", node->details->uname);
}
}
if (integrity_lost) {
if (is_set(data_set->flags, pe_flag_stonith_enabled) == FALSE) {
pe_warn("YOUR RESOURCES ARE NOW LIKELY COMPROMISED");
pe_err("ENABLE STONITH TO KEEP YOUR RESOURCES SAFE");
} else if (is_set(data_set->flags, pe_flag_have_quorum) == FALSE) {
crm_notice("Cannot fence unclean nodes until quorum is"
" attained (or no-quorum-policy is set to ignore)");
}
}
if (dc_down != NULL) {
/* Order any non-DC shutdowns before any DC shutdown, to avoid repeated
* DC elections. However, we don't want to order non-DC shutdowns before
* a DC *fencing*, because even though we don't want a node that's
* shutting down to become DC, the DC fencing could be ordered before a
* clone stop that's also ordered before the shutdowns, thus leading to
* a graph loop.
*/
if (safe_str_eq(dc_down->task, CRM_OP_SHUTDOWN)) {
for (gIter = shutdown_ops; gIter != NULL; gIter = gIter->next) {
action_t *node_stop = (action_t *) gIter->data;
crm_debug("Ordering shutdown on %s before %s on DC %s",
node_stop->node->details->uname,
dc_down->task, dc_down->node->details->uname);
order_actions(node_stop, dc_down, pe_order_optional);
}
}
// Order any non-DC fencing before any DC fencing or shutdown
if (is_set(data_set->flags, pe_flag_concurrent_fencing)) {
/* With concurrent fencing, order each non-DC fencing action
* separately before any DC fencing or shutdown.
*/
for (gIter = stonith_ops; gIter != NULL; gIter = gIter->next) {
order_actions((pe_action_t *) gIter->data, dc_down,
pe_order_optional);
}
} else if (stonith_ops) {
/* Without concurrent fencing, the non-DC fencing actions are
* already ordered relative to each other, so we just need to order
* the DC fencing after the last action in the chain (which is the
* first item in the list).
*/
order_actions((pe_action_t *) stonith_ops->data, dc_down,
pe_order_optional);
}
}
g_list_free(stonith_ops);
g_list_free(shutdown_ops);
return TRUE;
}
/*
* Determine the sets of independent actions and the correct order for the
* actions in each set.
*
* Mark dependencies of un-runnable actions un-runnable
*
*/
static GListPtr
find_actions_by_task(GListPtr actions, resource_t * rsc, const char *original_key)
{
GListPtr list = NULL;
list = find_actions(actions, original_key, NULL);
if (list == NULL) {
/* we're potentially searching a child of the original resource */
char *key = NULL;
char *task = NULL;
guint interval_ms = 0;
if (parse_op_key(original_key, NULL, &task, &interval_ms)) {
key = generate_op_key(rsc->id, task, interval_ms);
list = find_actions(actions, key, NULL);
} else {
crm_err("search key: %s", original_key);
}
free(key);
free(task);
}
return list;
}
static void
rsc_order_then(pe_action_t *lh_action, pe_resource_t *rsc,
pe__ordering_t *order)
{
GListPtr gIter = NULL;
GListPtr rh_actions = NULL;
action_t *rh_action = NULL;
enum pe_ordering type;
CRM_CHECK(rsc != NULL, return);
CRM_CHECK(order != NULL, return);
type = order->type;
rh_action = order->rh_action;
crm_trace("Processing RH of ordering constraint %d", order->id);
if (rh_action != NULL) {
rh_actions = g_list_prepend(NULL, rh_action);
} else if (rsc != NULL) {
rh_actions = find_actions_by_task(rsc->actions, rsc, order->rh_action_task);
}
if (rh_actions == NULL) {
pe_rsc_trace(rsc, "No RH-Side (%s/%s) found for constraint..."
" ignoring", rsc->id, order->rh_action_task);
if (lh_action) {
pe_rsc_trace(rsc, "LH-Side was: %s", lh_action->uuid);
}
return;
}
if (lh_action && lh_action->rsc == rsc && is_set(lh_action->flags, pe_action_dangle)) {
pe_rsc_trace(rsc, "Detected dangling operation %s -> %s", lh_action->uuid,
order->rh_action_task);
clear_bit(type, pe_order_implies_then);
}
gIter = rh_actions;
for (; gIter != NULL; gIter = gIter->next) {
action_t *rh_action_iter = (action_t *) gIter->data;
if (lh_action) {
order_actions(lh_action, rh_action_iter, type);
} else if (type & pe_order_implies_then) {
update_action_flags(rh_action_iter, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
crm_warn("Unrunnable %s 0x%.6x", rh_action_iter->uuid, type);
} else {
crm_warn("neither %s 0x%.6x", rh_action_iter->uuid, type);
}
}
g_list_free(rh_actions);
}
static void
rsc_order_first(pe_resource_t *lh_rsc, pe__ordering_t *order,
pe_working_set_t *data_set)
{
GListPtr gIter = NULL;
GListPtr lh_actions = NULL;
action_t *lh_action = order->lh_action;
resource_t *rh_rsc = order->rh_rsc;
crm_trace("Processing LH of ordering constraint %d", order->id);
CRM_ASSERT(lh_rsc != NULL);
if (lh_action != NULL) {
lh_actions = g_list_prepend(NULL, lh_action);
} else {
lh_actions = find_actions_by_task(lh_rsc->actions, lh_rsc, order->lh_action_task);
}
if (lh_actions == NULL && lh_rsc != rh_rsc) {
char *key = NULL;
char *op_type = NULL;
guint interval_ms = 0;
parse_op_key(order->lh_action_task, NULL, &op_type, &interval_ms);
key = generate_op_key(lh_rsc->id, op_type, interval_ms);
if (lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_STOPPED && safe_str_eq(op_type, RSC_STOP)) {
free(key);
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - ignoring",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
} else if (lh_rsc->fns->state(lh_rsc, TRUE) == RSC_ROLE_SLAVE && safe_str_eq(op_type, RSC_DEMOTE)) {
free(key);
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - ignoring",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
} else {
pe_rsc_trace(lh_rsc, "No LH-Side (%s/%s) found for constraint %d with %s - creating",
lh_rsc->id, order->lh_action_task, order->id, order->rh_action_task);
lh_action = custom_action(lh_rsc, key, op_type, NULL, TRUE, TRUE, data_set);
lh_actions = g_list_prepend(NULL, lh_action);
}
free(op_type);
}
gIter = lh_actions;
for (; gIter != NULL; gIter = gIter->next) {
action_t *lh_action_iter = (action_t *) gIter->data;
if (rh_rsc == NULL && order->rh_action) {
rh_rsc = order->rh_action->rsc;
}
if (rh_rsc) {
rsc_order_then(lh_action_iter, rh_rsc, order);
} else if (order->rh_action) {
order_actions(lh_action_iter, order->rh_action, order->type);
}
}
g_list_free(lh_actions);
}
extern void update_colo_start_chain(pe_action_t *action,
pe_working_set_t *data_set);
static int
is_recurring_action(action_t *action)
{
const char *interval_ms_s = g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS);
guint interval_ms = crm_parse_ms(interval_ms_s);
return (interval_ms > 0);
}
static void
apply_container_ordering(action_t *action, pe_working_set_t *data_set)
{
/* VMs are also classified as containers for these purposes... in
* that they both involve a 'thing' running on a real or remote
* cluster node.
*
* This allows us to be smarter about the type and extent of
* recovery actions required in various scenarios
*/
resource_t *remote_rsc = NULL;
resource_t *container = NULL;
enum action_tasks task = text2task(action->task);
CRM_ASSERT(action->rsc);
CRM_ASSERT(action->node);
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
container = remote_rsc->container;
CRM_ASSERT(container);
if(is_set(container->flags, pe_rsc_failed)) {
pe_fence_node(data_set, action->node, "container failed");
}
crm_trace("Order %s action %s relative to %s%s for %s%s",
action->task, action->uuid,
is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "",
remote_rsc->id,
is_set(container->flags, pe_rsc_failed)? "failed " : "",
container->id);
if (safe_str_eq(action->task, CRMD_ACTION_MIGRATE)
|| safe_str_eq(action->task, CRMD_ACTION_MIGRATED)) {
/* Migration ops map to "no_action", but we need to apply the same
* ordering as for stop or demote (see get_router_node()).
*/
task = stop_rsc;
}
switch (task) {
case start_rsc:
case action_promote:
/* Force resource recovery if the container is recovered */
order_start_then_action(container, action, pe_order_implies_then,
data_set);
/* Wait for the connection resource to be up too */
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
break;
case stop_rsc:
case action_demote:
if (is_set(container->flags, pe_rsc_failed)) {
/* When the container representing a guest node fails, any stop
* or demote actions for resources running on the guest node
* are implied by the container stopping. This is similar to
* how fencing operations work for cluster nodes and remote
* nodes.
*/
} else {
/* Ensure the operation happens before the connection is brought
* down.
*
* If we really wanted to, we could order these after the
* connection start, IFF the container's current role was
* stopped (otherwise we re-introduce an ordering loop when the
* connection is restarting).
*/
order_action_then_stop(action, remote_rsc, pe_order_none,
data_set);
}
break;
default:
/* Wait for the connection resource to be up */
if (is_recurring_action(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
if(task != no_action) {
order_start_then_action(remote_rsc, action,
pe_order_implies_then, data_set);
}
} else {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
}
break;
}
}
static enum remote_connection_state
get_remote_node_state(pe_node_t *node)
{
resource_t *remote_rsc = NULL;
node_t *cluster_node = NULL;
CRM_ASSERT(node);
remote_rsc = node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
cluster_node = pe__current_node(remote_rsc);
/* If the cluster node the remote connection resource resides on
* is unclean or went offline, we can't process any operations
* on that remote node until after it starts elsewhere.
*/
if(remote_rsc->next_role == RSC_ROLE_STOPPED || remote_rsc->allocated_to == NULL) {
/* The connection resource is not going to run anywhere */
if (cluster_node && cluster_node->details->unclean) {
/* The remote connection is failed because its resource is on a
* failed node and can't be recovered elsewhere, so we must fence.
*/
return remote_state_failed;
}
if (is_not_set(remote_rsc->flags, pe_rsc_failed)) {
/* Connection resource is cleanly stopped */
return remote_state_stopped;
}
/* Connection resource is failed */
if ((remote_rsc->next_role == RSC_ROLE_STOPPED)
&& remote_rsc->remote_reconnect_ms
&& node->details->remote_was_fenced
&& !pe__shutdown_requested(node)) {
/* We won't know whether the connection is recoverable until the
* reconnect interval expires and we reattempt connection.
*/
return remote_state_unknown;
}
/* The remote connection is in a failed state. If there are any
* resources known to be active on it (stop) or in an unknown state
* (probe), we must assume the worst and fence it.
*/
return remote_state_failed;
} else if (cluster_node == NULL) {
/* Connection is recoverable but not currently running anywhere, see if we can recover it first */
return remote_state_unknown;
} else if(cluster_node->details->unclean == TRUE
|| cluster_node->details->online == FALSE) {
/* Connection is running on a dead node, see if we can recover it first */
return remote_state_resting;
} else if (pcmk__list_of_multiple(remote_rsc->running_on)
&& remote_rsc->partial_migration_source
&& remote_rsc->partial_migration_target) {
/* We're in the middle of migrating a connection resource,
* wait until after the resource migrates before performing
* any actions.
*/
return remote_state_resting;
}
return remote_state_alive;
}
/*!
* \internal
* \brief Order actions on remote node relative to actions for the connection
*/
static void
apply_remote_ordering(action_t *action, pe_working_set_t *data_set)
{
resource_t *remote_rsc = NULL;
enum action_tasks task = text2task(action->task);
enum remote_connection_state state = get_remote_node_state(action->node);
enum pe_ordering order_opts = pe_order_none;
if (action->rsc == NULL) {
return;
}
CRM_ASSERT(action->node);
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc);
crm_trace("Order %s action %s relative to %s%s (state: %s)",
action->task, action->uuid,
is_set(remote_rsc->flags, pe_rsc_failed)? "failed " : "",
remote_rsc->id, state2text(state));
if (safe_str_eq(action->task, CRMD_ACTION_MIGRATE)
|| safe_str_eq(action->task, CRMD_ACTION_MIGRATED)) {
/* Migration ops map to "no_action", but we need to apply the same
* ordering as for stop or demote (see get_router_node()).
*/
task = stop_rsc;
}
switch (task) {
case start_rsc:
case action_promote:
order_opts = pe_order_none;
if (state == remote_state_failed) {
/* Force recovery, by making this action required */
order_opts |= pe_order_implies_then;
}
/* Ensure connection is up before running this action */
order_start_then_action(remote_rsc, action, order_opts, data_set);
break;
case stop_rsc:
if(state == remote_state_alive) {
order_action_then_stop(action, remote_rsc,
pe_order_implies_first, data_set);
} else if(state == remote_state_failed) {
/* The resource is active on the node, but since we don't have a
* valid connection, the only way to stop the resource is by
* fencing the node. There is no need to order the stop relative
* to the remote connection, since the stop will become implied
* by the fencing.
*/
pe_fence_node(data_set, action->node, "resources are active and the connection is unrecoverable");
} else if(remote_rsc->next_role == RSC_ROLE_STOPPED) {
/* State must be remote_state_unknown or remote_state_stopped.
* Since the connection is not coming back up in this
* transition, stop this resource first.
*/
order_action_then_stop(action, remote_rsc,
pe_order_implies_first, data_set);
} else {
/* The connection is going to be started somewhere else, so
* stop this resource after that completes.
*/
order_start_then_action(remote_rsc, action, pe_order_none, data_set);
}
break;
case action_demote:
/* Only order this demote relative to the connection start if the
* connection isn't being torn down. Otherwise, the demote would be
* blocked because the connection start would not be allowed.
*/
if(state == remote_state_resting || state == remote_state_unknown) {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
} /* Otherwise we can rely on the stop ordering */
break;
default:
/* Wait for the connection resource to be up */
if (is_recurring_action(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
order_start_then_action(remote_rsc, action,
pe_order_implies_then, data_set);
} else {
node_t *cluster_node = pe__current_node(remote_rsc);
if(task == monitor_rsc && state == remote_state_failed) {
/* We would only be here if we do not know the
* state of the resource on the remote node.
* Since we have no way to find out, it is
* necessary to fence the node.
*/
pe_fence_node(data_set, action->node, "resources are in an unknown state and the connection is unrecoverable");
}
if(cluster_node && state == remote_state_stopped) {
/* The connection is currently up, but is going
* down permanently.
*
* Make sure we check services are actually
* stopped _before_ we let the connection get
* closed
*/
order_action_then_stop(action, remote_rsc,
pe_order_runnable_left, data_set);
} else {
order_start_then_action(remote_rsc, action, pe_order_none,
data_set);
}
}
break;
}
}
static void
apply_remote_node_ordering(pe_working_set_t *data_set)
{
if (is_set(data_set->flags, pe_flag_have_remote_nodes) == FALSE) {
return;
}
for (GListPtr gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
resource_t *remote = NULL;
// We are only interested in resource actions
if (action->rsc == NULL) {
continue;
}
/* Special case: If we are clearing the failcount of an actual
* remote connection resource, then make sure this happens before
* any start of the resource in this transition.
*/
if (action->rsc->is_remote_node &&
safe_str_eq(action->task, CRM_OP_CLEAR_FAILCOUNT)) {
custom_action_order(action->rsc,
NULL,
action,
action->rsc,
generate_op_key(action->rsc->id, RSC_START, 0),
NULL,
pe_order_optional,
data_set);
continue;
}
// We are only interested in actions allocated to a node
if (action->node == NULL) {
continue;
}
if (!pe__is_guest_or_remote_node(action->node)) {
continue;
}
/* We are only interested in real actions.
*
* @TODO This is probably wrong; pseudo-actions might be converted to
* real actions and vice versa later in update_actions() at the end of
* stage7().
*/
if (is_set(action->flags, pe_action_pseudo)) {
continue;
}
remote = action->node->details->remote_rsc;
if (remote == NULL) {
// Orphaned
continue;
}
/* Another special case: if a resource is moving to a Pacemaker Remote
* node, order the stop on the original node after any start of the
* remote connection. This ensures that if the connection fails to
* start, we leave the resource running on the original node.
*/
if (safe_str_eq(action->task, RSC_START)) {
for (GList *item = action->rsc->actions; item != NULL;
item = item->next) {
pe_action_t *rsc_action = item->data;
if ((rsc_action->node->details != action->node->details)
&& safe_str_eq(rsc_action->task, RSC_STOP)) {
custom_action_order(remote, start_key(remote), NULL,
action->rsc, NULL, rsc_action,
pe_order_optional, data_set);
}
}
}
/* The action occurs across a remote connection, so create
* ordering constraints that guarantee the action occurs while the node
* is active (after start, before stop ... things like that).
*
* This is somewhat brittle in that we need to make sure the results of
* this ordering are compatible with the result of get_router_node().
* It would probably be better to add XML_LRM_ATTR_ROUTER_NODE as part
* of this logic rather than action2xml().
*/
if (remote->container) {
crm_trace("Container ordering for %s", action->uuid);
apply_container_ordering(action, data_set);
} else {
crm_trace("Remote ordering for %s", action->uuid);
apply_remote_ordering(action, data_set);
}
}
}
static gboolean
order_first_probe_unneeded(pe_action_t * probe, pe_action_t * rh_action)
{
/* No need to probe the resource on the node that is being
* unfenced. Otherwise it might introduce transition loop
* since probe will be performed after the node is
* unfenced.
*/
if (safe_str_eq(rh_action->task, CRM_OP_FENCE)
&& probe->node && rh_action->node
&& probe->node->details == rh_action->node->details) {
const char *op = g_hash_table_lookup(rh_action->meta, "stonith_action");
if (safe_str_eq(op, "on")) {
return TRUE;
}
}
// Shutdown waits for probe to complete only if it's on the same node
if ((safe_str_eq(rh_action->task, CRM_OP_SHUTDOWN))
&& probe->node && rh_action->node
&& probe->node->details != rh_action->node->details) {
return TRUE;
}
return FALSE;
}
static void
order_first_probes_imply_stops(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) {
pe__ordering_t *order = gIter->data;
enum pe_ordering order_type = pe_order_optional;
pe_resource_t *lh_rsc = order->lh_rsc;
pe_resource_t *rh_rsc = order->rh_rsc;
pe_action_t *lh_action = order->lh_action;
pe_action_t *rh_action = order->rh_action;
const char *lh_action_task = order->lh_action_task;
const char *rh_action_task = order->rh_action_task;
GListPtr probes = NULL;
GListPtr rh_actions = NULL;
GListPtr pIter = NULL;
if (lh_rsc == NULL) {
continue;
} else if (rh_rsc && lh_rsc == rh_rsc) {
continue;
}
if (lh_action == NULL && lh_action_task == NULL) {
continue;
}
if (rh_action == NULL && rh_action_task == NULL) {
continue;
}
/* Technically probe is expected to return "not running", which could be
* the alternative of stop action if the status of the resource is
* unknown yet.
*/
if (lh_action && safe_str_neq(lh_action->task, RSC_STOP)) {
continue;
} else if (lh_action == NULL
&& lh_action_task
&& crm_ends_with(lh_action_task, "_" RSC_STOP "_0") == FALSE) {
continue;
}
/* Do not probe the resource inside of a stopping container. Otherwise
* it might introduce transition loop since probe will be performed
* after the container starts again.
*/
if (rh_rsc && lh_rsc->container == rh_rsc) {
if (rh_action && safe_str_eq(rh_action->task, RSC_STOP)) {
continue;
} else if (rh_action == NULL && rh_action_task
&& crm_ends_with(rh_action_task,"_" RSC_STOP "_0")) {
continue;
}
}
if (order->type == pe_order_none) {
continue;
}
// Preserve the order options for future filtering
if (is_set(order->type, pe_order_apply_first_non_migratable)) {
set_bit(order_type, pe_order_apply_first_non_migratable);
}
if (is_set(order->type, pe_order_same_node)) {
set_bit(order_type, pe_order_same_node);
}
// Keep the order types for future filtering
if (order->type == pe_order_anti_colocation
|| order->type == pe_order_load) {
order_type = order->type;
}
probes = pe__resource_actions(lh_rsc, NULL, RSC_STATUS, FALSE);
if (probes == NULL) {
continue;
}
if (rh_action) {
rh_actions = g_list_prepend(rh_actions, rh_action);
} else if (rh_rsc && rh_action_task) {
rh_actions = find_actions(rh_rsc->actions, rh_action_task, NULL);
}
if (rh_actions == NULL) {
g_list_free(probes);
continue;
}
crm_trace("Processing for LH probe based on ordering constraint %s -> %s"
" (id=%d, type=%.6x)",
lh_action ? lh_action->uuid : lh_action_task,
rh_action ? rh_action->uuid : rh_action_task,
order->id, order->type);
for (pIter = probes; pIter != NULL; pIter = pIter->next) {
pe_action_t *probe = (pe_action_t *) pIter->data;
GListPtr rIter = NULL;
for (rIter = rh_actions; rIter != NULL; rIter = rIter->next) {
pe_action_t *rh_action_iter = (pe_action_t *) rIter->data;
if (order_first_probe_unneeded(probe, rh_action_iter)) {
continue;
}
order_actions(probe, rh_action_iter, order_type);
}
}
g_list_free(rh_actions);
g_list_free(probes);
}
}
static void
order_first_probe_then_restart_repromote(pe_action_t * probe,
pe_action_t * after,
pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
bool interleave = FALSE;
pe_resource_t *compatible_rsc = NULL;
if (probe == NULL
|| probe->rsc == NULL
|| probe->rsc->variant != pe_native) {
return;
}
if (after == NULL
// Avoid running into any possible loop
|| is_set(after->flags, pe_action_tracking)) {
return;
}
if (safe_str_neq(probe->task, RSC_STATUS)) {
return;
}
pe_set_action_bit(after, pe_action_tracking);
crm_trace("Processing based on %s %s -> %s %s",
probe->uuid,
probe->node ? probe->node->details->uname: "",
after->uuid,
after->node ? after->node->details->uname : "");
if (after->rsc
/* Better not build a dependency directly with a clone/group.
* We are going to proceed through the ordering chain and build
* dependencies with its children.
*/
&& after->rsc->variant == pe_native
&& probe->rsc != after->rsc) {
GListPtr then_actions = NULL;
enum pe_ordering probe_order_type = pe_order_optional;
if (safe_str_eq(after->task, RSC_START)) {
then_actions = pe__resource_actions(after->rsc, NULL, RSC_STOP, FALSE);
} else if (safe_str_eq(after->task, RSC_PROMOTE)) {
then_actions = pe__resource_actions(after->rsc, NULL, RSC_DEMOTE, FALSE);
}
for (gIter = then_actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *then = (pe_action_t *) gIter->data;
// Skip any pseudo action which for example is implied by fencing
if (is_set(then->flags, pe_action_pseudo)) {
continue;
}
order_actions(probe, then, probe_order_type);
}
g_list_free(then_actions);
}
if (after->rsc
&& after->rsc->variant > pe_group) {
const char *interleave_s = g_hash_table_lookup(after->rsc->meta,
XML_RSC_ATTR_INTERLEAVE);
interleave = crm_is_true(interleave_s);
if (interleave) {
/* For an interleaved clone, we should build a dependency only
* with the relevant clone child.
*/
compatible_rsc = find_compatible_child(probe->rsc,
after->rsc,
RSC_ROLE_UNKNOWN,
FALSE, data_set);
}
}
for (gIter = after->actions_after; gIter != NULL; gIter = gIter->next) {
pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) gIter->data;
/* pe_order_implies_then is the reason why a required A.start
* implies/enforces B.start to be required too, which is the cause of
* B.restart/re-promote.
*
* Not sure about pe_order_implies_then_on_node though. It's now only
* used for unfencing case, which tends to introduce transition
* loops...
*/
if (is_not_set(after_wrapper->type, pe_order_implies_then)) {
/* The order type between a group/clone and its child such as
* B.start-> B_child.start is:
* pe_order_implies_first_printed | pe_order_runnable_left
*
* Proceed through the ordering chain and build dependencies with
* its children.
*/
if (after->rsc == NULL
|| after->rsc->variant < pe_group
|| probe->rsc->parent == after->rsc
|| after_wrapper->action->rsc == NULL
|| after_wrapper->action->rsc->variant > pe_group
|| after->rsc != after_wrapper->action->rsc->parent) {
continue;
}
/* Proceed to the children of a group or a non-interleaved clone.
* For an interleaved clone, proceed only to the relevant child.
*/
if (after->rsc->variant > pe_group
&& interleave == TRUE
&& (compatible_rsc == NULL
|| compatible_rsc != after_wrapper->action->rsc)) {
continue;
}
}
crm_trace("Proceeding through %s %s -> %s %s (type=0x%.6x)",
after->uuid,
after->node ? after->node->details->uname: "",
after_wrapper->action->uuid,
after_wrapper->action->node ? after_wrapper->action->node->details->uname : "",
after_wrapper->type);
order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set);
}
}
static void clear_actions_tracking_flag(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (is_set(action->flags, pe_action_tracking)) {
pe_clear_action_bit(action, pe_action_tracking);
}
}
}
static void
order_first_rsc_probes(pe_resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
GListPtr probes = NULL;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe_resource_t * child = (pe_resource_t *) gIter->data;
order_first_rsc_probes(child, data_set);
}
if (rsc->variant != pe_native) {
return;
}
probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE);
for (gIter = probes; gIter != NULL; gIter= gIter->next) {
pe_action_t *probe = (pe_action_t *) gIter->data;
GListPtr aIter = NULL;
for (aIter = probe->actions_after; aIter != NULL; aIter = aIter->next) {
pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) aIter->data;
order_first_probe_then_restart_repromote(probe, after_wrapper->action, data_set);
clear_actions_tracking_flag(data_set);
}
}
g_list_free(probes);
}
static void
order_first_probes(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
order_first_rsc_probes(rsc, data_set);
}
order_first_probes_imply_stops(data_set);
}
static void
order_then_probes(pe_working_set_t * data_set)
{
#if 0
GListPtr gIter = NULL;
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
/* Given "A then B", we would prefer to wait for A to be
* started before probing B.
*
* If A was a filesystem on which the binaries and data for B
* lived, it would have been useful if the author of B's agent
* could assume that A is running before B.monitor will be
* called.
*
* However we can't _only_ probe once A is running, otherwise
* we'd not detect the state of B if A could not be started
* for some reason.
*
* In practice however, we cannot even do an opportunistic
* version of this because B may be moving:
*
* B.probe -> B.start
* B.probe -> B.stop
* B.stop -> B.start
* A.stop -> A.start
* A.start -> B.probe
*
* So far so good, but if we add the result of this code:
*
* B.stop -> A.stop
*
* Then we get a loop:
*
* B.probe -> B.stop -> A.stop -> A.start -> B.probe
*
* We could kill the 'B.probe -> B.stop' dependency, but that
* could mean stopping B "too" soon, because B.start must wait
* for the probes to complete.
*
* Another option is to allow it only if A is a non-unique
* clone with clone-max == node-max (since we'll never be
* moving it). However, we could still be stopping one
* instance at the same time as starting another.
* The complexity of checking for allowed conditions combined
* with the ever narrowing usecase suggests that this code
* should remain disabled until someone gets smarter.
*/
action_t *start = NULL;
GListPtr actions = NULL;
GListPtr probes = NULL;
actions = pe__resource_actions(rsc, NULL, RSC_START, FALSE);
if (actions) {
start = actions->data;
g_list_free(actions);
}
if(start == NULL) {
crm_err("No start action for %s", rsc->id);
continue;
}
probes = pe__resource_actions(rsc, NULL, RSC_STATUS, FALSE);
for (actions = start->actions_before; actions != NULL; actions = actions->next) {
action_wrapper_t *before = (action_wrapper_t *) actions->data;
GListPtr pIter = NULL;
action_t *first = before->action;
resource_t *first_rsc = first->rsc;
if(first->required_runnable_before) {
GListPtr clone_actions = NULL;
for (clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) {
before = (action_wrapper_t *) clone_actions->data;
crm_trace("Testing %s -> %s (%p) for %s", first->uuid, before->action->uuid, before->action->rsc, start->uuid);
CRM_ASSERT(before->action->rsc);
first_rsc = before->action->rsc;
break;
}
} else if(safe_str_neq(first->task, RSC_START)) {
crm_trace("Not a start op %s for %s", first->uuid, start->uuid);
}
if(first_rsc == NULL) {
continue;
} else if(uber_parent(first_rsc) == uber_parent(start->rsc)) {
crm_trace("Same parent %s for %s", first_rsc->id, start->uuid);
continue;
} else if(FALSE && pe_rsc_is_clone(uber_parent(first_rsc)) == FALSE) {
crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid);
continue;
}
crm_err("Applying %s before %s %d", first->uuid, start->uuid, uber_parent(first_rsc)->variant);
for (pIter = probes; pIter != NULL; pIter = pIter->next) {
action_t *probe = (action_t *) pIter->data;
crm_err("Ordering %s before %s", first->uuid, probe->uuid);
order_actions(first, probe, pe_order_optional);
}
}
}
#endif
}
static void
order_probes(pe_working_set_t * data_set)
{
order_first_probes(data_set);
order_then_probes(data_set);
}
gboolean
stage7(pe_working_set_t * data_set)
{
GList *gIter = NULL;
crm_trace("Applying ordering constraints");
/* Don't ask me why, but apparently they need to be processed in
* the order they were created in... go figure
*
* Also g_list_append() has horrendous performance characteristics
* So we need to use g_list_prepend() and then reverse the list here
*/
data_set->ordering_constraints = g_list_reverse(data_set->ordering_constraints);
for (gIter = data_set->ordering_constraints; gIter != NULL; gIter = gIter->next) {
pe__ordering_t *order = gIter->data;
resource_t *rsc = order->lh_rsc;
crm_trace("Applying ordering constraint: %d", order->id);
if (rsc != NULL) {
crm_trace("rsc_action-to-*");
rsc_order_first(rsc, order, data_set);
continue;
}
rsc = order->rh_rsc;
if (rsc != NULL) {
crm_trace("action-to-rsc_action");
rsc_order_then(order->lh_action, rsc, order);
} else {
crm_trace("action-to-action");
order_actions(order->lh_action, order->rh_action, order->type);
}
}
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
update_colo_start_chain(action, data_set);
}
crm_trace("Ordering probes");
order_probes(data_set);
crm_trace("Updating %d actions", g_list_length(data_set->actions));
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
update_action(action, data_set);
}
// Check for invalid orderings
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
pe_action_wrapper_t *input = NULL;
for (GList *input_iter = action->actions_before;
input_iter != NULL; input_iter = input_iter->next) {
input = (pe_action_wrapper_t *) input_iter->data;
if (pcmk__ordering_is_invalid(action, input)) {
input->type = pe_order_none;
}
}
}
LogNodeActions(data_set, FALSE);
for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
LogActions(rsc, data_set, FALSE);
}
return TRUE;
}
static int transition_id = -1;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const char *filename)
{
if (was_processing_error) {
crm_err("Calculated transition %d (with errors), saving inputs in %s",
transition_id, filename);
} else if (was_processing_warning) {
crm_warn("Calculated transition %d (with warnings), saving inputs in %s",
transition_id, filename);
} else {
crm_notice("Calculated transition %d, saving inputs in %s",
transition_id, filename);
}
if (crm_config_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*
* Create a dependency graph to send to the transitioner (via the controller)
*/
gboolean
stage8(pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
const char *value = NULL;
transition_id++;
crm_trace("Creating transition graph %d.", transition_id);
data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
value = pe_pref(data_set->config_hash, "cluster-delay");
crm_xml_add(data_set->graph, "cluster-delay", value);
value = pe_pref(data_set->config_hash, "stonith-timeout");
crm_xml_add(data_set->graph, "stonith-timeout", value);
crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY");
if (is_set(data_set->flags, pe_flag_start_failure_fatal)) {
crm_xml_add(data_set->graph, "failed-start-offset", "INFINITY");
} else {
crm_xml_add(data_set->graph, "failed-start-offset", "1");
}
value = pe_pref(data_set->config_hash, "batch-limit");
crm_xml_add(data_set->graph, "batch-limit", value);
crm_xml_add_int(data_set->graph, "transition_id", transition_id);
value = pe_pref(data_set->config_hash, "migration-limit");
if (crm_int_helper(value, NULL) > 0) {
crm_xml_add(data_set->graph, "migration-limit", value);
}
if (data_set->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu",
(long long) data_set->recheck_by);
crm_xml_add(data_set->graph, "recheck-by", recheck_epoch);
free(recheck_epoch);
}
/* errors...
slist_iter(action, action_t, action_list, lpc,
if(action->optional == FALSE && action->runnable == FALSE) {
print_action("Ignoring", action, TRUE);
}
);
*/
/* The following code will de-duplicate action inputs, so nothing past this
* should rely on the action input type flags retaining their original
* values.
*/
gIter = data_set->resources;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
pe_rsc_trace(rsc, "processing actions for rsc=%s", rsc->id);
rsc->cmds->expand(rsc, data_set);
}
crm_log_xml_trace(data_set->graph, "created resource-driven action list");
/* pseudo action to distribute list of nodes with maintenance state update */
add_maintenance_update(data_set);
/* catch any non-resource specific actions */
crm_trace("processing non-resource actions");
gIter = data_set->actions;
for (; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
if (action->rsc
&& action->node
&& action->node->details->shutdown
&& is_not_set(action->rsc->flags, pe_rsc_maintenance)
&& is_not_set(action->flags, pe_action_optional)
&& is_not_set(action->flags, pe_action_runnable)
&& crm_str_eq(action->task, RSC_STOP, TRUE)
) {
/* Eventually we should just ignore the 'fence' case
* But for now it's the best way to detect (in CTS) when
* CIB resource updates are being lost
*/
if (is_set(data_set->flags, pe_flag_have_quorum)
|| data_set->no_quorum_policy == no_quorum_ignore) {
crm_crit("Cannot %s node '%s' because of %s:%s%s (%s)",
action->node->details->unclean ? "fence" : "shut down",
action->node->details->uname, action->rsc->id,
is_not_set(action->rsc->flags, pe_rsc_managed) ? " unmanaged" : " blocked",
is_set(action->rsc->flags, pe_rsc_failed) ? " failed" : "",
action->uuid);
}
}
graph_element_from_action(action, data_set);
}
crm_log_xml_trace(data_set->graph, "created generic action list");
crm_trace("Created transition graph %d.", transition_id);
return TRUE;
}
void
LogNodeActions(pe_working_set_t * data_set, gboolean terminal)
{
GListPtr gIter = NULL;
for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) {
char *node_name = NULL;
char *task = NULL;
action_t *action = (action_t *) gIter->data;
if (action->rsc != NULL) {
continue;
} else if (is_set(action->flags, pe_action_optional)) {
continue;
}
if (pe__is_guest_node(action->node)) {
node_name = crm_strdup_printf("%s (resource: %s)", action->node->details->uname, action->node->details->remote_rsc->container->id);
} else if(action->node) {
node_name = crm_strdup_printf("%s", action->node->details->uname);
}
if (safe_str_eq(action->task, CRM_OP_SHUTDOWN)) {
task = strdup("Shutdown");
} else if (safe_str_eq(action->task, CRM_OP_FENCE)) {
const char *op = g_hash_table_lookup(action->meta, "stonith_action");
task = crm_strdup_printf("Fence (%s)", op);
}
if(task == NULL) {
/* Nothing to report */
} else if(terminal && action->reason) {
printf(" * %s %s '%s'\n", task, node_name, action->reason);
} else if(terminal) {
printf(" * %s %s\n", task, node_name);
} else if(action->reason) {
crm_notice(" * %s %s '%s'\n", task, node_name, action->reason);
} else {
crm_notice(" * %s %s\n", task, node_name);
}
free(node_name);
free(task);
}
}
diff --git a/lib/pacemaker/pcmk_sched_graph.c b/lib/pacemaker/pcmk_sched_graph.c
index a6967fe522..355ffcad09 100644
--- a/lib/pacemaker/pcmk_sched_graph.c
+++ b/lib/pacemaker/pcmk_sched_graph.c
@@ -1,1804 +1,1844 @@
/*
- * Copyright 2004-2019 the Pacemaker project contributors
+ * Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
void update_colo_start_chain(pe_action_t *action, pe_working_set_t *data_set);
gboolean rsc_update_action(action_t * first, action_t * then, enum pe_ordering type);
static enum pe_action_flags
get_action_flags(action_t * action, node_t * node)
{
enum pe_action_flags flags = action->flags;
if (action->rsc) {
flags = action->rsc->cmds->action_flags(action, NULL);
if (pe_rsc_is_clone(action->rsc) && node) {
/* We only care about activity on $node */
enum pe_action_flags clone_flags = action->rsc->cmds->action_flags(action, node);
/* Go to great lengths to ensure the correct value for pe_action_runnable...
*
* If we are a clone, then for _ordering_ constraints, it's only relevant
* if we are runnable _anywhere_.
*
* This only applies to _runnable_ though, and only for ordering constraints.
* If this function is ever used during colocation, then we'll need additional logic
*
* Not very satisfying, but it's logical and appears to work well.
*/
if (is_not_set(clone_flags, pe_action_runnable)
&& is_set(flags, pe_action_runnable)) {
pe_rsc_trace(action->rsc, "Fixing up runnable flag for %s", action->uuid);
set_bit(clone_flags, pe_action_runnable);
}
flags = clone_flags;
}
}
return flags;
}
static char *
convert_non_atomic_uuid(char *old_uuid, resource_t * rsc, gboolean allow_notify,
gboolean free_original)
{
guint interval_ms = 0;
char *uuid = NULL;
char *rid = NULL;
char *raw_task = NULL;
int task = no_action;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "Processing %s", old_uuid);
if (old_uuid == NULL) {
return NULL;
} else if (strstr(old_uuid, "notify") != NULL) {
goto done; /* no conversion */
} else if (rsc->variant < pe_group) {
goto done; /* no conversion */
}
CRM_ASSERT(parse_op_key(old_uuid, &rid, &raw_task, &interval_ms));
if (interval_ms > 0) {
goto done; /* no conversion */
}
task = text2task(raw_task);
switch (task) {
case stop_rsc:
case start_rsc:
case action_notify:
case action_promote:
case action_demote:
break;
case stopped_rsc:
case started_rsc:
case action_notified:
case action_promoted:
case action_demoted:
task--;
break;
case monitor_rsc:
case shutdown_crm:
case stonith_node:
task = no_action;
break;
default:
crm_err("Unknown action: %s", raw_task);
task = no_action;
break;
}
if (task != no_action) {
if (is_set(rsc->flags, pe_rsc_notify) && allow_notify) {
uuid = generate_notify_key(rid, "confirmed-post", task2text(task + 1));
} else {
uuid = generate_op_key(rid, task2text(task + 1), 0);
}
pe_rsc_trace(rsc, "Converted %s -> %s", old_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(old_uuid);
}
if (free_original) {
free(old_uuid);
}
free(raw_task);
free(rid);
return uuid;
}
static action_t *
rsc_expand_action(action_t * action)
{
gboolean notify = FALSE;
action_t *result = action;
resource_t *rsc = action->rsc;
if (rsc == NULL) {
return action;
}
if ((rsc->parent == NULL)
|| (pe_rsc_is_clone(rsc) && (rsc->parent->variant == pe_container))) {
/* Only outermost resources have notification actions.
* The exception is those in bundles.
*/
notify = is_set(rsc->flags, pe_rsc_notify);
}
if (rsc->variant >= pe_group) {
/* Expand 'start' -> 'started' */
char *uuid = NULL;
uuid = convert_non_atomic_uuid(action->uuid, rsc, notify, FALSE);
if (uuid) {
pe_rsc_trace(rsc, "Converting %s to %s %d", action->uuid, uuid,
is_set(rsc->flags, pe_rsc_notify));
result = find_first_action(rsc->actions, uuid, NULL, NULL);
if (result == NULL) {
crm_err("Couldn't expand %s to %s in %s", action->uuid, uuid, rsc->id);
result = action;
}
free(uuid);
}
}
return result;
}
static enum pe_graph_flags
graph_update_action(action_t * first, action_t * then, node_t * node,
enum pe_action_flags first_flags, enum pe_action_flags then_flags,
action_wrapper_t *order, pe_working_set_t *data_set)
{
enum pe_graph_flags changed = pe_graph_none;
enum pe_ordering type = order->type;
gboolean processed = FALSE;
/* TODO: Do as many of these in parallel as possible */
if(is_set(type, pe_order_implies_then_on_node)) {
/* Normally we want the _whole_ 'then' clone to
* restart if 'first' is restarted, so then->node is
* needed.
*
* However for unfencing, we want to limit this to
* instances on the same node as 'first' (the
* unfencing operation), so first->node is supplied.
*
* Swap the node, from then on we can can treat it
* like any other 'pe_order_implies_then'
*/
clear_bit(type, pe_order_implies_then_on_node);
set_bit(type, pe_order_implies_then);
node = first->node;
}
clear_bit(first_flags, pe_action_pseudo);
if (type & pe_order_implies_then) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags & pe_action_optional, pe_action_optional,
pe_order_implies_then, data_set);
} else if (is_set(first_flags, pe_action_optional) == FALSE) {
if (update_action_flags(then, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__)) {
changed |= pe_graph_updated_then;
}
}
if (changed) {
pe_rsc_trace(then->rsc, "implies right: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("implies right: %s then %s %p", first->uuid, then->uuid, then->rsc);
}
}
if ((type & pe_order_restart) && then->rsc) {
enum pe_action_flags restart = (pe_action_optional | pe_action_runnable);
processed = TRUE;
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, restart,
pe_order_restart, data_set);
if (changed) {
pe_rsc_trace(then->rsc, "restart: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("restart: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_implies_first) {
processed = TRUE;
if (first->rsc) {
changed |= first->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_optional, pe_order_implies_first,
data_set);
} else if (is_set(first_flags, pe_action_optional) == FALSE) {
pe_rsc_trace(first->rsc, "first unrunnable: %s (%d) then %s (%d)",
first->uuid, is_set(first_flags, pe_action_optional),
then->uuid, is_set(then_flags, pe_action_optional));
if (update_action_flags(first, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__)) {
changed |= pe_graph_updated_first;
}
}
if (changed) {
pe_rsc_trace(then->rsc, "implies left: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("implies left: %s (%d) then %s (%d)",
first->uuid, is_set(first_flags, pe_action_optional),
then->uuid, is_set(then_flags, pe_action_optional));
}
}
if (type & pe_order_implies_first_master) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags & pe_action_optional, pe_action_optional,
pe_order_implies_first_master, data_set);
}
if (changed) {
pe_rsc_trace(then->rsc,
"implies left when right rsc is Master role: %s then %s: changed",
first->uuid, then->uuid);
} else {
crm_trace("implies left when right rsc is Master role: %s then %s", first->uuid,
then->uuid);
}
}
if (type & pe_order_one_or_more) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_runnable, pe_order_one_or_more,
data_set);
} else if (is_set(first_flags, pe_action_runnable)) {
/* alright. a "first" action is considered runnable, incremente
* the 'runnable_before' counter */
then->runnable_before++;
/* if the runnable before count for then exceeds the required number
* of "before" runnable actions... mark then as runnable */
if (then->runnable_before >= then->required_runnable_before) {
if (update_action_flags(then, pe_action_runnable, __FUNCTION__, __LINE__)) {
changed |= pe_graph_updated_then;
}
}
}
if (changed) {
pe_rsc_trace(then->rsc, "runnable_one_or_more: %s then %s: changed", first->uuid,
then->uuid);
} else {
crm_trace("runnable_one_or_more: %s then %s", first->uuid, then->uuid);
}
}
if (then->rsc && is_set(type, pe_order_probe)) {
processed = TRUE;
if (is_not_set(first_flags, pe_action_runnable) && first->rsc->running_on != NULL) {
pe_rsc_trace(then->rsc, "Ignoring %s then %s - %s is about to be stopped",
first->uuid, then->uuid, first->rsc->id);
type = pe_order_none;
order->type = pe_order_none;
} else {
pe_rsc_trace(then->rsc, "Enforcing %s then %s", first->uuid, then->uuid);
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_runnable, pe_order_runnable_left,
data_set);
}
if (changed) {
pe_rsc_trace(then->rsc, "runnable: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("runnable: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_runnable_left) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_runnable, pe_order_runnable_left,
data_set);
} else if (is_set(first_flags, pe_action_runnable) == FALSE) {
pe_rsc_trace(then->rsc, "then unrunnable: %s then %s", first->uuid, then->uuid);
if (update_action_flags(then, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__)) {
changed |= pe_graph_updated_then;
}
}
if (changed) {
pe_rsc_trace(then->rsc, "runnable: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("runnable: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_implies_first_migratable) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_optional,
pe_order_implies_first_migratable, data_set);
}
if (changed) {
pe_rsc_trace(then->rsc, "optional: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("optional: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_pseudo_left) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_optional, pe_order_pseudo_left,
data_set);
}
if (changed) {
pe_rsc_trace(then->rsc, "optional: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("optional: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_optional) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_runnable, pe_order_optional, data_set);
}
if (changed) {
pe_rsc_trace(then->rsc, "optional: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("optional: %s then %s", first->uuid, then->uuid);
}
}
if (type & pe_order_asymmetrical) {
processed = TRUE;
if (then->rsc) {
changed |= then->rsc->cmds->update_actions(first, then, node,
first_flags, pe_action_runnable, pe_order_asymmetrical,
data_set);
}
if (changed) {
pe_rsc_trace(then->rsc, "asymmetrical: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("asymmetrical: %s then %s", first->uuid, then->uuid);
}
}
if ((first->flags & pe_action_runnable) && (type & pe_order_implies_then_printed)
&& (first_flags & pe_action_optional) == 0) {
processed = TRUE;
crm_trace("%s implies %s printed", first->uuid, then->uuid);
update_action_flags(then, pe_action_print_always, __FUNCTION__, __LINE__); /* don't care about changed */
}
if (is_set(type, pe_order_implies_first_printed) && is_set(then_flags, pe_action_optional) == FALSE) {
processed = TRUE;
crm_trace("%s implies %s printed", then->uuid, first->uuid);
update_action_flags(first, pe_action_print_always, __FUNCTION__, __LINE__); /* don't care about changed */
}
if ((type & pe_order_implies_then
|| type & pe_order_implies_first
|| type & pe_order_restart)
&& first->rsc
&& safe_str_eq(first->task, RSC_STOP)
&& is_not_set(first->rsc->flags, pe_rsc_managed)
&& is_set(first->rsc->flags, pe_rsc_block)
&& is_not_set(first->flags, pe_action_runnable)) {
if (update_action_flags(then, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__)) {
changed |= pe_graph_updated_then;
}
if (changed) {
pe_rsc_trace(then->rsc, "unmanaged left: %s then %s: changed", first->uuid, then->uuid);
} else {
crm_trace("unmanaged left: %s then %s", first->uuid, then->uuid);
}
}
if (processed == FALSE) {
crm_trace("Constraint 0x%.6x not applicable", type);
}
return changed;
}
static void
mark_start_blocked(pe_resource_t *rsc, pe_resource_t *reason,
pe_working_set_t *data_set)
{
GListPtr gIter = rsc->actions;
char *reason_text = crm_strdup_printf("colocation with %s", reason->id);
for (; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
if (safe_str_neq(action->task, RSC_START)) {
continue;
}
if (is_set(action->flags, pe_action_runnable)) {
pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, NULL, reason_text, pe_action_runnable, FALSE);
update_colo_start_chain(action, data_set);
update_action(action, data_set);
}
}
free(reason_text);
}
void
update_colo_start_chain(pe_action_t *action, pe_working_set_t *data_set)
{
GListPtr gIter = NULL;
resource_t *rsc = NULL;
if (is_not_set(action->flags, pe_action_runnable) && safe_str_eq(action->task, RSC_START)) {
rsc = uber_parent(action->rsc);
if (rsc->parent) {
/* For bundles, uber_parent() returns the clone/master, not the
* bundle, so the existence of rsc->parent implies this is a bundle.
* In this case, we need the bundle resource, so that we can check
* if all containers are stopped/stopping.
*/
rsc = rsc->parent;
}
}
if (rsc == NULL || rsc->rsc_cons_lhs == NULL) {
return;
}
/* if rsc has children, all the children need to have start set to
* unrunnable before we follow the colo chain for the parent. */
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *)gIter->data;
action_t *start = find_first_action(child->actions, NULL, RSC_START, NULL);
if (start == NULL || is_set(start->flags, pe_action_runnable)) {
return;
}
}
for (gIter = rsc->rsc_cons_lhs; gIter != NULL; gIter = gIter->next) {
rsc_colocation_t *colocate_with = (rsc_colocation_t *)gIter->data;
if (colocate_with->score == INFINITY) {
mark_start_blocked(colocate_with->rsc_lh, action->rsc, data_set);
}
}
}
gboolean
update_action(pe_action_t *then, pe_working_set_t *data_set)
{
GListPtr lpc = NULL;
enum pe_graph_flags changed = pe_graph_none;
int last_flags = then->flags;
crm_trace("Processing %s (%s %s %s)",
then->uuid,
is_set(then->flags, pe_action_optional) ? "optional" : "required",
is_set(then->flags, pe_action_runnable) ? "runnable" : "unrunnable",
is_set(then->flags,
pe_action_pseudo) ? "pseudo" : then->node ? then->node->details->uname : "");
if (is_set(then->flags, pe_action_requires_any)) {
/* initialize current known runnable before actions to 0
* from here as graph_update_action is called for each of
* then's before actions, this number will increment as
* runnable 'first' actions are encountered */
then->runnable_before = 0;
/* for backwards compatibility with previous options that use
* the 'requires_any' flag, initialize required to 1 if it is
* not set. */
if (then->required_runnable_before == 0) {
then->required_runnable_before = 1;
}
pe_clear_action_bit(then, pe_action_runnable);
/* We are relying on the pe_order_one_or_more clause of
* graph_update_action(), called as part of the:
*
* 'if (first == other->action)'
*
* block below, to set this back if appropriate
*/
}
for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) {
action_wrapper_t *other = (action_wrapper_t *) lpc->data;
action_t *first = other->action;
node_t *then_node = then->node;
node_t *first_node = first->node;
enum pe_action_flags then_flags = 0;
enum pe_action_flags first_flags = 0;
if (first->rsc && first->rsc->variant == pe_group && safe_str_eq(first->task, RSC_START)) {
first_node = first->rsc->fns->location(first->rsc, NULL, FALSE);
if (first_node) {
crm_trace("First: Found node %s for %s", first_node->details->uname, first->uuid);
}
}
if (then->rsc && then->rsc->variant == pe_group && safe_str_eq(then->task, RSC_START)) {
then_node = then->rsc->fns->location(then->rsc, NULL, FALSE);
if (then_node) {
crm_trace("Then: Found node %s for %s", then_node->details->uname, then->uuid);
}
}
/* Disable constraint if it only applies when on same node, but isn't */
if (is_set(other->type, pe_order_same_node) && first_node && then_node
&& (first_node->details != then_node->details)) {
crm_trace("Disabled constraint %s on %s -> %s on %s",
other->action->uuid, first_node->details->uname,
then->uuid, then_node->details->uname);
other->type = pe_order_none;
continue;
}
clear_bit(changed, pe_graph_updated_first);
if (first->rsc && is_set(other->type, pe_order_then_cancels_first)
&& is_not_set(then->flags, pe_action_optional)) {
/* 'then' is required, so we must abandon 'first'
* (e.g. a required stop cancels any reload).
- * Only used with reload actions as 'first'.
*/
set_bit(other->action->flags, pe_action_optional);
- clear_bit(first->rsc->flags, pe_rsc_reload);
+ if (!strcmp(first->task, CRMD_ACTION_RELOAD)) {
+ clear_bit(first->rsc->flags, pe_rsc_reload);
+ }
}
if (first->rsc && then->rsc && (first->rsc != then->rsc)
&& (is_parent(then->rsc, first->rsc) == FALSE)) {
first = rsc_expand_action(first);
}
if (first != other->action) {
crm_trace("Ordering %s after %s instead of %s", then->uuid, first->uuid,
other->action->uuid);
}
first_flags = get_action_flags(first, then_node);
then_flags = get_action_flags(then, first_node);
crm_trace("Checking %s (%s %s %s) against %s (%s %s %s) filter=0x%.6x type=0x%.6x",
then->uuid,
is_set(then_flags, pe_action_optional) ? "optional" : "required",
is_set(then_flags, pe_action_runnable) ? "runnable" : "unrunnable",
is_set(then_flags,
pe_action_pseudo) ? "pseudo" : then->node ? then->node->details->
uname : "", first->uuid, is_set(first_flags,
pe_action_optional) ? "optional" : "required",
is_set(first_flags, pe_action_runnable) ? "runnable" : "unrunnable",
is_set(first_flags,
pe_action_pseudo) ? "pseudo" : first->node ? first->node->details->
uname : "", first_flags, other->type);
if (first == other->action) {
/*
* 'first' was not expanded (e.g. from 'start' to 'running'), which could mean it:
* - has no associated resource,
* - was a primitive,
* - was pre-expanded (e.g. 'running' instead of 'start')
*
* The third argument here to graph_update_action() is a node which is used under two conditions:
* - Interleaving, in which case first->node and
* then->node are equal (and NULL)
* - If 'then' is a clone, to limit the scope of the
* constraint to instances on the supplied node
*
*/
node_t *node = then->node;
changed |= graph_update_action(first, then, node, first_flags,
then_flags, other, data_set);
/* 'first' was for a complex resource (clone, group, etc),
* create a new dependency if necessary
*/
} else if (order_actions(first, then, other->type)) {
/* This was the first time 'first' and 'then' were associated,
* start again to get the new actions_before list
*/
changed |= (pe_graph_updated_then | pe_graph_disable);
}
if (changed & pe_graph_disable) {
crm_trace("Disabled constraint %s -> %s in favor of %s -> %s",
other->action->uuid, then->uuid, first->uuid, then->uuid);
clear_bit(changed, pe_graph_disable);
other->type = pe_order_none;
}
if (changed & pe_graph_updated_first) {
GListPtr lpc2 = NULL;
crm_trace("Updated %s (first %s %s %s), processing dependents ",
first->uuid,
is_set(first->flags, pe_action_optional) ? "optional" : "required",
is_set(first->flags, pe_action_runnable) ? "runnable" : "unrunnable",
is_set(first->flags,
pe_action_pseudo) ? "pseudo" : first->node ? first->node->details->
uname : "");
for (lpc2 = first->actions_after; lpc2 != NULL; lpc2 = lpc2->next) {
action_wrapper_t *other = (action_wrapper_t *) lpc2->data;
update_action(other->action, data_set);
}
update_action(first, data_set);
}
}
if (is_set(then->flags, pe_action_requires_any)) {
if (last_flags != then->flags) {
changed |= pe_graph_updated_then;
} else {
clear_bit(changed, pe_graph_updated_then);
}
}
if (changed & pe_graph_updated_then) {
crm_trace("Updated %s (then %s %s %s), processing dependents ",
then->uuid,
is_set(then->flags, pe_action_optional) ? "optional" : "required",
is_set(then->flags, pe_action_runnable) ? "runnable" : "unrunnable",
is_set(then->flags,
pe_action_pseudo) ? "pseudo" : then->node ? then->node->details->
uname : "");
if (is_set(last_flags, pe_action_runnable) && is_not_set(then->flags, pe_action_runnable)) {
update_colo_start_chain(then, data_set);
}
update_action(then, data_set);
for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) {
action_wrapper_t *other = (action_wrapper_t *) lpc->data;
update_action(other->action, data_set);
}
}
return FALSE;
}
gboolean
shutdown_constraints(node_t * node, action_t * shutdown_op, pe_working_set_t * data_set)
{
/* add the stop to the before lists so it counts as a pre-req
* for the shutdown
*/
GListPtr lpc = NULL;
for (lpc = data_set->actions; lpc != NULL; lpc = lpc->next) {
action_t *action = (action_t *) lpc->data;
if (action->rsc == NULL || action->node == NULL) {
continue;
} else if (action->node->details != node->details) {
continue;
} else if (is_set(action->rsc->flags, pe_rsc_maintenance)) {
pe_rsc_trace(action->rsc, "Skipping %s: maintenance mode", action->uuid);
continue;
} else if (node->details->maintenance) {
pe_rsc_trace(action->rsc, "Skipping %s: node %s is in maintenance mode",
action->uuid, node->details->uname);
continue;
} else if (safe_str_neq(action->task, RSC_STOP)) {
continue;
} else if (is_not_set(action->rsc->flags, pe_rsc_managed)
&& is_not_set(action->rsc->flags, pe_rsc_block)) {
/*
* If another action depends on this one, we may still end up blocking
*/
pe_rsc_trace(action->rsc, "Skipping %s: unmanaged", action->uuid);
continue;
}
pe_rsc_trace(action->rsc, "Ordering %s before shutdown on %s", action->uuid,
node->details->uname);
pe_clear_action_bit(action, pe_action_optional);
custom_action_order(action->rsc, NULL, action,
NULL, strdup(CRM_OP_SHUTDOWN), shutdown_op,
pe_order_optional | pe_order_runnable_left, data_set);
}
return TRUE;
}
/*!
* \internal
* \brief Order all actions appropriately relative to a fencing operation
*
* Ensure start operations of affected resources are ordered after fencing,
* imply stop and demote operations of affected resources by marking them as
* pseudo-actions, etc.
*
* \param[in] stonith_op Fencing operation
* \param[in,out] data_set Working set of cluster
*/
void
pcmk__order_vs_fence(pe_action_t *stonith_op, pe_working_set_t *data_set)
{
CRM_CHECK(stonith_op && data_set, return);
for (GList *r = data_set->resources; r != NULL; r = r->next) {
rsc_stonith_ordering((pe_resource_t *) r->data, stonith_op, data_set);
}
}
static node_t *
get_router_node(action_t *action)
{
node_t *began_on = NULL;
node_t *ended_on = NULL;
node_t *router_node = NULL;
bool partial_migration = FALSE;
const char *task = action->task;
if (safe_str_eq(task, CRM_OP_FENCE)
|| !pe__is_guest_or_remote_node(action->node)) {
return NULL;
}
CRM_ASSERT(action->node->details->remote_rsc != NULL);
began_on = pe__current_node(action->node->details->remote_rsc);
ended_on = action->node->details->remote_rsc->allocated_to;
if (action->node->details->remote_rsc
&& (action->node->details->remote_rsc->container == NULL)
&& action->node->details->remote_rsc->partial_migration_target) {
partial_migration = TRUE;
}
/* if there is only one location to choose from,
* this is easy. Check for those conditions first */
if (!began_on || !ended_on) {
/* remote rsc is either shutting down or starting up */
return began_on ? began_on : ended_on;
} else if (began_on->details == ended_on->details) {
/* remote rsc didn't move nodes. */
return began_on;
}
/* If we have get here, we know the remote resource
* began on one node and is moving to another node.
*
* This means some actions will get routed through the cluster
* node the connection rsc began on, and others are routed through
* the cluster node the connection rsc ends up on.
*
* 1. stop, demote, migrate actions of resources living in the remote
* node _MUST_ occur _BEFORE_ the connection can move (these actions
* are all required before the remote rsc stop action can occur.) In
* this case, we know these actions have to be routed through the initial
* cluster node the connection resource lived on before the move takes place.
* The exception is a partial migration of a (non-guest) remote
* connection resource; in that case, all actions (even these) will be
* ordered after the connection's pseudo-start on the migration target,
* so the target is the router node.
*
* 2. Everything else (start, promote, monitor, probe, refresh, clear failcount
* delete ....) must occur after the resource starts on the node it is
* moving to.
*/
if (safe_str_eq(task, "notify")) {
task = g_hash_table_lookup(action->meta, "notify_operation");
}
/* 1. before connection rsc moves. */
if ((safe_str_eq(task, "stop") ||
safe_str_eq(task, "demote") ||
safe_str_eq(task, "migrate_from") ||
safe_str_eq(task, "migrate_to")) && !partial_migration) {
router_node = began_on;
/* 2. after connection rsc moves. */
} else {
router_node = ended_on;
}
return router_node;
}
/*!
* \internal
* \brief Add an XML node tag for a specified ID
*
* \param[in] id Node UUID to add
* \param[in,out] xml Parent XML tag to add to
*/
static xmlNode*
add_node_to_xml_by_id(const char *id, xmlNode *xml)
{
xmlNode *node_xml;
node_xml = create_xml_node(xml, XML_CIB_TAG_NODE);
crm_xml_add(node_xml, XML_ATTR_UUID, id);
return node_xml;
}
/*!
* \internal
* \brief Add an XML node tag for a specified node
*
* \param[in] node Node to add
* \param[in,out] xml XML to add node to
*/
static void
add_node_to_xml(const node_t *node, void *xml)
{
add_node_to_xml_by_id(node->details->id, (xmlNode *) xml);
}
/*!
* \internal
* \brief Add XML with nodes that need an update of their maintenance state
*
* \param[in,out] xml Parent XML tag to add to
* \param[in] data_set Working set for cluster
*/
static int
add_maintenance_nodes(xmlNode *xml, const pe_working_set_t *data_set)
{
GListPtr gIter = NULL;
xmlNode *maintenance =
xml?create_xml_node(xml, XML_GRAPH_TAG_MAINTENANCE):NULL;
int count = 0;
for (gIter = data_set->nodes; gIter != NULL;
gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
struct pe_node_shared_s *details = node->details;
if (!pe__is_guest_or_remote_node(node)) {
continue; /* just remote nodes need to know atm */
}
if (details->maintenance != details->remote_maintenance) {
if (maintenance) {
crm_xml_add(
add_node_to_xml_by_id(node->details->id, maintenance),
XML_NODE_IS_MAINTENANCE, details->maintenance?"1":"0");
}
count++;
}
}
crm_trace("%s %d nodes to adjust maintenance-mode "
"to transition", maintenance?"Added":"Counted", count);
return count;
}
/*!
* \internal
* \brief Add pseudo action with nodes needing maintenance state update
*
* \param[in,out] data_set Working set for cluster
*/
void
add_maintenance_update(pe_working_set_t *data_set)
{
action_t *action = NULL;
if (add_maintenance_nodes(NULL, data_set)) {
crm_trace("adding maintenance state update pseudo action");
action = get_pseudo_op(CRM_OP_MAINTENANCE_NODES, data_set);
set_bit(action->flags, pe_action_print_always);
}
}
/*!
* \internal
* \brief Add XML with nodes that an action is expected to bring down
*
* If a specified action is expected to bring any nodes down, add an XML block
* with their UUIDs. When a node is lost, this allows the controller to
* determine whether it was expected.
*
* \param[in,out] xml Parent XML tag to add to
* \param[in] action Action to check for downed nodes
* \param[in] data_set Working set for cluster
*/
static void
add_downed_nodes(xmlNode *xml, const action_t *action,
const pe_working_set_t *data_set)
{
CRM_CHECK(xml && action && action->node && data_set, return);
if (safe_str_eq(action->task, CRM_OP_SHUTDOWN)) {
/* Shutdown makes the action's node down */
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
} else if (safe_str_eq(action->task, CRM_OP_FENCE)) {
/* Fencing makes the action's node and any hosted guest nodes down */
const char *fence = g_hash_table_lookup(action->meta, "stonith_action");
if (safe_str_eq(fence, "off") || safe_str_eq(fence, "reboot")) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
pe_foreach_guest_node(data_set, action->node, add_node_to_xml, downed);
}
} else if (action->rsc && action->rsc->is_remote_node
&& safe_str_eq(action->task, CRMD_ACTION_STOP)) {
/* Stopping a remote connection resource makes connected node down,
* unless it's part of a migration
*/
GListPtr iter;
action_t *input;
gboolean migrating = FALSE;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((action_wrapper_t *) iter->data)->action;
if (input->rsc && safe_str_eq(action->rsc->id, input->rsc->id)
&& safe_str_eq(input->task, CRMD_ACTION_MIGRATED)) {
migrating = TRUE;
break;
}
}
if (!migrating) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->rsc->id, downed);
}
}
}
+static bool
+should_lock_action(pe_action_t *action)
+{
+ // Only actions taking place on resource's lock node are locked
+ if ((action->rsc->lock_node == NULL) || (action->node == NULL)
+ || (action->node->details != action->rsc->lock_node->details)) {
+ return false;
+ }
+
+ /* During shutdown, only stops are locked (otherwise, another action such as
+ * a demote would cause the controller to clear the lock)
+ */
+ if (action->node->details->shutdown && action->task
+ && strcmp(action->task, RSC_STOP)) {
+ return false;
+ }
+
+ return true;
+}
+
static xmlNode *
action2xml(action_t * action, gboolean as_input, pe_working_set_t *data_set)
{
gboolean needs_node_info = TRUE;
gboolean needs_maintenance_info = FALSE;
xmlNode *action_xml = NULL;
xmlNode *args_xml = NULL;
#if ENABLE_VERSIONED_ATTRS
pe_rsc_action_details_t *rsc_details = NULL;
#endif
if (action == NULL) {
return NULL;
}
if (safe_str_eq(action->task, CRM_OP_FENCE)) {
/* All fences need node info; guest node fences are pseudo-events */
action_xml = create_xml_node(NULL,
is_set(action->flags, pe_action_pseudo)?
XML_GRAPH_TAG_PSEUDO_EVENT :
XML_GRAPH_TAG_CRM_EVENT);
} else if (safe_str_eq(action->task, CRM_OP_SHUTDOWN)) {
action_xml = create_xml_node(NULL, XML_GRAPH_TAG_CRM_EVENT);
} else if (safe_str_eq(action->task, CRM_OP_CLEAR_FAILCOUNT)) {
action_xml = create_xml_node(NULL, XML_GRAPH_TAG_CRM_EVENT);
} else if (safe_str_eq(action->task, CRM_OP_LRM_REFRESH)) {
action_xml = create_xml_node(NULL, XML_GRAPH_TAG_CRM_EVENT);
+ } else if (safe_str_eq(action->task, CRM_OP_LRM_DELETE)) {
+ // CIB-only clean-up for shutdown locks
+ action_xml = create_xml_node(NULL, XML_GRAPH_TAG_CRM_EVENT);
+ crm_xml_add(action_xml, PCMK__XA_MODE, XML_TAG_CIB);
+
/* } else if(safe_str_eq(action->task, RSC_PROBED)) { */
/* action_xml = create_xml_node(NULL, XML_GRAPH_TAG_CRM_EVENT); */
} else if (is_set(action->flags, pe_action_pseudo)) {
if (safe_str_eq(action->task, CRM_OP_MAINTENANCE_NODES)) {
needs_maintenance_info = TRUE;
}
action_xml = create_xml_node(NULL, XML_GRAPH_TAG_PSEUDO_EVENT);
needs_node_info = FALSE;
} else {
action_xml = create_xml_node(NULL, XML_GRAPH_TAG_RSC_OP);
+
#if ENABLE_VERSIONED_ATTRS
rsc_details = pe_rsc_action_details(action);
#endif
}
crm_xml_add_int(action_xml, XML_ATTR_ID, action->id);
crm_xml_add(action_xml, XML_LRM_ATTR_TASK, action->task);
if (action->rsc != NULL && action->rsc->clone_name != NULL) {
char *clone_key = NULL;
const char *interval_ms_s = g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS);
guint interval_ms = crm_parse_ms(interval_ms_s);
if (safe_str_eq(action->task, RSC_NOTIFY)) {
const char *n_type = g_hash_table_lookup(action->meta, "notify_type");
const char *n_task = g_hash_table_lookup(action->meta, "notify_operation");
CRM_CHECK(n_type != NULL, crm_err("No notify type value found for %s", action->uuid));
CRM_CHECK(n_task != NULL,
crm_err("No notify operation value found for %s", action->uuid));
clone_key = generate_notify_key(action->rsc->clone_name, n_type, n_task);
} else if(action->cancel_task) {
clone_key = generate_op_key(action->rsc->clone_name,
action->cancel_task, interval_ms);
} else {
clone_key = generate_op_key(action->rsc->clone_name,
action->task, interval_ms);
}
CRM_CHECK(clone_key != NULL, crm_err("Could not generate a key for %s", action->uuid));
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, clone_key);
crm_xml_add(action_xml, "internal_" XML_LRM_ATTR_TASK_KEY, action->uuid);
free(clone_key);
} else {
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, action->uuid);
}
if (needs_node_info && action->node != NULL) {
node_t *router_node = get_router_node(action);
crm_xml_add(action_xml, XML_LRM_ATTR_TARGET, action->node->details->uname);
crm_xml_add(action_xml, XML_LRM_ATTR_TARGET_UUID, action->node->details->id);
if (router_node) {
crm_xml_add(action_xml, XML_LRM_ATTR_ROUTER_NODE, router_node->details->uname);
}
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET), strdup(action->node->details->uname));
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET_UUID), strdup(action->node->details->id));
}
/* No details if this action is only being listed in the inputs section */
if (as_input) {
return action_xml;
}
if (action->rsc && is_not_set(action->flags, pe_action_pseudo)) {
int lpc = 0;
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER,
XML_ATTR_TYPE
};
+ /* If a resource is locked to a node via shutdown-lock, mark its actions
+ * so the controller can preserve the lock when the action completes.
+ */
+ if (should_lock_action(action)) {
+ crm_xml_add_ll(action_xml, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
+ (long long) action->rsc->lock_time);
+ }
+
// List affected resource
rsc_xml = create_xml_node(action_xml,
crm_element_name(action->rsc->xml));
if (is_set(action->rsc->flags, pe_rsc_orphan)
&& action->rsc->clone_name) {
/* Do not use the 'instance free' name here as that
* might interfere with the instance we plan to keep.
* Ie. if there are more than two named /anonymous/
* instances on a given node, we need to make sure the
* command goes to the right one.
*
* Keep this block, even when everyone is using
* 'instance free' anonymous clone names - it means
* we'll do the right thing if anyone toggles the
* unique flag to 'off'
*/
crm_debug("Using orphan clone name %s instead of %s", action->rsc->id,
action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
} else if (is_not_set(action->rsc->flags, pe_rsc_unique)) {
const char *xml_id = ID(action->rsc->xml);
crm_debug("Using anonymous clone name %s for %s (aka. %s)", xml_id, action->rsc->id,
action->rsc->clone_name);
/* ID is what we'd like client to use
* ID_LONG is what they might know it as instead
*
* ID_LONG is only strictly needed /here/ during the
* transition period until all nodes in the cluster
* are running the new software /and/ have rebooted
* once (meaning that they've only ever spoken to a DC
* supporting this feature).
*
* If anyone toggles the unique flag to 'on', the
* 'instance free' name will correspond to an orphan
* and fall into the clause above instead
*/
crm_xml_add(rsc_xml, XML_ATTR_ID, xml_id);
if (action->rsc->clone_name && safe_str_neq(xml_id, action->rsc->clone_name)) {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->clone_name);
} else {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
}
} else {
CRM_ASSERT(action->rsc->clone_name == NULL);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->id);
}
for (lpc = 0; lpc < DIMOF(attr_list); lpc++) {
crm_xml_add(rsc_xml, attr_list[lpc],
g_hash_table_lookup(action->rsc->meta, attr_list[lpc]));
}
}
/* List any attributes in effect */
args_xml = create_xml_node(NULL, XML_TAG_ATTRS);
crm_xml_add(args_xml, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
g_hash_table_foreach(action->extra, hash2field, args_xml);
if (action->rsc != NULL && action->node) {
GHashTable *p = crm_str_table_new();
get_rsc_attributes(p, action->rsc, action->node, data_set);
g_hash_table_foreach(p, hash2smartfield, args_xml);
g_hash_table_destroy(p);
#if ENABLE_VERSIONED_ATTRS
{
xmlNode *versioned_parameters = create_xml_node(NULL, XML_TAG_RSC_VER_ATTRS);
pe_get_versioned_attributes(versioned_parameters, action->rsc,
action->node, data_set);
if (xml_has_children(versioned_parameters)) {
add_node_copy(action_xml, versioned_parameters);
}
free_xml(versioned_parameters);
}
#endif
} else if(action->rsc && action->rsc->variant <= pe_native) {
g_hash_table_foreach(action->rsc->parameters, hash2smartfield, args_xml);
#if ENABLE_VERSIONED_ATTRS
if (xml_has_children(action->rsc->versioned_parameters)) {
add_node_copy(action_xml, action->rsc->versioned_parameters);
}
#endif
}
#if ENABLE_VERSIONED_ATTRS
if (rsc_details) {
if (xml_has_children(rsc_details->versioned_parameters)) {
add_node_copy(action_xml, rsc_details->versioned_parameters);
}
if (xml_has_children(rsc_details->versioned_meta)) {
add_node_copy(action_xml, rsc_details->versioned_meta);
}
}
#endif
g_hash_table_foreach(action->meta, hash2metafield, args_xml);
if (action->rsc != NULL) {
const char *value = g_hash_table_lookup(action->rsc->meta, "external-ip");
resource_t *parent = action->rsc;
while (parent != NULL) {
parent->cmds->append_meta(parent, args_xml);
parent = parent->parent;
}
if(value) {
hash2smartfield((gpointer)"pcmk_external_ip", (gpointer)value, (gpointer)args_xml);
}
if (pe__is_guest_node(action->node)) {
pe_node_t *host = NULL;
enum action_tasks task = text2task(action->task);
if(task == action_notify || task == action_notified) {
const char *n_task = g_hash_table_lookup(action->meta, "notify_operation");
task = text2task(n_task);
}
// Differentiate between up and down actions
switch (task) {
case stop_rsc:
case stopped_rsc:
case action_demote:
case action_demoted:
host = pe__current_node(action->node->details->remote_rsc->container);
break;
case start_rsc:
case started_rsc:
case monitor_rsc:
case action_promote:
case action_promoted:
host = action->node->details->remote_rsc->container->allocated_to;
break;
default:
break;
}
if(host) {
hash2metafield((gpointer)XML_RSC_ATTR_TARGET,
(gpointer)g_hash_table_lookup(action->rsc->meta, XML_RSC_ATTR_TARGET), (gpointer)args_xml);
hash2metafield((gpointer)PCMK_ENV_PHYSICAL_HOST, (gpointer)host->details->uname, (gpointer)args_xml);
}
}
} else if (safe_str_eq(action->task, CRM_OP_FENCE) && action->node) {
/* Pass the node's attributes as meta-attributes.
*
* @TODO: Determine whether it is still necessary to do this. It was
* added in 33d99707, probably for the libfence-based implementation in
* c9a90bd, which is no longer used.
*/
g_hash_table_foreach(action->node->details->attrs, hash2metafield, args_xml);
}
sorted_xml(args_xml, action_xml, FALSE);
free_xml(args_xml);
/* List any nodes this action is expected to make down */
if (needs_node_info && (action->node != NULL)) {
add_downed_nodes(action_xml, action, data_set);
}
if (needs_maintenance_info) {
add_maintenance_nodes(action_xml, data_set);
}
crm_log_xml_trace(action_xml, "dumped action");
return action_xml;
}
static bool
should_dump_action(pe_action_t *action)
{
CRM_CHECK(action != NULL, return false);
if (is_set(action->flags, pe_action_dumped)) {
crm_trace("Action %s (%d) already dumped", action->uuid, action->id);
return false;
} else if (is_set(action->flags, pe_action_pseudo)
&& safe_str_eq(action->task, CRM_OP_PROBED)) {
GListPtr lpc = NULL;
/* This is a horrible but convenient hack
*
* It mimimizes the number of actions with unsatisfied inputs
* (i.e. not included in the graph)
*
* This in turn, means we can be more concise when printing
* aborted/incomplete graphs.
*
* It also makes it obvious which node is preventing
* probe_complete from running (presumably because it is only
* partially up)
*
* For these reasons we tolerate such perversions
*/
for (lpc = action->actions_after; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *wrapper = (pe_action_wrapper_t *) lpc->data;
if (is_not_set(wrapper->action->flags, pe_action_runnable)) {
/* Only interested in runnable operations */
} else if (safe_str_neq(wrapper->action->task, RSC_START)) {
/* Only interested in start operations */
} else if (is_set(wrapper->action->flags, pe_action_dumped)
|| should_dump_action(wrapper->action)) {
crm_trace("Action %s (%d) should be dumped: "
"dependency of %s (%d)",
action->uuid, action->id,
wrapper->action->uuid, wrapper->action->id);
return true;
}
}
}
if (is_not_set(action->flags, pe_action_runnable)) {
crm_trace("Ignoring action %s (%d): unrunnable",
action->uuid, action->id);
return false;
} else if (is_set(action->flags, pe_action_optional)
&& is_not_set(action->flags, pe_action_print_always)) {
crm_trace("Ignoring action %s (%d): optional",
action->uuid, action->id);
return false;
// Monitors should be dumped even for unmanaged resources
} else if (action->rsc && is_not_set(action->rsc->flags, pe_rsc_managed)
&& safe_str_neq(action->task, RSC_STATUS)) {
const char *interval_ms_s = g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS);
// Cancellation of recurring monitors should still be dumped
if ((interval_ms_s == NULL) || !strcmp(interval_ms_s, "0")) {
crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)",
action->uuid, action->id, action->rsc->id);
return false;
}
}
if (is_set(action->flags, pe_action_pseudo)
|| safe_str_eq(action->task, CRM_OP_FENCE)
|| safe_str_eq(action->task, CRM_OP_SHUTDOWN)) {
/* skip the next checks */
return true;
}
if (action->node == NULL) {
pe_err("Skipping action %s (%d) "
"because it was not allocated to a node (bug?)",
action->uuid, action->id);
log_action(LOG_DEBUG, "Unallocated action", action, false);
return false;
+ } else if (is_set(action->flags, pe_action_dc)) {
+ crm_trace("Action %s (%d) should be dumped: "
+ "can run on DC instead of %s",
+ action->uuid, action->id, action->node->details->uname);
+
} else if (pe__is_guest_node(action->node)
&& !action->node->details->remote_requires_reset) {
crm_trace("Action %s (%d) should be dumped: "
"assuming will be runnable on guest node %s",
action->uuid, action->id, action->node->details->uname);
} else if (action->node->details->online == false) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for offline node (bug?)",
action->uuid, action->id);
log_action(LOG_DEBUG, "Action for offline node", action, FALSE);
return false;
#if 0
/* but this would also affect resources that can be safely
* migrated before a fencing op
*/
} else if (action->node->details->unclean == false) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for unclean node (bug?)",
action->uuid, action->id);
log_action(LOG_DEBUG, "Action for unclean node", action, false);
return false;
#endif
}
return true;
}
/* lowest to highest */
static gint
sort_action_id(gconstpointer a, gconstpointer b)
{
const action_wrapper_t *action_wrapper2 = (const action_wrapper_t *)a;
const action_wrapper_t *action_wrapper1 = (const action_wrapper_t *)b;
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (action_wrapper1->action->id > action_wrapper2->action->id) {
return -1;
}
if (action_wrapper1->action->id < action_wrapper2->action->id) {
return 1;
}
return 0;
}
/*!
* \internal
* \brief Check whether an action input should be in the transition graph
*
* \param[in] action Action to check
* \param[in,out] input Action input to check
*
* \return true if input should be in graph, false otherwise
* \note This function may not only check an input, but disable it under certian
* circumstances (load or anti-colocation orderings that are not needed).
*/
static bool
check_dump_input(pe_action_t *action, pe_action_wrapper_t *input)
{
int type = input->type;
if (input->state == pe_link_dumped) {
return true;
}
type &= ~pe_order_implies_first_printed;
type &= ~pe_order_implies_then_printed;
type &= ~pe_order_optional;
if (input->type == pe_order_none) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering disabled",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (is_not_set(input->action->flags, pe_action_runnable)
&& (type == pe_order_none)
&& safe_str_neq(input->action->uuid, CRM_OP_PROBED)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (is_not_set(input->action->flags, pe_action_runnable)
&& is_set(input->type, pe_order_one_or_more)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"one-or-more and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (is_set(action->flags, pe_action_pseudo)
&& is_set(input->type, pe_order_stonith_stop)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"stonith stop but action is pseudo",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (is_set(input->type, pe_order_implies_first_migratable)
&& is_not_set(input->action->flags, pe_action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"implies input migratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (is_set(input->type, pe_order_apply_first_non_migratable)
&& is_set(input->action->flags, pe_action_migrate_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"only if input unmigratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->type == pe_order_optional)
&& is_set(input->action->flags, pe_action_migrate_runnable)
&& crm_ends_with(input->action->uuid, "_stop_0")) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional but stop in migration",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (input->type == pe_order_load) {
pe_node_t *input_node = input->action->node;
// load orderings are relevant only if actions are for same node
if (action->rsc && safe_str_eq(action->task, RSC_MIGRATE)) {
pe_node_t *allocated = action->rsc->allocated_to;
/* For load_stopped -> migrate_to orderings, we care about where it
* has been allocated to, not where it will be executed.
*/
if ((input_node == NULL) || (allocated == NULL)
|| (input_node->details != allocated->details)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(allocated? allocated->details->uname : ""),
(input_node? input_node->details->uname : ""));
input->type = pe_order_none;
return false;
}
} else if ((input_node == NULL) || (action->node == NULL)
|| (input_node->details != action->node->details)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->details->uname : ""),
(input_node? input_node->details->uname : ""));
input->type = pe_order_none;
return false;
} else if (is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->type == pe_order_anti_colocation) {
if (input->action->node && action->node
&& (input->action->node->details != action->node->details)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
action->node->details->uname,
input->action->node->details->uname);
input->type = pe_order_none;
return false;
} else if (is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& is_set(input->action->rsc->flags, pe_rsc_failed)
&& is_not_set(input->action->rsc->flags, pe_rsc_managed)
&& crm_ends_with(input->action->uuid, "_stop_0")
&& action->rsc && pe_rsc_is_clone(action->rsc)) {
crm_warn("Ignoring requirement that %s complete before %s:"
" unmanaged failed resources cannot prevent clone shutdown",
input->action->uuid, action->uuid);
return false;
} else if (is_set(input->action->flags, pe_action_optional)
&& is_not_set(input->action->flags, pe_action_print_always)
&& is_not_set(input->action->flags, pe_action_dumped)
&& !should_dump_action(input->action)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
}
crm_trace("%s (%d) input %s (%d) @ %s should be dumped: %s, %s, %s, 0x%.6x",
action->uuid, action->id,
input->action->uuid, input->action->id,
input->action->node? input->action->node->details->uname : "no node",
is_set(input->action->flags, pe_action_pseudo)? "pseudo" : "real",
is_set(input->action->flags, pe_action_runnable)? "runnable" : "unrunnable",
is_set(input->action->flags, pe_action_optional)? "optional" : "required",
input->type);
return true;
}
static bool
graph_has_loop(pe_action_t *init_action, pe_action_t *action,
pe_action_wrapper_t *input)
{
bool has_loop = false;
if (is_set(input->action->flags, pe_action_tracking)) {
crm_trace("Breaking tracking loop: %s@%s -> %s@%s (0x%.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
return false;
}
// Don't need to check inputs that won't be used
if (!check_dump_input(action, input)) {
return false;
}
if (input->action == init_action) {
crm_debug("Input loop found in %s@%s ->...-> %s@%s",
action->uuid,
action->node? action->node->details->uname : "",
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
return true;
}
set_bit(input->action->flags, pe_action_tracking);
crm_trace("Checking inputs of action %s@%s input %s@%s (0x%.6x)"
"for graph loop with %s@%s ",
action->uuid,
action->node? action->node->details->uname : "",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
input->type,
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
// Recursively check input itself for loops
for (GList *iter = input->action->actions_before;
iter != NULL; iter = iter->next) {
if (graph_has_loop(init_action, input->action,
(pe_action_wrapper_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
goto done;
}
}
done:
pe_clear_action_bit(input->action, pe_action_tracking);
if (!has_loop) {
crm_trace("No input loop found in %s@%s -> %s@%s (0x%.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
}
return has_loop;
}
bool
pcmk__ordering_is_invalid(pe_action_t *action, pe_action_wrapper_t *input)
{
/* Prevent user-defined ordering constraints between resources
* running in a guest node and the resource that defines that node.
*/
if (is_not_set(input->type, pe_order_preserve)
&& action->rsc && action->rsc->fillers
&& input->action->rsc && input->action->node
&& input->action->node->details->remote_rsc
&& (input->action->node->details->remote_rsc->container == action->rsc)) {
crm_warn("Invalid ordering constraint between %s and %s",
input->action->rsc->id, action->rsc->id);
return true;
}
/* If there's an order like
* "rscB_stop node2"-> "load_stopped_node2" -> "rscA_migrate_to node1"
*
* then rscA is being migrated from node1 to node2, while rscB is being
* migrated from node2 to node1. If there would be a graph loop,
* break the order "load_stopped_node2" -> "rscA_migrate_to node1".
*/
if ((input->type == pe_order_load) && action->rsc
&& safe_str_eq(action->task, RSC_MIGRATE)
&& graph_has_loop(action, action, input)) {
return true;
}
return false;
}
// Remove duplicate inputs (regardless of flags)
static void
deduplicate_inputs(pe_action_t *action)
{
GList *item = NULL;
GList *next = NULL;
pe_action_wrapper_t *last_input = NULL;
action->actions_before = g_list_sort(action->actions_before,
sort_action_id);
for (item = action->actions_before; item != NULL; item = next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) item->data;
next = item->next;
if (last_input && (input->action->id == last_input->action->id)) {
crm_trace("Input %s (%d) duplicate skipped for action %s (%d)",
input->action->uuid, input->action->id,
action->uuid, action->id);
/* For the purposes of scheduling, the ordering flags no longer
* matter, but crm_simulate looks at certain ones when creating a
* dot graph. Combining the flags is sufficient for that purpose.
*/
last_input->type |= input->type;
if (input->state == pe_link_dumped) {
last_input->state = pe_link_dumped;
}
free(item->data);
action->actions_before = g_list_delete_link(action->actions_before,
item);
} else {
last_input = input;
input->state = pe_link_not_dumped;
}
}
}
/*!
* \internal
* \brief Add an action to the transition graph XML if appropriate
*
* \param[in] action Action to possibly add
* \param[in] data_set Cluster working set
*
* \note This will de-duplicate the action inputs, meaning that the
* pe_action_wrapper_t:type flags can no longer be relied on to retain
* their original settings. That means this MUST be called after stage7()
* is complete, and nothing after this should rely on those type flags.
* (For example, some code looks for type equal to some flag rather than
* whether the flag is set, and some code looks for particular
* combinations of flags -- such code must be done before stage8().)
*/
void
graph_element_from_action(pe_action_t *action, pe_working_set_t *data_set)
{
GList *lpc = NULL;
int synapse_priority = 0;
xmlNode *syn = NULL;
xmlNode *set = NULL;
xmlNode *in = NULL;
xmlNode *xml_action = NULL;
pe_action_wrapper_t *input = NULL;
/* If we haven't already, de-duplicate inputs -- even if we won't be dumping
* the action, so that crm_simulate dot graphs don't have duplicates.
*/
if (is_not_set(action->flags, pe_action_dedup)) {
deduplicate_inputs(action);
set_bit(action->flags, pe_action_dedup);
}
if (should_dump_action(action) == FALSE) {
return;
}
set_bit(action->flags, pe_action_dumped);
syn = create_xml_node(data_set->graph, "synapse");
set = create_xml_node(syn, "action_set");
in = create_xml_node(syn, "inputs");
crm_xml_add_int(syn, XML_ATTR_ID, data_set->num_synapse);
data_set->num_synapse++;
if (action->rsc != NULL) {
synapse_priority = action->rsc->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, XML_CIB_ATTR_PRIORITY, synapse_priority);
}
xml_action = action2xml(action, FALSE, data_set);
add_node_nocopy(set, crm_element_name(xml_action), xml_action);
for (lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
input = (pe_action_wrapper_t *) lpc->data;
if (check_dump_input(action, input)) {
xmlNode *input_xml = create_xml_node(in, "trigger");
input->state = pe_link_dumped;
xml_action = action2xml(input->action, TRUE, data_set);
add_node_nocopy(input_xml, crm_element_name(xml_action), xml_action);
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_native.c b/lib/pacemaker/pcmk_sched_native.c
index 9ebdd35856..714a7a0af7 100644
--- a/lib/pacemaker/pcmk_sched_native.c
+++ b/lib/pacemaker/pcmk_sched_native.c
@@ -1,3326 +1,3332 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
// The controller removes the resource from the CIB, making this redundant
// #define DELETE_THEN_REFRESH 1
#define INFINITY_HACK (INFINITY * -100)
#define VARIANT_NATIVE 1
#include
void native_rsc_colocation_rh_must(resource_t * rsc_lh, gboolean update_lh,
resource_t * rsc_rh, gboolean update_rh);
void native_rsc_colocation_rh_mustnot(resource_t * rsc_lh, gboolean update_lh,
resource_t * rsc_rh, gboolean update_rh);
static void Recurring(resource_t *rsc, action_t *start, node_t *node,
pe_working_set_t *data_set);
static void RecurringOp(resource_t *rsc, action_t *start, node_t *node,
xmlNode *operation, pe_working_set_t *data_set);
static void Recurring_Stopped(resource_t *rsc, action_t *start, node_t *node,
pe_working_set_t *data_set);
static void RecurringOp_Stopped(resource_t *rsc, action_t *start, node_t *node,
xmlNode *operation, pe_working_set_t *data_set);
void ReloadRsc(resource_t * rsc, node_t *node, pe_working_set_t * data_set);
gboolean DeleteRsc(resource_t * rsc, node_t * node, gboolean optional, pe_working_set_t * data_set);
gboolean StopRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set);
gboolean StartRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set);
gboolean DemoteRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set);
gboolean PromoteRsc(resource_t * rsc, node_t * next, gboolean optional,
pe_working_set_t * data_set);
gboolean RoleError(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set);
gboolean NullOp(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set);
/* *INDENT-OFF* */
enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* Current State */
/* Next State: Unknown Stopped Started Slave Master */
/* Unknown */ { RSC_ROLE_UNKNOWN, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, },
/* Stopped */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STARTED, RSC_ROLE_SLAVE, RSC_ROLE_SLAVE, },
/* Started */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STARTED, RSC_ROLE_SLAVE, RSC_ROLE_MASTER, },
/* Slave */ { RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_STOPPED, RSC_ROLE_SLAVE, RSC_ROLE_MASTER, },
/* Master */ { RSC_ROLE_STOPPED, RSC_ROLE_SLAVE, RSC_ROLE_SLAVE, RSC_ROLE_SLAVE, RSC_ROLE_MASTER, },
};
gboolean (*rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX])(resource_t*,node_t*,gboolean,pe_working_set_t*) = {
/* Current State */
/* Next State: Unknown Stopped Started Slave Master */
/* Unknown */ { RoleError, StopRsc, RoleError, RoleError, RoleError, },
/* Stopped */ { RoleError, NullOp, StartRsc, StartRsc, RoleError, },
/* Started */ { RoleError, StopRsc, NullOp, NullOp, PromoteRsc, },
/* Slave */ { RoleError, StopRsc, StopRsc, NullOp, PromoteRsc, },
/* Master */ { RoleError, DemoteRsc, DemoteRsc, DemoteRsc, NullOp, },
};
/* *INDENT-ON* */
static gboolean
native_choose_node(resource_t * rsc, node_t * prefer, pe_working_set_t * data_set)
{
GListPtr nodes = NULL;
node_t *chosen = NULL;
node_t *best = NULL;
int multiple = 1;
int length = 0;
gboolean result = FALSE;
process_utilization(rsc, &prefer, data_set);
if (is_not_set(rsc->flags, pe_rsc_provisional)) {
return rsc->allocated_to ? TRUE : FALSE;
}
// Sort allowed nodes by weight
if (rsc->allowed_nodes) {
length = g_hash_table_size(rsc->allowed_nodes);
}
if (length > 0) {
nodes = g_hash_table_get_values(rsc->allowed_nodes);
nodes = sort_nodes_by_weight(nodes, pe__current_node(rsc), data_set);
// First node in sorted list has the best score
best = g_list_nth_data(nodes, 0);
}
if (prefer && nodes) {
chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id);
if (chosen == NULL) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unknown",
prefer->details->uname, rsc->id);
/* Favor the preferred node as long as its weight is at least as good as
* the best allowed node's.
*
* An alternative would be to favor the preferred node even if the best
* node is better, when the best node's weight is less than INFINITY.
*/
} else if ((chosen->weight < 0) || (chosen->weight < best->weight)) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
chosen->details->uname, rsc->id);
chosen = NULL;
} else if (!can_run_resources(chosen)) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable",
chosen->details->uname, rsc->id);
chosen = NULL;
} else {
pe_rsc_trace(rsc,
"Chose preferred node %s for %s (ignoring %d candidates)",
chosen->details->uname, rsc->id, length);
}
}
if ((chosen == NULL) && nodes) {
/* Either there is no preferred node, or the preferred node is not
* available, but there are other nodes allowed to run the resource.
*/
chosen = best;
pe_rsc_trace(rsc, "Chose node %s for %s from %d candidates",
chosen ? chosen->details->uname : "", rsc->id, length);
if (!pe_rsc_is_unique_clone(rsc->parent)
&& chosen && (chosen->weight > 0) && can_run_resources(chosen)) {
/* If the resource is already running on a node, prefer that node if
* it is just as good as the chosen node.
*
* We don't do this for unique clone instances, because
* distribute_children() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unallocated instances to prefer a node that's already
* running another instance.
*/
node_t *running = pe__current_node(rsc);
if (running && (can_run_resources(running) == FALSE)) {
pe_rsc_trace(rsc, "Current node for %s (%s) can't run resources",
rsc->id, running->details->uname);
} else if (running) {
for (GList *iter = nodes->next; iter; iter = iter->next) {
node_t *tmp = (node_t *) iter->data;
if (tmp->weight != chosen->weight) {
// The nodes are sorted by weight, so no more are equal
break;
}
if (tmp->details == running->details) {
// Scores are equal, so prefer the current node
chosen = tmp;
}
multiple++;
}
}
}
}
if (multiple > 1) {
static char score[33];
int log_level = (chosen->weight >= INFINITY)? LOG_WARNING : LOG_INFO;
score2char_stack(chosen->weight, score, sizeof(score));
do_crm_log(log_level,
"Chose node %s for %s from %d nodes with score %s",
chosen->details->uname, rsc->id, multiple, score);
}
result = native_assign_node(rsc, nodes, chosen, FALSE);
g_list_free(nodes);
return result;
}
static int
node_list_attr_score(GHashTable * list, const char *attr, const char *value)
{
GHashTableIter iter;
node_t *node = NULL;
int best_score = -INFINITY;
const char *best_node = NULL;
if (attr == NULL) {
attr = CRM_ATTR_UNAME;
}
g_hash_table_iter_init(&iter, list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
int weight = node->weight;
if (can_run_resources(node) == FALSE) {
weight = -INFINITY;
}
if (weight > best_score || best_node == NULL) {
const char *tmp = pe_node_attribute_raw(node, attr);
if (safe_str_eq(value, tmp)) {
best_score = weight;
best_node = node->details->uname;
}
}
}
if (safe_str_neq(attr, CRM_ATTR_UNAME)) {
crm_info("Best score for %s=%s was %s with %d",
attr, value, best_node ? best_node : "", best_score);
}
return best_score;
}
static void
node_hash_update(GHashTable * list1, GHashTable * list2, const char *attr, float factor,
gboolean only_positive)
{
int score = 0;
int new_score = 0;
GHashTableIter iter;
node_t *node = NULL;
if (attr == NULL) {
attr = CRM_ATTR_UNAME;
}
g_hash_table_iter_init(&iter, list1);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
float weight_f = 0;
int weight = 0;
CRM_LOG_ASSERT(node != NULL);
if(node == NULL) { continue; };
score = node_list_attr_score(list2, attr, pe_node_attribute_raw(node, attr));
weight_f = factor * score;
/* Round the number */
/* http://c-faq.com/fp/round.html */
weight = (int)(weight_f < 0 ? weight_f - 0.5 : weight_f + 0.5);
new_score = merge_weights(weight, node->weight);
if (factor < 0 && score < 0) {
/* Negative preference for a node with a negative score
* should not become a positive preference
*
* TODO - Decide if we want to filter only if weight == -INFINITY
*
*/
crm_trace("%s: Filtering %d + %f*%d (factor * score)",
node->details->uname, node->weight, factor, score);
} else if (node->weight == INFINITY_HACK) {
crm_trace("%s: Filtering %d + %f*%d (node < 0)",
node->details->uname, node->weight, factor, score);
} else if (only_positive && new_score < 0 && node->weight > 0) {
node->weight = INFINITY_HACK;
crm_trace("%s: Filtering %d + %f*%d (score > 0)",
node->details->uname, node->weight, factor, score);
} else if (only_positive && new_score < 0 && node->weight == 0) {
crm_trace("%s: Filtering %d + %f*%d (score == 0)",
node->details->uname, node->weight, factor, score);
} else {
crm_trace("%s: %d + %f*%d", node->details->uname, node->weight, factor, score);
node->weight = new_score;
}
}
}
GHashTable *
node_hash_dup(GHashTable * hash)
{
/* Hack! */
GListPtr list = g_hash_table_get_values(hash);
GHashTable *result = node_hash_from_list(list);
g_list_free(list);
return result;
}
GHashTable *
native_merge_weights(resource_t * rsc, const char *rhs, GHashTable * nodes, const char *attr,
float factor, enum pe_weights flags)
{
return rsc_merge_weights(rsc, rhs, nodes, attr, factor, flags);
}
GHashTable *
rsc_merge_weights(resource_t * rsc, const char *rhs, GHashTable * nodes, const char *attr,
float factor, enum pe_weights flags)
{
GHashTable *work = NULL;
int multiplier = 1;
if (factor < 0) {
multiplier = -1;
}
if (is_set(rsc->flags, pe_rsc_merging)) {
pe_rsc_info(rsc, "%s: Breaking dependency loop at %s", rhs, rsc->id);
return nodes;
}
set_bit(rsc->flags, pe_rsc_merging);
if (is_set(flags, pe_weights_init)) {
if (rsc->variant == pe_group && rsc->children) {
GListPtr last = rsc->children;
while (last->next != NULL) {
last = last->next;
}
pe_rsc_trace(rsc, "Merging %s as a group %p %p", rsc->id, rsc->children, last);
work = rsc_merge_weights(last->data, rhs, NULL, attr, factor, flags);
} else {
work = node_hash_dup(rsc->allowed_nodes);
}
clear_bit(flags, pe_weights_init);
} else if (rsc->variant == pe_group && rsc->children) {
GListPtr iter = rsc->children;
pe_rsc_trace(rsc, "%s: Combining scores from %d children of %s", rhs, g_list_length(iter), rsc->id);
work = node_hash_dup(nodes);
for(iter = rsc->children; iter->next != NULL; iter = iter->next) {
work = rsc_merge_weights(iter->data, rhs, work, attr, factor, flags);
}
} else {
pe_rsc_trace(rsc, "%s: Combining scores from %s", rhs, rsc->id);
work = node_hash_dup(nodes);
node_hash_update(work, rsc->allowed_nodes, attr, factor,
is_set(flags, pe_weights_positive));
}
if (is_set(flags, pe_weights_rollback) && can_run_any(work) == FALSE) {
pe_rsc_info(rsc, "%s: Rolling back scores from %s", rhs, rsc->id);
g_hash_table_destroy(work);
clear_bit(rsc->flags, pe_rsc_merging);
return nodes;
}
if (can_run_any(work)) {
GListPtr gIter = NULL;
if (is_set(flags, pe_weights_forward)) {
gIter = rsc->rsc_cons;
crm_trace("Checking %d additional colocation constraints", g_list_length(gIter));
} else if(rsc->variant == pe_group && rsc->children) {
GListPtr last = rsc->children;
while (last->next != NULL) {
last = last->next;
}
gIter = ((resource_t*)last->data)->rsc_cons_lhs;
crm_trace("Checking %d additional optional group colocation constraints from %s",
g_list_length(gIter), ((resource_t*)last->data)->id);
} else {
gIter = rsc->rsc_cons_lhs;
crm_trace("Checking %d additional optional colocation constraints %s", g_list_length(gIter), rsc->id);
}
for (; gIter != NULL; gIter = gIter->next) {
resource_t *other = NULL;
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
if (is_set(flags, pe_weights_forward)) {
other = constraint->rsc_rh;
} else {
other = constraint->rsc_lh;
}
pe_rsc_trace(rsc, "Applying %s (%s)", constraint->id, other->id);
work = rsc_merge_weights(other, rhs, work, constraint->node_attribute,
multiplier * (float)constraint->score / INFINITY, flags|pe_weights_rollback);
dump_node_scores(LOG_TRACE, NULL, rhs, work);
}
}
if (is_set(flags, pe_weights_positive)) {
node_t *node = NULL;
GHashTableIter iter;
g_hash_table_iter_init(&iter, work);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if (node->weight == INFINITY_HACK) {
node->weight = 1;
}
}
}
if (nodes) {
g_hash_table_destroy(nodes);
}
clear_bit(rsc->flags, pe_rsc_merging);
return work;
}
static inline bool
node_has_been_unfenced(node_t *node)
{
const char *unfenced = pe_node_attribute_raw(node, CRM_ATTR_UNFENCED);
return unfenced && strcmp("0", unfenced);
}
static inline bool
is_unfence_device(resource_t *rsc, pe_working_set_t *data_set)
{
return is_set(rsc->flags, pe_rsc_fence_device)
&& is_set(data_set->flags, pe_flag_enable_unfencing);
}
node_t *
native_color(resource_t * rsc, node_t * prefer, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
int alloc_details = scores_log_level + 1;
if (rsc->parent && is_not_set(rsc->parent->flags, pe_rsc_allocating)) {
/* never allocate children on their own */
pe_rsc_debug(rsc, "Escalating allocation of %s to its parent: %s", rsc->id,
rsc->parent->id);
rsc->parent->cmds->allocate(rsc->parent, prefer, data_set);
}
if (is_not_set(rsc->flags, pe_rsc_provisional)) {
return rsc->allocated_to;
}
if (is_set(rsc->flags, pe_rsc_allocating)) {
pe_rsc_debug(rsc, "Dependency loop detected involving %s", rsc->id);
return NULL;
}
set_bit(rsc->flags, pe_rsc_allocating);
dump_node_scores(alloc_details, rsc, "Pre-alloc", rsc->allowed_nodes);
for (gIter = rsc->rsc_cons; gIter != NULL; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
GHashTable *archive = NULL;
resource_t *rsc_rh = constraint->rsc_rh;
pe_rsc_trace(rsc, "%s: Pre-Processing %s (%s, %d, %s)",
rsc->id, constraint->id, rsc_rh->id,
constraint->score, role2text(constraint->role_lh));
if (constraint->role_lh >= RSC_ROLE_MASTER
|| (constraint->score < 0 && constraint->score > -INFINITY)) {
archive = node_hash_dup(rsc->allowed_nodes);
}
rsc_rh->cmds->allocate(rsc_rh, NULL, data_set);
rsc->cmds->rsc_colocation_lh(rsc, rsc_rh, constraint, data_set);
if (archive && can_run_any(rsc->allowed_nodes) == FALSE) {
pe_rsc_info(rsc, "%s: Rolling back scores from %s", rsc->id, rsc_rh->id);
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = archive;
archive = NULL;
}
if (archive) {
g_hash_table_destroy(archive);
}
}
dump_node_scores(alloc_details, rsc, "Post-coloc", rsc->allowed_nodes);
for (gIter = rsc->rsc_cons_lhs; gIter != NULL; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
rsc->allowed_nodes =
constraint->rsc_lh->cmds->merge_weights(constraint->rsc_lh, rsc->id, rsc->allowed_nodes,
constraint->node_attribute,
(float)constraint->score / INFINITY,
pe_weights_rollback);
}
if (rsc->next_role == RSC_ROLE_STOPPED) {
pe_rsc_trace(rsc, "Making sure %s doesn't get allocated", rsc->id);
/* make sure it doesn't come up again */
resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE, data_set);
} else if(rsc->next_role > rsc->role
&& is_set(data_set->flags, pe_flag_have_quorum) == FALSE
&& data_set->no_quorum_policy == no_quorum_freeze) {
crm_notice("Resource %s cannot be elevated from %s to %s: no-quorum-policy=freeze",
rsc->id, role2text(rsc->role), role2text(rsc->next_role));
rsc->next_role = rsc->role;
}
dump_node_scores((show_scores? LOG_STDOUT : scores_log_level), rsc,
__FUNCTION__, rsc->allowed_nodes);
if (is_set(data_set->flags, pe_flag_stonith_enabled)
&& is_set(data_set->flags, pe_flag_have_stonith_resource) == FALSE) {
clear_bit(rsc->flags, pe_rsc_managed);
}
if (is_not_set(rsc->flags, pe_rsc_managed)) {
const char *reason = NULL;
node_t *assign_to = NULL;
rsc->next_role = rsc->role;
assign_to = pe__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->role == RSC_ROLE_MASTER) {
reason = "master";
} else if (is_set(rsc->flags, pe_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pe_rsc_info(rsc, "Unmanaged resource %s allocated to %s: %s", rsc->id,
(assign_to? assign_to->details->uname : "no node"), reason);
native_assign_node(rsc, NULL, assign_to, TRUE);
} else if (is_set(data_set->flags, pe_flag_stop_everything)) {
pe_rsc_debug(rsc, "Forcing %s to stop", rsc->id);
native_assign_node(rsc, NULL, NULL, TRUE);
} else if (is_set(rsc->flags, pe_rsc_provisional)
&& native_choose_node(rsc, prefer, data_set)) {
pe_rsc_trace(rsc, "Allocated resource %s to %s", rsc->id,
rsc->allocated_to->details->uname);
} else if (rsc->allocated_to == NULL) {
if (is_not_set(rsc->flags, pe_rsc_orphan)) {
pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if (rsc->running_on != NULL) {
pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
} else {
pe_rsc_debug(rsc, "Pre-Allocated resource %s to %s", rsc->id,
rsc->allocated_to->details->uname);
}
clear_bit(rsc->flags, pe_rsc_allocating);
if (rsc->is_remote_node) {
node_t *remote_node = pe_find_node(data_set->nodes, rsc->id);
CRM_ASSERT(remote_node != NULL);
if (rsc->allocated_to && rsc->next_role != RSC_ROLE_STOPPED) {
crm_trace("Setting Pacemaker Remote node %s to ONLINE",
remote_node->details->id);
remote_node->details->online = TRUE;
/* We shouldn't consider an unseen remote-node unclean if we are going
* to try and connect to it. Otherwise we get an unnecessary fence */
if (remote_node->details->unseen == TRUE) {
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Setting Pacemaker Remote node %s to SHUTDOWN (next role %s, %sallocated)",
remote_node->details->id, role2text(rsc->next_role),
(rsc->allocated_to? "" : "un"));
remote_node->details->shutdown = TRUE;
}
}
return rsc->allocated_to;
}
static gboolean
is_op_dup(resource_t *rsc, const char *name, guint interval_ms)
{
gboolean dup = FALSE;
const char *id = NULL;
const char *value = NULL;
xmlNode *operation = NULL;
guint interval2_ms = 0;
CRM_ASSERT(rsc);
for (operation = __xml_first_child_element(rsc->ops_xml); operation != NULL;
operation = __xml_next_element(operation)) {
if (crm_str_eq((const char *)operation->name, "op", TRUE)) {
value = crm_element_value(operation, "name");
if (safe_str_neq(value, name)) {
continue;
}
value = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
interval2_ms = crm_parse_interval_spec(value);
if (interval_ms != interval2_ms) {
continue;
}
if (id == NULL) {
id = ID(operation);
} else {
crm_config_err("Operation %s is a duplicate of %s", ID(operation), id);
crm_config_err
("Do not use the same (name, interval) combination more than once per resource");
dup = TRUE;
}
}
}
return dup;
}
static bool
op_cannot_recur(const char *name)
{
return safe_str_eq(name, RSC_STOP)
|| safe_str_eq(name, RSC_START)
|| safe_str_eq(name, RSC_DEMOTE)
|| safe_str_eq(name, RSC_PROMOTE);
}
static void
RecurringOp(resource_t * rsc, action_t * start, node_t * node,
xmlNode * operation, pe_working_set_t * data_set)
{
char *key = NULL;
const char *name = NULL;
const char *role = NULL;
const char *interval_spec = NULL;
const char *node_uname = node? node->details->uname : "n/a";
guint interval_ms = 0;
action_t *mon = NULL;
gboolean is_optional = TRUE;
GListPtr possible_matches = NULL;
CRM_ASSERT(rsc);
/* Only process for the operations without role="Stopped" */
role = crm_element_value(operation, "role");
if (role && text2role(role) == RSC_ROLE_STOPPED) {
return;
}
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
interval_ms = crm_parse_interval_spec(interval_spec);
if (interval_ms == 0) {
return;
}
name = crm_element_value(operation, "name");
if (is_op_dup(rsc, name, interval_ms)) {
crm_trace("Not creating duplicate recurring action %s for %dms %s",
ID(operation), interval_ms, name);
return;
}
if (op_cannot_recur(name)) {
crm_config_err("Ignoring %s because action '%s' cannot be recurring",
ID(operation), name);
return;
}
key = generate_op_key(rsc->id, name, interval_ms);
if (find_rsc_op_entry(rsc, key) == NULL) {
crm_trace("Not creating recurring action %s for disabled resource %s",
ID(operation), rsc->id);
free(key);
return;
}
pe_rsc_trace(rsc, "Creating recurring action %s for %s in role %s on %s",
ID(operation), rsc->id, role2text(rsc->next_role), node_uname);
if (start != NULL) {
pe_rsc_trace(rsc, "Marking %s %s due to %s",
key, is_set(start->flags, pe_action_optional) ? "optional" : "mandatory",
start->uuid);
is_optional = (rsc->cmds->action_flags(start, NULL) & pe_action_optional);
} else {
pe_rsc_trace(rsc, "Marking %s optional", key);
is_optional = TRUE;
}
/* start a monitor for an already active resource */
possible_matches = find_actions_exact(rsc->actions, key, node);
if (possible_matches == NULL) {
is_optional = FALSE;
pe_rsc_trace(rsc, "Marking %s mandatory: not active", key);
} else {
GListPtr gIter = NULL;
for (gIter = possible_matches; gIter != NULL; gIter = gIter->next) {
action_t *op = (action_t *) gIter->data;
if (is_set(op->flags, pe_action_reschedule)) {
is_optional = FALSE;
break;
}
}
g_list_free(possible_matches);
}
if ((rsc->next_role == RSC_ROLE_MASTER && role == NULL)
|| (role != NULL && text2role(role) != rsc->next_role)) {
int log_level = LOG_TRACE;
const char *result = "Ignoring";
if (is_optional) {
char *after_key = NULL;
action_t *cancel_op = NULL;
// It's running, so cancel it
log_level = LOG_INFO;
result = "Cancelling";
cancel_op = pe_cancel_op(rsc, name, interval_ms, node, data_set);
switch (rsc->role) {
case RSC_ROLE_SLAVE:
case RSC_ROLE_STARTED:
if (rsc->next_role == RSC_ROLE_MASTER) {
after_key = promote_key(rsc);
} else if (rsc->next_role == RSC_ROLE_STOPPED) {
after_key = stop_key(rsc);
}
break;
case RSC_ROLE_MASTER:
after_key = demote_key(rsc);
break;
default:
break;
}
if (after_key) {
custom_action_order(rsc, NULL, cancel_op, rsc, after_key, NULL,
pe_order_runnable_left, data_set);
}
}
do_crm_log(log_level, "%s action %s (%s vs. %s)",
result, key, role ? role : role2text(RSC_ROLE_SLAVE),
role2text(rsc->next_role));
free(key);
return;
}
mon = custom_action(rsc, key, name, node, is_optional, TRUE, data_set);
key = mon->uuid;
if (is_optional) {
pe_rsc_trace(rsc, "%s\t %s (optional)", node_uname, mon->uuid);
}
if (start == NULL || is_set(start->flags, pe_action_runnable) == FALSE) {
pe_rsc_debug(rsc, "%s\t %s (cancelled : start un-runnable)",
node_uname, mon->uuid);
update_action_flags(mon, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
} else if (node == NULL || node->details->online == FALSE || node->details->unclean) {
pe_rsc_debug(rsc, "%s\t %s (cancelled : no node available)",
node_uname, mon->uuid);
update_action_flags(mon, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
} else if (is_set(mon->flags, pe_action_optional) == FALSE) {
pe_rsc_info(rsc, " Start recurring %s (%us) for %s on %s",
mon->task, interval_ms / 1000, rsc->id, node_uname);
}
if (rsc->next_role == RSC_ROLE_MASTER) {
char *running_master = crm_itoa(PCMK_OCF_RUNNING_MASTER);
add_hash_param(mon->meta, XML_ATTR_TE_TARGET_RC, running_master);
free(running_master);
}
if (node == NULL || is_set(rsc->flags, pe_rsc_managed)) {
custom_action_order(rsc, start_key(rsc), NULL,
NULL, strdup(key), mon,
pe_order_implies_then | pe_order_runnable_left, data_set);
custom_action_order(rsc, reload_key(rsc), NULL,
NULL, strdup(key), mon,
pe_order_implies_then | pe_order_runnable_left, data_set);
if (rsc->next_role == RSC_ROLE_MASTER) {
custom_action_order(rsc, promote_key(rsc), NULL,
rsc, NULL, mon,
pe_order_optional | pe_order_runnable_left, data_set);
} else if (rsc->role == RSC_ROLE_MASTER) {
custom_action_order(rsc, demote_key(rsc), NULL,
rsc, NULL, mon,
pe_order_optional | pe_order_runnable_left, data_set);
}
}
}
static void
Recurring(resource_t * rsc, action_t * start, node_t * node, pe_working_set_t * data_set)
{
if (is_not_set(rsc->flags, pe_rsc_maintenance) &&
(node == NULL || node->details->maintenance == FALSE)) {
xmlNode *operation = NULL;
for (operation = __xml_first_child_element(rsc->ops_xml);
operation != NULL;
operation = __xml_next_element(operation)) {
if (crm_str_eq((const char *)operation->name, "op", TRUE)) {
RecurringOp(rsc, start, node, operation, data_set);
}
}
}
}
static void
RecurringOp_Stopped(resource_t * rsc, action_t * start, node_t * node,
xmlNode * operation, pe_working_set_t * data_set)
{
char *key = NULL;
const char *name = NULL;
const char *role = NULL;
const char *interval_spec = NULL;
const char *node_uname = node? node->details->uname : "n/a";
guint interval_ms = 0;
GListPtr possible_matches = NULL;
GListPtr gIter = NULL;
/* Only process for the operations with role="Stopped" */
role = crm_element_value(operation, "role");
if (role == NULL || text2role(role) != RSC_ROLE_STOPPED) {
return;
}
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
interval_ms = crm_parse_interval_spec(interval_spec);
if (interval_ms == 0) {
return;
}
name = crm_element_value(operation, "name");
if (is_op_dup(rsc, name, interval_ms)) {
crm_trace("Not creating duplicate recurring action %s for %dms %s",
ID(operation), interval_ms, name);
return;
}
if (op_cannot_recur(name)) {
crm_config_err("Invalid recurring action %s wth name: '%s'", ID(operation), name);
return;
}
key = generate_op_key(rsc->id, name, interval_ms);
if (find_rsc_op_entry(rsc, key) == NULL) {
crm_trace("Not creating recurring action %s for disabled resource %s",
ID(operation), rsc->id);
free(key);
return;
}
// @TODO add support
if (is_set(rsc->flags, pe_rsc_unique) == FALSE) {
crm_notice("Ignoring %s (recurring monitors for Stopped role are "
"not supported for anonymous clones)",
ID(operation));
return;
}
pe_rsc_trace(rsc,
"Creating recurring action %s for %s in role %s on nodes where it should not be running",
ID(operation), rsc->id, role2text(rsc->next_role));
/* if the monitor exists on the node where the resource will be running, cancel it */
if (node != NULL) {
possible_matches = find_actions_exact(rsc->actions, key, node);
if (possible_matches) {
action_t *cancel_op = NULL;
g_list_free(possible_matches);
cancel_op = pe_cancel_op(rsc, name, interval_ms, node, data_set);
if (rsc->next_role == RSC_ROLE_STARTED || rsc->next_role == RSC_ROLE_SLAVE) {
/* rsc->role == RSC_ROLE_STOPPED: cancel the monitor before start */
/* rsc->role == RSC_ROLE_STARTED: for a migration, cancel the monitor on the target node before start */
custom_action_order(rsc, NULL, cancel_op, rsc, start_key(rsc), NULL,
pe_order_runnable_left, data_set);
}
pe_rsc_info(rsc, "Cancel action %s (%s vs. %s) on %s",
key, role, role2text(rsc->next_role), node_uname);
}
}
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
node_t *stop_node = (node_t *) gIter->data;
const char *stop_node_uname = stop_node->details->uname;
gboolean is_optional = TRUE;
gboolean probe_is_optional = TRUE;
gboolean stop_is_optional = TRUE;
action_t *stopped_mon = NULL;
char *rc_inactive = NULL;
GListPtr probe_complete_ops = NULL;
GListPtr stop_ops = NULL;
GListPtr local_gIter = NULL;
if (node && safe_str_eq(stop_node_uname, node_uname)) {
continue;
}
pe_rsc_trace(rsc, "Creating recurring action %s for %s on %s",
ID(operation), rsc->id, crm_str(stop_node_uname));
/* start a monitor for an already stopped resource */
possible_matches = find_actions_exact(rsc->actions, key, stop_node);
if (possible_matches == NULL) {
pe_rsc_trace(rsc, "Marking %s mandatory on %s: not active", key,
crm_str(stop_node_uname));
is_optional = FALSE;
} else {
pe_rsc_trace(rsc, "Marking %s optional on %s: already active", key,
crm_str(stop_node_uname));
is_optional = TRUE;
g_list_free(possible_matches);
}
stopped_mon = custom_action(rsc, strdup(key), name, stop_node, is_optional, TRUE, data_set);
rc_inactive = crm_itoa(PCMK_OCF_NOT_RUNNING);
add_hash_param(stopped_mon->meta, XML_ATTR_TE_TARGET_RC, rc_inactive);
free(rc_inactive);
if (is_set(rsc->flags, pe_rsc_managed)) {
GList *probes = pe__resource_actions(rsc, stop_node, RSC_STATUS,
FALSE);
GListPtr pIter = NULL;
for (pIter = probes; pIter != NULL; pIter = pIter->next) {
action_t *probe = (action_t *) pIter->data;
order_actions(probe, stopped_mon, pe_order_runnable_left);
crm_trace("%s then %s on %s", probe->uuid, stopped_mon->uuid, stop_node->details->uname);
}
g_list_free(probes);
}
if (probe_complete_ops) {
g_list_free(probe_complete_ops);
}
stop_ops = pe__resource_actions(rsc, stop_node, RSC_STOP, TRUE);
for (local_gIter = stop_ops; local_gIter != NULL; local_gIter = local_gIter->next) {
action_t *stop = (action_t *) local_gIter->data;
if (is_set(stop->flags, pe_action_optional) == FALSE) {
stop_is_optional = FALSE;
}
if (is_set(stop->flags, pe_action_runnable) == FALSE) {
crm_debug("%s\t %s (cancelled : stop un-runnable)",
crm_str(stop_node_uname), stopped_mon->uuid);
update_action_flags(stopped_mon, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
}
if (is_set(rsc->flags, pe_rsc_managed)) {
custom_action_order(rsc, stop_key(rsc), stop,
NULL, strdup(key), stopped_mon,
pe_order_implies_then | pe_order_runnable_left, data_set);
}
}
if (stop_ops) {
g_list_free(stop_ops);
}
if (is_optional == FALSE && probe_is_optional && stop_is_optional
&& is_set(rsc->flags, pe_rsc_managed) == FALSE) {
pe_rsc_trace(rsc, "Marking %s optional on %s due to unmanaged",
key, crm_str(stop_node_uname));
update_action_flags(stopped_mon, pe_action_optional, __FUNCTION__, __LINE__);
}
if (is_set(stopped_mon->flags, pe_action_optional)) {
pe_rsc_trace(rsc, "%s\t %s (optional)", crm_str(stop_node_uname), stopped_mon->uuid);
}
if (stop_node->details->online == FALSE || stop_node->details->unclean) {
pe_rsc_debug(rsc, "%s\t %s (cancelled : no node available)",
crm_str(stop_node_uname), stopped_mon->uuid);
update_action_flags(stopped_mon, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
}
if (is_set(stopped_mon->flags, pe_action_runnable)
&& is_set(stopped_mon->flags, pe_action_optional) == FALSE) {
crm_notice(" Start recurring %s (%us) for %s on %s", stopped_mon->task,
interval_ms / 1000, rsc->id, crm_str(stop_node_uname));
}
}
free(key);
}
static void
Recurring_Stopped(resource_t * rsc, action_t * start, node_t * node, pe_working_set_t * data_set)
{
if (is_not_set(rsc->flags, pe_rsc_maintenance) &&
(node == NULL || node->details->maintenance == FALSE)) {
xmlNode *operation = NULL;
for (operation = __xml_first_child_element(rsc->ops_xml);
operation != NULL;
operation = __xml_next_element(operation)) {
if (crm_str_eq((const char *)operation->name, "op", TRUE)) {
RecurringOp_Stopped(rsc, start, node, operation, data_set);
}
}
}
}
static void
handle_migration_actions(resource_t * rsc, node_t *current, node_t *chosen, pe_working_set_t * data_set)
{
action_t *migrate_to = NULL;
action_t *migrate_from = NULL;
action_t *start = NULL;
action_t *stop = NULL;
gboolean partial = rsc->partial_migration_target ? TRUE : FALSE;
pe_rsc_trace(rsc, "Processing migration actions %s moving from %s to %s . partial migration = %s",
rsc->id, current->details->id, chosen->details->id, partial ? "TRUE" : "FALSE");
start = start_action(rsc, chosen, TRUE);
stop = stop_action(rsc, current, TRUE);
if (partial == FALSE) {
migrate_to = custom_action(rsc, generate_op_key(rsc->id, RSC_MIGRATE, 0), RSC_MIGRATE, current, TRUE, TRUE, data_set);
}
migrate_from = custom_action(rsc, generate_op_key(rsc->id, RSC_MIGRATED, 0), RSC_MIGRATED, chosen, TRUE, TRUE, data_set);
if ((migrate_to && migrate_from) || (migrate_from && partial)) {
set_bit(start->flags, pe_action_migrate_runnable);
set_bit(stop->flags, pe_action_migrate_runnable);
update_action_flags(start, pe_action_pseudo, __FUNCTION__, __LINE__); /* easier than trying to delete it from the graph */
/* order probes before migrations */
if (partial) {
set_bit(migrate_from->flags, pe_action_migrate_runnable);
migrate_from->needs = start->needs;
custom_action_order(rsc, generate_op_key(rsc->id, RSC_STATUS, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_MIGRATED, 0), NULL, pe_order_optional, data_set);
} else {
set_bit(migrate_from->flags, pe_action_migrate_runnable);
set_bit(migrate_to->flags, pe_action_migrate_runnable);
migrate_to->needs = start->needs;
custom_action_order(rsc, generate_op_key(rsc->id, RSC_STATUS, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_MIGRATE, 0), NULL, pe_order_optional, data_set);
custom_action_order(rsc, generate_op_key(rsc->id, RSC_MIGRATE, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_MIGRATED, 0), NULL, pe_order_optional | pe_order_implies_first_migratable, data_set);
}
custom_action_order(rsc, generate_op_key(rsc->id, RSC_MIGRATED, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_STOP, 0), NULL, pe_order_optional | pe_order_implies_first_migratable, data_set);
custom_action_order(rsc, generate_op_key(rsc->id, RSC_MIGRATED, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional | pe_order_implies_first_migratable | pe_order_pseudo_left, data_set);
}
if (migrate_to) {
add_hash_param(migrate_to->meta, XML_LRM_ATTR_MIGRATE_SOURCE, current->details->uname);
add_hash_param(migrate_to->meta, XML_LRM_ATTR_MIGRATE_TARGET, chosen->details->uname);
/* Pacemaker Remote connections don't require pending to be recorded in
* the CIB. We can reduce CIB writes by not setting PENDING for them.
*/
if (rsc->is_remote_node == FALSE) {
/* migrate_to takes place on the source node, but can
* have an effect on the target node depending on how
* the agent is written. Because of this, we have to maintain
* a record that the migrate_to occurred, in case the source node
* loses membership while the migrate_to action is still in-flight.
*/
add_hash_param(migrate_to->meta, XML_OP_ATTR_PENDING, "true");
}
}
if (migrate_from) {
add_hash_param(migrate_from->meta, XML_LRM_ATTR_MIGRATE_SOURCE, current->details->uname);
add_hash_param(migrate_from->meta, XML_LRM_ATTR_MIGRATE_TARGET, chosen->details->uname);
}
}
void
native_create_actions(resource_t * rsc, pe_working_set_t * data_set)
{
action_t *start = NULL;
node_t *chosen = NULL;
node_t *current = NULL;
gboolean need_stop = FALSE;
gboolean is_moving = FALSE;
gboolean allow_migrate = is_set(rsc->flags, pe_rsc_allow_migrate) ? TRUE : FALSE;
GListPtr gIter = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
bool multiply_active = FALSE;
enum rsc_role_e role = RSC_ROLE_UNKNOWN;
enum rsc_role_e next_role = RSC_ROLE_UNKNOWN;
CRM_ASSERT(rsc);
chosen = rsc->allocated_to;
if (chosen != NULL && rsc->next_role == RSC_ROLE_UNKNOWN) {
rsc->next_role = RSC_ROLE_STARTED;
pe_rsc_trace(rsc, "Fixed next_role: unknown -> %s", role2text(rsc->next_role));
} else if (rsc->next_role == RSC_ROLE_UNKNOWN) {
rsc->next_role = RSC_ROLE_STOPPED;
pe_rsc_trace(rsc, "Fixed next_role: unknown -> %s", role2text(rsc->next_role));
}
pe_rsc_trace(rsc, "Processing state transition for %s %p: %s->%s", rsc->id, rsc,
role2text(rsc->role), role2text(rsc->next_role));
current = pe__find_active_on(rsc, &num_all_active, &num_clean_active);
for (gIter = rsc->dangling_migrations; gIter != NULL; gIter = gIter->next) {
node_t *dangling_source = (node_t *) gIter->data;
action_t *stop = stop_action(rsc, dangling_source, FALSE);
set_bit(stop->flags, pe_action_dangle);
pe_rsc_trace(rsc, "Forcing a cleanup of %s on %s",
rsc->id, dangling_source->details->uname);
if (is_set(data_set->flags, pe_flag_remove_after_stop)) {
DeleteRsc(rsc, dangling_source, FALSE, data_set);
}
}
if ((num_all_active == 2) && (num_clean_active == 2) && chosen
&& rsc->partial_migration_source && rsc->partial_migration_target
&& (current->details == rsc->partial_migration_source->details)
&& (chosen->details == rsc->partial_migration_target->details)) {
/* The chosen node is still the migration target from a partial
* migration. Attempt to continue the migration instead of recovering
* by stopping the resource everywhere and starting it on a single node.
*/
pe_rsc_trace(rsc,
"Will attempt to continue with a partial migration to target %s from %s",
rsc->partial_migration_target->details->id,
rsc->partial_migration_source->details->id);
} else if (is_not_set(rsc->flags, pe_rsc_needs_fencing)) {
/* If a resource has "requires" set to nothing or quorum, don't consider
* it active on unclean nodes (similar to how all resources behave when
* stonith-enabled is false). We can start such resources elsewhere
* before fencing completes, and if we considered the resource active on
* the failed node, we would attempt recovery for being active on
* multiple nodes.
*/
multiply_active = (num_clean_active > 1);
} else {
multiply_active = (num_all_active > 1);
}
if (multiply_active) {
if (rsc->partial_migration_target && rsc->partial_migration_source) {
// Migration was in progress, but we've chosen a different target
crm_notice("Resource %s can no longer migrate to %s. Stopping on %s too",
rsc->id, rsc->partial_migration_target->details->uname,
rsc->partial_migration_source->details->uname);
} else {
// Resource was incorrectly multiply active
pe_proc_err("Resource %s is active on %u nodes (%s)",
rsc->id, num_all_active,
recovery2text(rsc->recovery_type));
crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ#Resource_is_Too_Active for more information");
}
if (rsc->recovery_type == recovery_stop_start) {
need_stop = TRUE;
}
/* If by chance a partial migration is in process, but the migration
* target is not chosen still, clear all partial migration data.
*/
rsc->partial_migration_source = rsc->partial_migration_target = NULL;
allow_migrate = FALSE;
}
if (is_set(rsc->flags, pe_rsc_start_pending)) {
start = start_action(rsc, chosen, TRUE);
set_bit(start->flags, pe_action_print_always);
}
if (current && chosen && current->details != chosen->details) {
pe_rsc_trace(rsc, "Moving %s", rsc->id);
is_moving = TRUE;
need_stop = TRUE;
} else if (is_set(rsc->flags, pe_rsc_failed)) {
pe_rsc_trace(rsc, "Recovering %s", rsc->id);
need_stop = TRUE;
} else if (is_set(rsc->flags, pe_rsc_block)) {
pe_rsc_trace(rsc, "Block %s", rsc->id);
need_stop = TRUE;
} else if (rsc->role > RSC_ROLE_STARTED && current != NULL && chosen != NULL) {
/* Recovery of a promoted resource */
start = start_action(rsc, chosen, TRUE);
if (is_set(start->flags, pe_action_optional) == FALSE) {
pe_rsc_trace(rsc, "Forced start %s", rsc->id);
need_stop = TRUE;
}
}
pe_rsc_trace(rsc, "Creating actions for %s: %s->%s", rsc->id,
role2text(rsc->role), role2text(rsc->next_role));
/* Create any additional actions required when bringing resource down and
* back up to same level.
*/
role = rsc->role;
while (role != RSC_ROLE_STOPPED) {
next_role = rsc_state_matrix[role][RSC_ROLE_STOPPED];
pe_rsc_trace(rsc, "Down: Executing: %s->%s (%s)%s", role2text(role), role2text(next_role),
rsc->id, need_stop ? " required" : "");
if (rsc_action_matrix[role][next_role] (rsc, current, !need_stop, data_set) == FALSE) {
break;
}
role = next_role;
}
while (rsc->role <= rsc->next_role && role != rsc->role && is_not_set(rsc->flags, pe_rsc_block)) {
next_role = rsc_state_matrix[role][rsc->role];
pe_rsc_trace(rsc, "Up: Executing: %s->%s (%s)%s", role2text(role), role2text(next_role),
rsc->id, need_stop ? " required" : "");
if (rsc_action_matrix[role][next_role] (rsc, chosen, !need_stop, data_set) == FALSE) {
break;
}
role = next_role;
}
role = rsc->role;
/* Required steps from this role to the next */
while (role != rsc->next_role) {
next_role = rsc_state_matrix[role][rsc->next_role];
pe_rsc_trace(rsc, "Role: Executing: %s->%s = (%s on %s)", role2text(role), role2text(next_role), rsc->id, chosen?chosen->details->uname:"NA");
if (rsc_action_matrix[role][next_role] (rsc, chosen, FALSE, data_set) == FALSE) {
break;
}
role = next_role;
}
if(is_set(rsc->flags, pe_rsc_block)) {
pe_rsc_trace(rsc, "No monitor additional ops for blocked resource");
} else if (rsc->next_role != RSC_ROLE_STOPPED || is_set(rsc->flags, pe_rsc_managed) == FALSE) {
pe_rsc_trace(rsc, "Monitor ops for active resource");
start = start_action(rsc, chosen, TRUE);
Recurring(rsc, start, chosen, data_set);
Recurring_Stopped(rsc, start, chosen, data_set);
} else {
pe_rsc_trace(rsc, "Monitor ops for inactive resource");
Recurring_Stopped(rsc, NULL, NULL, data_set);
}
/* if we are stuck in a partial migration, where the target
* of the partial migration no longer matches the chosen target.
* A full stop/start is required */
if (rsc->partial_migration_target && (chosen == NULL || rsc->partial_migration_target->details != chosen->details)) {
pe_rsc_trace(rsc, "Not allowing partial migration to continue. %s", rsc->id);
allow_migrate = FALSE;
} else if (is_moving == FALSE ||
is_not_set(rsc->flags, pe_rsc_managed) ||
is_set(rsc->flags, pe_rsc_failed) ||
is_set(rsc->flags, pe_rsc_start_pending) ||
(current && current->details->unclean) ||
rsc->next_role < RSC_ROLE_STARTED) {
allow_migrate = FALSE;
}
if (allow_migrate) {
handle_migration_actions(rsc, current, chosen, data_set);
}
}
static void
rsc_avoids_remote_nodes(resource_t *rsc)
{
GHashTableIter iter;
node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if (node->details->remote_rsc) {
node->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Return allowed nodes as (possibly sorted) list
*
* Convert a resource's hash table of allowed nodes to a list. If printing to
* stdout, sort the list, to keep action ID numbers consistent for regression
* test output (while avoiding the performance hit on a live cluster).
*
* \param[in] rsc Resource to check for allowed nodes
* \param[in] data_set Cluster working set
*
* \return List of resource's allowed nodes
* \note Callers should take care not to rely on the list being sorted.
*/
static GList *
allowed_nodes_as_list(pe_resource_t *rsc, pe_working_set_t *data_set)
{
GList *allowed_nodes = NULL;
if (rsc->allowed_nodes) {
allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
if (is_set(data_set->flags, pe_flag_stdout)) {
allowed_nodes = g_list_sort(allowed_nodes, sort_node_uname);
}
return allowed_nodes;
}
void
native_internal_constraints(resource_t * rsc, pe_working_set_t * data_set)
{
/* This function is on the critical path and worth optimizing as much as possible */
pe_resource_t *top = NULL;
GList *allowed_nodes = NULL;
bool check_unfencing = FALSE;
bool check_utilization = FALSE;
if (is_not_set(rsc->flags, pe_rsc_managed)) {
pe_rsc_trace(rsc,
"Skipping native constraints for unmanaged resource: %s",
rsc->id);
return;
}
top = uber_parent(rsc);
// Whether resource requires unfencing
check_unfencing = is_not_set(rsc->flags, pe_rsc_fence_device)
&& is_set(data_set->flags, pe_flag_enable_unfencing)
&& is_set(rsc->flags, pe_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& safe_str_neq(data_set->placement_strategy, "default");
// Order stops before starts (i.e. restart)
custom_action_order(rsc, generate_op_key(rsc->id, RSC_STOP, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_START, 0), NULL,
pe_order_optional | pe_order_implies_then | pe_order_restart,
data_set);
// Promotable ordering: demote before stop, start before promote
if (is_set(top->flags, pe_rsc_promotable) || (rsc->role > RSC_ROLE_SLAVE)) {
custom_action_order(rsc, generate_op_key(rsc->id, RSC_DEMOTE, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_STOP, 0), NULL,
pe_order_implies_first_master, data_set);
custom_action_order(rsc, generate_op_key(rsc->id, RSC_START, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_PROMOTE, 0), NULL,
pe_order_runnable_left, data_set);
}
+ // Don't clear resource history if probing on same node
+ custom_action_order(rsc, generate_op_key(rsc->id, CRM_OP_LRM_DELETE, 0),
+ NULL, rsc, generate_op_key(rsc->id, RSC_STATUS, 0),
+ NULL, pe_order_same_node|pe_order_then_cancels_first,
+ data_set);
+
// Certain checks need allowed nodes
if (check_unfencing || check_utilization || rsc->container) {
allowed_nodes = allowed_nodes_as_list(rsc, data_set);
}
if (check_unfencing) {
/* Check if the node needs to be unfenced first */
for (GList *item = allowed_nodes; item; item = item->next) {
pe_node_t *node = item->data;
pe_action_t *unfence = pe_fence_op(node, "on", TRUE, NULL, data_set);
crm_debug("Ordering any stops of %s before %s, and any starts after",
rsc->id, unfence->uuid);
/*
* It would be more efficient to order clone resources once,
* rather than order each instance, but ordering the instance
* allows us to avoid unnecessary dependencies that might conflict
* with user constraints.
*
* @TODO: This constraint can still produce a transition loop if the
* resource has a stop scheduled on the node being unfenced, and
* there is a user ordering constraint to start some other resource
* (which will be ordered after the unfence) before stopping this
* resource. An example is "start some slow-starting cloned service
* before stopping an associated virtual IP that may be moving to
* it":
* stop this -> unfencing -> start that -> stop this
*/
custom_action_order(rsc, stop_key(rsc), NULL,
NULL, strdup(unfence->uuid), unfence,
pe_order_optional|pe_order_same_node, data_set);
custom_action_order(NULL, strdup(unfence->uuid), unfence,
rsc, start_key(rsc), NULL,
pe_order_implies_then_on_node|pe_order_same_node,
data_set);
}
}
if (check_utilization) {
GListPtr gIter = NULL;
pe_rsc_trace(rsc, "Creating utilization constraints for %s - strategy: %s",
rsc->id, data_set->placement_strategy);
for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) {
node_t *current = (node_t *) gIter->data;
char *load_stopped_task = crm_concat(LOAD_STOPPED, current->details->uname, '_');
action_t *load_stopped = get_pseudo_op(load_stopped_task, data_set);
if (load_stopped->node == NULL) {
load_stopped->node = node_copy(current);
update_action_flags(load_stopped, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__);
}
custom_action_order(rsc, stop_key(rsc), NULL,
NULL, load_stopped_task, load_stopped, pe_order_load, data_set);
}
for (GList *item = allowed_nodes; item; item = item->next) {
pe_node_t *next = item->data;
char *load_stopped_task = crm_concat(LOAD_STOPPED, next->details->uname, '_');
action_t *load_stopped = get_pseudo_op(load_stopped_task, data_set);
if (load_stopped->node == NULL) {
load_stopped->node = node_copy(next);
update_action_flags(load_stopped, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__);
}
custom_action_order(NULL, strdup(load_stopped_task), load_stopped,
rsc, start_key(rsc), NULL, pe_order_load, data_set);
custom_action_order(NULL, strdup(load_stopped_task), load_stopped,
rsc, generate_op_key(rsc->id, RSC_MIGRATE, 0), NULL,
pe_order_load, data_set);
free(load_stopped_task);
}
}
if (rsc->container) {
resource_t *remote_rsc = NULL;
if (rsc->is_remote_node) {
// rsc is the implicit remote connection for a guest or bundle node
/* Do not allow a guest resource to live on a Pacemaker Remote node,
* to avoid nesting remotes. However, allow bundles to run on remote
* nodes.
*/
if (is_not_set(rsc->flags, pe_rsc_allow_remote_remotes)) {
rsc_avoids_remote_nodes(rsc->container);
}
/* If someone cleans up a guest or bundle node's container, we will
* likely schedule a (re-)probe of the container and recovery of the
* connection. Order the connection stop after the container probe,
* so that if we detect the container running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
new_rsc_order(rsc->container, RSC_STATUS, rsc, RSC_STOP,
pe_order_optional, data_set);
/* A user can specify that a resource must start on a Pacemaker Remote
* node by explicitly configuring it with the container=NODENAME
* meta-attribute. This is of questionable merit, since location
* constraints can accomplish the same thing. But we support it, so here
* we check whether a resource (that is not itself a remote connection)
* has container set to a remote node or guest node resource.
*/
} else if (rsc->container->is_remote_node) {
remote_rsc = rsc->container;
} else {
remote_rsc = pe__resource_contains_guest_node(data_set,
rsc->container);
}
if (remote_rsc) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the container resource.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pe_node_t *node = item->data;
if (node->details->remote_rsc != remote_rsc) {
node->weight = -INFINITY;
}
}
} else {
/* This resource is either a filler for a container that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its container %s",
rsc->id, rsc->container->id);
custom_action_order(rsc->container, generate_op_key(rsc->container->id, RSC_START, 0), NULL,
rsc, generate_op_key(rsc->id, RSC_START, 0), NULL,
pe_order_implies_then | pe_order_runnable_left, data_set);
custom_action_order(rsc, generate_op_key(rsc->id, RSC_STOP, 0), NULL,
rsc->container, generate_op_key(rsc->container->id, RSC_STOP, 0), NULL,
pe_order_implies_first, data_set);
if (is_set(rsc->flags, pe_rsc_allow_remote_remotes)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
rsc_colocation_new("resource-with-container", NULL, score, rsc,
rsc->container, NULL, NULL, data_set);
}
}
if (rsc->is_remote_node || is_set(rsc->flags, pe_rsc_fence_device)) {
/* don't allow remote nodes to run stonith devices
* or remote connection resources.*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
void
native_rsc_colocation_lh(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh,
rsc_colocation_t *constraint,
pe_working_set_t *data_set)
{
if (rsc_lh == NULL) {
pe_err("rsc_lh was NULL for %s", constraint->id);
return;
} else if (constraint->rsc_rh == NULL) {
pe_err("rsc_rh was NULL for %s", constraint->id);
return;
}
pe_rsc_trace(rsc_lh, "Processing colocation constraint between %s and %s", rsc_lh->id,
rsc_rh->id);
rsc_rh->cmds->rsc_colocation_rh(rsc_lh, rsc_rh, constraint, data_set);
}
enum filter_colocation_res
filter_colocation_constraint(resource_t * rsc_lh, resource_t * rsc_rh,
rsc_colocation_t * constraint, gboolean preview)
{
if (constraint->score == 0) {
return influence_nothing;
}
/* rh side must be allocated before we can process constraint */
if (preview == FALSE && is_set(rsc_rh->flags, pe_rsc_provisional)) {
return influence_nothing;
}
if ((constraint->role_lh >= RSC_ROLE_SLAVE) &&
rsc_lh->parent && is_set(rsc_lh->parent->flags, pe_rsc_promotable)
&& is_not_set(rsc_lh->flags, pe_rsc_provisional)) {
/* LH and RH resources have already been allocated, place the correct
* priority on LH rsc for the given promotable clone resource role */
return influence_rsc_priority;
}
if (preview == FALSE && is_not_set(rsc_lh->flags, pe_rsc_provisional)) {
// Log an error if we violated a mandatory colocation constraint
const pe_node_t *rh_node = rsc_rh->allocated_to;
if (rsc_lh->allocated_to == NULL) {
// Dependent resource isn't allocated, so constraint doesn't matter
return influence_nothing;
}
if (constraint->score >= INFINITY) {
// Dependent resource must colocate with rh_node
if ((rh_node == NULL)
|| (rh_node->details != rsc_lh->allocated_to->details)) {
crm_err("%s must be colocated with %s but is not (%s vs. %s)",
rsc_lh->id, rsc_rh->id,
rsc_lh->allocated_to->details->uname,
(rh_node? rh_node->details->uname : "unallocated"));
}
} else if (constraint->score <= -INFINITY) {
// Dependent resource must anti-colocate with rh_node
if ((rh_node != NULL)
&& (rsc_lh->allocated_to->details == rh_node->details)) {
crm_err("%s and %s must be anti-colocated but are allocated "
"to the same node (%s)",
rsc_lh->id, rsc_rh->id, rh_node->details->uname);
}
}
return influence_nothing;
}
if (constraint->score > 0
&& constraint->role_lh != RSC_ROLE_UNKNOWN && constraint->role_lh != rsc_lh->next_role) {
crm_trace("LH: Skipping constraint: \"%s\" state filter nextrole is %s",
role2text(constraint->role_lh), role2text(rsc_lh->next_role));
return influence_nothing;
}
if (constraint->score > 0
&& constraint->role_rh != RSC_ROLE_UNKNOWN && constraint->role_rh != rsc_rh->next_role) {
crm_trace("RH: Skipping constraint: \"%s\" state filter", role2text(constraint->role_rh));
return FALSE;
}
if (constraint->score < 0
&& constraint->role_lh != RSC_ROLE_UNKNOWN && constraint->role_lh == rsc_lh->next_role) {
crm_trace("LH: Skipping negative constraint: \"%s\" state filter",
role2text(constraint->role_lh));
return influence_nothing;
}
if (constraint->score < 0
&& constraint->role_rh != RSC_ROLE_UNKNOWN && constraint->role_rh == rsc_rh->next_role) {
crm_trace("RH: Skipping negative constraint: \"%s\" state filter",
role2text(constraint->role_rh));
return influence_nothing;
}
return influence_rsc_location;
}
static void
influence_priority(resource_t * rsc_lh, resource_t * rsc_rh, rsc_colocation_t * constraint)
{
const char *rh_value = NULL;
const char *lh_value = NULL;
const char *attribute = CRM_ATTR_ID;
int score_multiplier = 1;
if (constraint->node_attribute != NULL) {
attribute = constraint->node_attribute;
}
if (!rsc_rh->allocated_to || !rsc_lh->allocated_to) {
return;
}
lh_value = pe_node_attribute_raw(rsc_lh->allocated_to, attribute);
rh_value = pe_node_attribute_raw(rsc_rh->allocated_to, attribute);
if (!safe_str_eq(lh_value, rh_value)) {
if(constraint->score == INFINITY && constraint->role_lh == RSC_ROLE_MASTER) {
rsc_lh->priority = -INFINITY;
}
return;
}
if (constraint->role_rh && (constraint->role_rh != rsc_rh->next_role)) {
return;
}
if (constraint->role_lh == RSC_ROLE_SLAVE) {
score_multiplier = -1;
}
rsc_lh->priority = merge_weights(score_multiplier * constraint->score, rsc_lh->priority);
}
static void
colocation_match(resource_t * rsc_lh, resource_t * rsc_rh, rsc_colocation_t * constraint)
{
const char *tmp = NULL;
const char *value = NULL;
const char *attribute = CRM_ATTR_ID;
GHashTable *work = NULL;
gboolean do_check = FALSE;
GHashTableIter iter;
node_t *node = NULL;
if (constraint->node_attribute != NULL) {
attribute = constraint->node_attribute;
}
if (rsc_rh->allocated_to) {
value = pe_node_attribute_raw(rsc_rh->allocated_to, attribute);
do_check = TRUE;
} else if (constraint->score < 0) {
/* nothing to do:
* anti-colocation with something that is not running
*/
return;
}
work = node_hash_dup(rsc_lh->allowed_nodes);
g_hash_table_iter_init(&iter, work);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
tmp = pe_node_attribute_raw(node, attribute);
if (do_check && safe_str_eq(tmp, value)) {
if (constraint->score < INFINITY) {
pe_rsc_trace(rsc_lh, "%s: %s.%s += %d", constraint->id, rsc_lh->id,
node->details->uname, constraint->score);
node->weight = merge_weights(constraint->score, node->weight);
}
} else if (do_check == FALSE || constraint->score >= INFINITY) {
pe_rsc_trace(rsc_lh, "%s: %s.%s -= %d (%s)", constraint->id, rsc_lh->id,
node->details->uname, constraint->score,
do_check ? "failed" : "unallocated");
node->weight = merge_weights(-constraint->score, node->weight);
}
}
if (can_run_any(work)
|| constraint->score <= -INFINITY || constraint->score >= INFINITY) {
g_hash_table_destroy(rsc_lh->allowed_nodes);
rsc_lh->allowed_nodes = work;
work = NULL;
} else {
static char score[33];
score2char_stack(constraint->score, score, sizeof(score));
pe_rsc_info(rsc_lh, "%s: Rolling back scores from %s (%d, %s)",
rsc_lh->id, rsc_rh->id, do_check, score);
}
if (work) {
g_hash_table_destroy(work);
}
}
void
native_rsc_colocation_rh(pe_resource_t *rsc_lh, pe_resource_t *rsc_rh,
rsc_colocation_t *constraint,
pe_working_set_t *data_set)
{
enum filter_colocation_res filter_results;
CRM_ASSERT(rsc_lh);
CRM_ASSERT(rsc_rh);
filter_results = filter_colocation_constraint(rsc_lh, rsc_rh, constraint, FALSE);
pe_rsc_trace(rsc_lh, "%sColocating %s with %s (%s, weight=%d, filter=%d)",
constraint->score >= 0 ? "" : "Anti-",
rsc_lh->id, rsc_rh->id, constraint->id, constraint->score, filter_results);
switch (filter_results) {
case influence_rsc_priority:
influence_priority(rsc_lh, rsc_rh, constraint);
break;
case influence_rsc_location:
pe_rsc_trace(rsc_lh, "%sColocating %s with %s (%s, weight=%d)",
constraint->score >= 0 ? "" : "Anti-",
rsc_lh->id, rsc_rh->id, constraint->id, constraint->score);
colocation_match(rsc_lh, rsc_rh, constraint);
break;
case influence_nothing:
default:
return;
}
}
static gboolean
filter_rsc_ticket(resource_t * rsc_lh, rsc_ticket_t * rsc_ticket)
{
if (rsc_ticket->role_lh != RSC_ROLE_UNKNOWN && rsc_ticket->role_lh != rsc_lh->role) {
pe_rsc_trace(rsc_lh, "LH: Skipping constraint: \"%s\" state filter",
role2text(rsc_ticket->role_lh));
return FALSE;
}
return TRUE;
}
void
rsc_ticket_constraint(resource_t * rsc_lh, rsc_ticket_t * rsc_ticket, pe_working_set_t * data_set)
{
if (rsc_ticket == NULL) {
pe_err("rsc_ticket was NULL");
return;
}
if (rsc_lh == NULL) {
pe_err("rsc_lh was NULL for %s", rsc_ticket->id);
return;
}
if (rsc_ticket->ticket->granted && rsc_ticket->ticket->standby == FALSE) {
return;
}
if (rsc_lh->children) {
GListPtr gIter = rsc_lh->children;
pe_rsc_trace(rsc_lh, "Processing ticket dependencies from %s", rsc_lh->id);
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
rsc_ticket_constraint(child_rsc, rsc_ticket, data_set);
}
return;
}
pe_rsc_trace(rsc_lh, "%s: Processing ticket dependency on %s (%s, %s)",
rsc_lh->id, rsc_ticket->ticket->id, rsc_ticket->id,
role2text(rsc_ticket->role_lh));
if ((rsc_ticket->ticket->granted == FALSE)
&& (rsc_lh->running_on != NULL)) {
GListPtr gIter = NULL;
switch (rsc_ticket->loss_policy) {
case loss_ticket_stop:
resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set);
break;
case loss_ticket_demote:
// Promotion score will be set to -INFINITY in promotion_order()
if (rsc_ticket->role_lh != RSC_ROLE_MASTER) {
resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set);
}
break;
case loss_ticket_fence:
if (filter_rsc_ticket(rsc_lh, rsc_ticket) == FALSE) {
return;
}
resource_location(rsc_lh, NULL, -INFINITY, "__loss_of_ticket__", data_set);
for (gIter = rsc_lh->running_on; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
pe_fence_node(data_set, node, "deadman ticket was lost");
}
break;
case loss_ticket_freeze:
if (filter_rsc_ticket(rsc_lh, rsc_ticket) == FALSE) {
return;
}
if (rsc_lh->running_on != NULL) {
clear_bit(rsc_lh->flags, pe_rsc_managed);
set_bit(rsc_lh->flags, pe_rsc_block);
}
break;
}
} else if (rsc_ticket->ticket->granted == FALSE) {
if (rsc_ticket->role_lh != RSC_ROLE_MASTER || rsc_ticket->loss_policy == loss_ticket_stop) {
resource_location(rsc_lh, NULL, -INFINITY, "__no_ticket__", data_set);
}
} else if (rsc_ticket->ticket->standby) {
if (rsc_ticket->role_lh != RSC_ROLE_MASTER || rsc_ticket->loss_policy == loss_ticket_stop) {
resource_location(rsc_lh, NULL, -INFINITY, "__ticket_standby__", data_set);
}
}
}
enum pe_action_flags
native_action_flags(action_t * action, node_t * node)
{
return action->flags;
}
static inline bool
is_primitive_action(pe_action_t *action)
{
return action && action->rsc && (action->rsc->variant == pe_native);
}
/*!
* \internal
* \brief Set action bits appropriately when pe_restart_order is used
*
* \param[in] first 'First' action in an ordering with pe_restart_order
* \param[in] then 'Then' action in an ordering with pe_restart_order
* \param[in] filter What ordering flags to care about
*
* \note pe_restart_order is set for "stop resource before starting it" and
* "stop later group member before stopping earlier group member"
*/
static void
handle_restart_ordering(pe_action_t *first, pe_action_t *then,
enum pe_action_flags filter)
{
const char *reason = NULL;
CRM_ASSERT(is_primitive_action(first));
CRM_ASSERT(is_primitive_action(then));
// We need to update the action in two cases:
// ... if 'then' is required
if (is_set(filter, pe_action_optional)
&& is_not_set(then->flags, pe_action_optional)) {
reason = "restart";
}
/* ... if 'then' is unrunnable start of managed resource (if a resource
* should restart but can't start, we still want to stop)
*/
if (is_set(filter, pe_action_runnable)
&& is_not_set(then->flags, pe_action_runnable)
&& is_set(then->rsc->flags, pe_rsc_managed)
&& safe_str_eq(then->task, RSC_START)) {
reason = "stop";
}
if (reason == NULL) {
return;
}
pe_rsc_trace(first->rsc, "Handling %s -> %s for %s",
first->uuid, then->uuid, reason);
// Make 'first' required if it is runnable
if (is_set(first->flags, pe_action_runnable)) {
pe_action_implies(first, then, pe_action_optional);
}
// Make 'first' required if 'then' is required
if (is_not_set(then->flags, pe_action_optional)) {
pe_action_implies(first, then, pe_action_optional);
}
// Make 'first' unmigratable if 'then' is unmigratable
if (is_not_set(then->flags, pe_action_migrate_runnable)) {
pe_action_implies(first, then, pe_action_migrate_runnable);
}
// Make 'then' unrunnable if 'first' is required but unrunnable
if (is_not_set(first->flags, pe_action_optional)
&& is_not_set(first->flags, pe_action_runnable)) {
pe_action_implies(then, first, pe_action_runnable);
}
}
enum pe_graph_flags
native_update_actions(pe_action_t *first, pe_action_t *then, pe_node_t *node,
enum pe_action_flags flags, enum pe_action_flags filter,
enum pe_ordering type, pe_working_set_t *data_set)
{
/* flags == get_action_flags(first, then_node) called from update_action() */
enum pe_graph_flags changed = pe_graph_none;
enum pe_action_flags then_flags = then->flags;
enum pe_action_flags first_flags = first->flags;
crm_trace( "Testing %s on %s (0x%.6x) with %s 0x%.6x",
first->uuid, first->node ? first->node->details->uname : "[none]",
first->flags, then->uuid, then->flags);
if (type & pe_order_asymmetrical) {
resource_t *then_rsc = then->rsc;
enum rsc_role_e then_rsc_role = then_rsc ? then_rsc->fns->state(then_rsc, TRUE) : 0;
if (!then_rsc) {
/* ignore */
} else if ((then_rsc_role == RSC_ROLE_STOPPED) && safe_str_eq(then->task, RSC_STOP)) {
/* ignore... if 'then' is supposed to be stopped after 'first', but
* then is already stopped, there is nothing to be done when non-symmetrical. */
} else if ((then_rsc_role >= RSC_ROLE_STARTED)
&& safe_str_eq(then->task, RSC_START)
&& is_set(then->flags, pe_action_optional)
&& then->node
&& pcmk__list_of_1(then_rsc->running_on)
&& then->node->details == ((node_t *) then_rsc->running_on->data)->details) {
/* Ignore. If 'then' is supposed to be started after 'first', but
* 'then' is already started, there is nothing to be done when
* asymmetrical -- unless the start is mandatory, which indicates
* the resource is restarting, and the ordering is still needed.
*/
} else if (!(first->flags & pe_action_runnable)) {
/* prevent 'then' action from happening if 'first' is not runnable and
* 'then' has not yet occurred. */
pe_action_implies(then, first, pe_action_optional);
pe_action_implies(then, first, pe_action_runnable);
pe_rsc_trace(then->rsc, "Unset optional and runnable on %s", then->uuid);
} else {
/* ignore... then is allowed to start/stop if it wants to. */
}
}
if (type & pe_order_implies_first) {
if (is_set(filter, pe_action_optional) && is_not_set(flags /* Should be then_flags? */, pe_action_optional)) {
// Needs is_set(first_flags, pe_action_optional) too?
pe_rsc_trace(first->rsc, "Unset optional on %s because of %s", first->uuid, then->uuid);
pe_action_implies(first, then, pe_action_optional);
}
if (is_set(flags, pe_action_migrate_runnable) &&
is_set(then->flags, pe_action_migrate_runnable) == FALSE &&
is_set(then->flags, pe_action_optional) == FALSE) {
pe_rsc_trace(first->rsc, "Unset migrate runnable on %s because of %s",
first->uuid, then->uuid);
pe_action_implies(first, then, pe_action_migrate_runnable);
}
}
if (type & pe_order_implies_first_master) {
if ((filter & pe_action_optional) &&
((then->flags & pe_action_optional) == FALSE) &&
then->rsc && (then->rsc->role == RSC_ROLE_MASTER)) {
pe_action_implies(first, then, pe_action_optional);
if (is_set(first->flags, pe_action_migrate_runnable) &&
is_set(then->flags, pe_action_migrate_runnable) == FALSE) {
pe_rsc_trace(first->rsc, "Unset migrate runnable on %s because of %s", first->uuid, then->uuid);
pe_action_implies(first, then, pe_action_migrate_runnable);
}
pe_rsc_trace(then->rsc, "Unset optional on %s because of %s", first->uuid, then->uuid);
}
}
if ((type & pe_order_implies_first_migratable)
&& is_set(filter, pe_action_optional)) {
if (((then->flags & pe_action_migrate_runnable) == FALSE) ||
((then->flags & pe_action_runnable) == FALSE)) {
pe_rsc_trace(then->rsc, "Unset runnable on %s because %s is neither runnable or migratable", first->uuid, then->uuid);
pe_action_implies(first, then, pe_action_runnable);
}
if ((then->flags & pe_action_optional) == 0) {
pe_rsc_trace(then->rsc, "Unset optional on %s because %s is not optional", first->uuid, then->uuid);
pe_action_implies(first, then, pe_action_optional);
}
}
if ((type & pe_order_pseudo_left)
&& is_set(filter, pe_action_optional)) {
if ((first->flags & pe_action_runnable) == FALSE) {
pe_action_implies(then, first, pe_action_migrate_runnable);
pe_clear_action_bit(then, pe_action_pseudo);
pe_rsc_trace(then->rsc, "Unset pseudo on %s because %s is not runnable", then->uuid, first->uuid);
}
}
if (is_set(type, pe_order_runnable_left)
&& is_set(filter, pe_action_runnable)
&& is_set(then->flags, pe_action_runnable)
&& is_set(flags, pe_action_runnable) == FALSE) {
pe_rsc_trace(then->rsc, "Unset runnable on %s because of %s", then->uuid, first->uuid);
pe_action_implies(then, first, pe_action_runnable);
pe_action_implies(then, first, pe_action_migrate_runnable);
}
if (is_set(type, pe_order_implies_then)
&& is_set(filter, pe_action_optional)
&& is_set(then->flags, pe_action_optional)
&& is_set(flags, pe_action_optional) == FALSE) {
/* in this case, treat migrate_runnable as if first is optional */
if (is_set(first->flags, pe_action_migrate_runnable) == FALSE) {
pe_rsc_trace(then->rsc, "Unset optional on %s because of %s", then->uuid, first->uuid);
pe_action_implies(then, first, pe_action_optional);
}
}
if (is_set(type, pe_order_restart)) {
handle_restart_ordering(first, then, filter);
}
if (then_flags != then->flags) {
changed |= pe_graph_updated_then;
pe_rsc_trace(then->rsc,
"Then: Flags for %s on %s are now 0x%.6x (was 0x%.6x) because of %s 0x%.6x",
then->uuid, then->node ? then->node->details->uname : "[none]", then->flags,
then_flags, first->uuid, first->flags);
if(then->rsc && then->rsc->parent) {
/* "X_stop then X_start" doesn't get handled for cloned groups unless we do this */
update_action(then, data_set);
}
}
if (first_flags != first->flags) {
changed |= pe_graph_updated_first;
pe_rsc_trace(first->rsc,
"First: Flags for %s on %s are now 0x%.6x (was 0x%.6x) because of %s 0x%.6x",
first->uuid, first->node ? first->node->details->uname : "[none]",
first->flags, first_flags, then->uuid, then->flags);
}
return changed;
}
void
native_rsc_location(pe_resource_t *rsc, pe__location_t *constraint)
{
GListPtr gIter = NULL;
GHashTableIter iter;
node_t *node = NULL;
if (constraint == NULL) {
pe_err("Constraint is NULL");
return;
} else if (rsc == NULL) {
pe_err("LHS of rsc_to_node (%s) is NULL", constraint->id);
return;
}
pe_rsc_trace(rsc, "Applying %s (%s) to %s", constraint->id,
role2text(constraint->role_filter), rsc->id);
/* take "lifetime" into account */
if (constraint->role_filter > RSC_ROLE_UNKNOWN && constraint->role_filter != rsc->next_role) {
pe_rsc_debug(rsc, "Constraint (%s) is not active (role : %s vs. %s)",
constraint->id, role2text(constraint->role_filter), role2text(rsc->next_role));
return;
}
if (constraint->node_list_rh == NULL) {
pe_rsc_trace(rsc, "RHS of constraint %s is NULL", constraint->id);
return;
}
for (gIter = constraint->node_list_rh; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
node_t *other_node = NULL;
other_node = (node_t *) pe_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (other_node != NULL) {
pe_rsc_trace(rsc, "%s + %s: %d + %d",
node->details->uname,
other_node->details->uname, node->weight, other_node->weight);
other_node->weight = merge_weights(other_node->weight, node->weight);
} else {
other_node = node_copy(node);
pe_rsc_trace(rsc, "%s: %d (insert %d)", other_node->details->uname, other_node->weight, constraint->discover_mode);
g_hash_table_insert(rsc->allowed_nodes, (gpointer) other_node->details->id, other_node);
}
if (other_node->rsc_discover_mode < constraint->discover_mode) {
if (constraint->discover_mode == pe_discover_exclusive) {
rsc->exclusive_discover = TRUE;
}
/* exclusive > never > always... always is default */
other_node->rsc_discover_mode = constraint->discover_mode;
}
}
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
pe_rsc_trace(rsc, "%s + %s : %d", rsc->id, node->details->uname, node->weight);
}
}
void
native_expand(resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "Processing actions from %s", rsc->id);
for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
crm_trace("processing action %d for rsc=%s", action->id, rsc->id);
graph_element_from_action(action, data_set);
}
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
child_rsc->cmds->expand(child_rsc, data_set);
}
}
#define log_change(a, fmt, args...) do { \
if(a && a->reason && terminal) { \
printf(" * "fmt" \tdue to %s\n", ##args, a->reason); \
} else if(a && a->reason) { \
crm_notice(fmt" \tdue to %s", ##args, a->reason); \
} else if(terminal) { \
printf(" * "fmt"\n", ##args); \
} else { \
crm_notice(fmt, ##args); \
} \
} while(0)
#define STOP_SANITY_ASSERT(lineno) do { \
if(current && current->details->unclean) { \
/* It will be a pseudo op */ \
} else if(stop == NULL) { \
crm_err("%s:%d: No stop action exists for %s", __FUNCTION__, lineno, rsc->id); \
CRM_ASSERT(stop != NULL); \
} else if(is_set(stop->flags, pe_action_optional)) { \
crm_err("%s:%d: Action %s is still optional", __FUNCTION__, lineno, stop->uuid); \
CRM_ASSERT(is_not_set(stop->flags, pe_action_optional)); \
} \
} while(0)
static int rsc_width = 5;
static int detail_width = 5;
static void
LogAction(const char *change, resource_t *rsc, pe_node_t *origin, pe_node_t *destination, pe_action_t *action, pe_action_t *source, gboolean terminal)
{
int len = 0;
char *reason = NULL;
char *details = NULL;
bool same_host = FALSE;
bool same_role = FALSE;
bool need_role = FALSE;
CRM_ASSERT(action);
CRM_ASSERT(destination != NULL || origin != NULL);
if(source == NULL) {
source = action;
}
len = strlen(rsc->id);
if(len > rsc_width) {
rsc_width = len + 2;
}
if(rsc->role > RSC_ROLE_STARTED || rsc->next_role > RSC_ROLE_SLAVE) {
need_role = TRUE;
}
if(origin != NULL && destination != NULL && origin->details == destination->details) {
same_host = TRUE;
}
if(rsc->role == rsc->next_role) {
same_role = TRUE;
}
if(need_role && origin == NULL) {
/* Promoting from Stopped */
details = crm_strdup_printf("%s -> %s %s", role2text(rsc->role), role2text(rsc->next_role), destination->details->uname);
} else if(need_role && destination == NULL) {
/* Demoting a Master or Stopping a Slave */
details = crm_strdup_printf("%s %s", role2text(rsc->role), origin->details->uname);
} else if(origin == NULL || destination == NULL) {
/* Starting or stopping a resource */
details = crm_strdup_printf("%s", origin?origin->details->uname:destination->details->uname);
} else if(need_role && same_role && same_host) {
/* Recovering or restarting a promotable clone resource */
details = crm_strdup_printf("%s %s", role2text(rsc->role), origin->details->uname);
} else if(same_role && same_host) {
/* Recovering or Restarting a normal resource */
details = crm_strdup_printf("%s", origin->details->uname);
} else if(same_role && need_role) {
/* Moving a promotable clone resource */
details = crm_strdup_printf("%s -> %s %s", origin->details->uname, destination->details->uname, role2text(rsc->role));
} else if(same_role) {
/* Moving a normal resource */
details = crm_strdup_printf("%s -> %s", origin->details->uname, destination->details->uname);
} else if(same_host) {
/* Promoting or demoting a promotable clone resource */
details = crm_strdup_printf("%s -> %s %s", role2text(rsc->role), role2text(rsc->next_role), origin->details->uname);
} else {
/* Moving and promoting/demoting */
details = crm_strdup_printf("%s %s -> %s %s", role2text(rsc->role), origin->details->uname, role2text(rsc->next_role), destination->details->uname);
}
len = strlen(details);
if(len > detail_width) {
detail_width = len;
}
if(source->reason && is_not_set(action->flags, pe_action_runnable)) {
reason = crm_strdup_printf(" due to %s (blocked)", source->reason);
} else if(source->reason) {
reason = crm_strdup_printf(" due to %s", source->reason);
} else if(is_not_set(action->flags, pe_action_runnable)) {
reason = strdup(" blocked");
} else {
reason = strdup("");
}
if(terminal) {
printf(" * %-8s %-*s ( %*s ) %s\n", change, rsc_width, rsc->id, detail_width, details, reason);
} else {
crm_notice(" * %-8s %-*s ( %*s ) %s", change, rsc_width, rsc->id, detail_width, details, reason);
}
free(details);
free(reason);
}
void
LogActions(resource_t * rsc, pe_working_set_t * data_set, gboolean terminal)
{
node_t *next = NULL;
node_t *current = NULL;
pe_node_t *start_node = NULL;
action_t *stop = NULL;
action_t *start = NULL;
action_t *demote = NULL;
action_t *promote = NULL;
char *key = NULL;
gboolean moving = FALSE;
GListPtr possible_matches = NULL;
if(rsc->variant == pe_container) {
pcmk__bundle_log_actions(rsc, data_set, terminal);
return;
}
if (rsc->children) {
GListPtr gIter = NULL;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
LogActions(child_rsc, data_set, terminal);
}
return;
}
next = rsc->allocated_to;
if (rsc->running_on) {
current = pe__current_node(rsc);
if (rsc->role == RSC_ROLE_STOPPED) {
/*
* This can occur when resources are being recovered
* We fiddle with the current role in native_create_actions()
*/
rsc->role = RSC_ROLE_STARTED;
}
}
if (current == NULL && is_set(rsc->flags, pe_rsc_orphan)) {
/* Don't log stopped orphans */
return;
}
if (is_not_set(rsc->flags, pe_rsc_managed)
|| (current == NULL && next == NULL)) {
pe_rsc_info(rsc, "Leave %s\t(%s%s)",
rsc->id, role2text(rsc->role), is_not_set(rsc->flags,
pe_rsc_managed) ? " unmanaged" : "");
return;
}
if (current != NULL && next != NULL && safe_str_neq(current->details->id, next->details->id)) {
moving = TRUE;
}
possible_matches = pe__resource_actions(rsc, next, RSC_START, FALSE);
if (possible_matches) {
start = possible_matches->data;
g_list_free(possible_matches);
}
if ((start == NULL) || is_not_set(start->flags, pe_action_runnable)) {
start_node = NULL;
} else {
start_node = current;
}
possible_matches = pe__resource_actions(rsc, start_node, RSC_STOP, FALSE);
if (possible_matches) {
stop = possible_matches->data;
g_list_free(possible_matches);
}
possible_matches = pe__resource_actions(rsc, next, RSC_PROMOTE, FALSE);
if (possible_matches) {
promote = possible_matches->data;
g_list_free(possible_matches);
}
possible_matches = pe__resource_actions(rsc, next, RSC_DEMOTE, FALSE);
if (possible_matches) {
demote = possible_matches->data;
g_list_free(possible_matches);
}
if (rsc->role == rsc->next_role) {
action_t *migrate_op = NULL;
possible_matches = pe__resource_actions(rsc, next, RSC_MIGRATED, FALSE);
if (possible_matches) {
migrate_op = possible_matches->data;
}
CRM_CHECK(next != NULL,);
if (next == NULL) {
} else if (migrate_op && is_set(migrate_op->flags, pe_action_runnable) && current) {
LogAction("Migrate", rsc, current, next, start, NULL, terminal);
} else if (is_set(rsc->flags, pe_rsc_reload)) {
LogAction("Reload", rsc, current, next, start, NULL, terminal);
} else if (start == NULL || is_set(start->flags, pe_action_optional)) {
pe_rsc_info(rsc, "Leave %s\t(%s %s)", rsc->id, role2text(rsc->role),
next->details->uname);
} else if (start && is_set(start->flags, pe_action_runnable) == FALSE) {
LogAction("Stop", rsc, current, NULL, stop,
(stop && stop->reason)? stop : start, terminal);
STOP_SANITY_ASSERT(__LINE__);
} else if (moving && current) {
LogAction(is_set(rsc->flags, pe_rsc_failed) ? "Recover" : "Move",
rsc, current, next, stop, NULL, terminal);
} else if (is_set(rsc->flags, pe_rsc_failed)) {
LogAction("Recover", rsc, current, NULL, stop, NULL, terminal);
STOP_SANITY_ASSERT(__LINE__);
} else {
LogAction("Restart", rsc, current, next, start, NULL, terminal);
/* STOP_SANITY_ASSERT(__LINE__); False positive for migrate-fail-7 */
}
g_list_free(possible_matches);
return;
}
if(stop
&& (rsc->next_role == RSC_ROLE_STOPPED
|| (start && is_not_set(start->flags, pe_action_runnable)))) {
GListPtr gIter = NULL;
key = stop_key(rsc);
for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
action_t *stop_op = NULL;
possible_matches = find_actions(rsc->actions, key, node);
if (possible_matches) {
stop_op = possible_matches->data;
g_list_free(possible_matches);
}
if (stop_op && (stop_op->flags & pe_action_runnable)) {
STOP_SANITY_ASSERT(__LINE__);
}
LogAction("Stop", rsc, node, NULL, stop_op,
(stop_op && stop_op->reason)? stop_op : start, terminal);
}
free(key);
} else if (stop && is_set(rsc->flags, pe_rsc_failed)) {
/* 'stop' may be NULL if the failure was ignored */
LogAction("Recover", rsc, current, next, stop, start, terminal);
STOP_SANITY_ASSERT(__LINE__);
} else if (moving) {
LogAction("Move", rsc, current, next, stop, NULL, terminal);
STOP_SANITY_ASSERT(__LINE__);
} else if (is_set(rsc->flags, pe_rsc_reload)) {
LogAction("Reload", rsc, current, next, start, NULL, terminal);
} else if (stop != NULL && is_not_set(stop->flags, pe_action_optional)) {
LogAction("Restart", rsc, current, next, start, NULL, terminal);
STOP_SANITY_ASSERT(__LINE__);
} else if (rsc->role == RSC_ROLE_MASTER) {
CRM_LOG_ASSERT(current != NULL);
LogAction("Demote", rsc, current, next, demote, NULL, terminal);
} else if(rsc->next_role == RSC_ROLE_MASTER) {
CRM_LOG_ASSERT(next);
LogAction("Promote", rsc, current, next, promote, NULL, terminal);
} else if (rsc->role == RSC_ROLE_STOPPED && rsc->next_role > RSC_ROLE_STOPPED) {
LogAction("Start", rsc, current, next, start, NULL, terminal);
}
}
gboolean
StopRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "%s", rsc->id);
for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) {
node_t *current = (node_t *) gIter->data;
action_t *stop;
if (rsc->partial_migration_target) {
if (rsc->partial_migration_target->details == current->details) {
pe_rsc_trace(rsc, "Filtered %s -> %s %s", current->details->uname,
next->details->uname, rsc->id);
continue;
} else {
pe_rsc_trace(rsc, "Forced on %s %s", current->details->uname, rsc->id);
optional = FALSE;
}
}
pe_rsc_trace(rsc, "%s on %s", rsc->id, current->details->uname);
stop = stop_action(rsc, current, optional);
if(rsc->allocated_to == NULL) {
pe_action_set_reason(stop, "node availability", TRUE);
}
if (is_not_set(rsc->flags, pe_rsc_managed)) {
update_action_flags(stop, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
}
if (is_set(data_set->flags, pe_flag_remove_after_stop)) {
DeleteRsc(rsc, current, optional, data_set);
}
if(is_set(rsc->flags, pe_rsc_needs_unfencing)) {
action_t *unfence = pe_fence_op(current, "on", TRUE, NULL, data_set);
order_actions(stop, unfence, pe_order_implies_first);
if (!node_has_been_unfenced(current)) {
pe_proc_err("Stopping %s until %s can be unfenced", rsc->id, current->details->uname);
}
}
}
return TRUE;
}
static void
order_after_unfencing(resource_t *rsc, pe_node_t *node, action_t *action,
enum pe_ordering order, pe_working_set_t *data_set)
{
/* When unfencing is in use, we order unfence actions before any probe or
* start of resources that require unfencing, and also of fence devices.
*
* This might seem to violate the principle that fence devices require
* only quorum. However, fence agents that unfence often don't have enough
* information to even probe or start unless the node is first unfenced.
*/
if (is_unfence_device(rsc, data_set)
|| is_set(rsc->flags, pe_rsc_needs_unfencing)) {
/* Start with an optional ordering. Requiring unfencing would result in
* the node being unfenced, and all its resources being stopped,
* whenever a new resource is added -- which would be highly suboptimal.
*/
action_t *unfence = pe_fence_op(node, "on", TRUE, NULL, data_set);
order_actions(unfence, action, order);
if (!node_has_been_unfenced(node)) {
// But unfencing is required if it has never been done
char *reason = crm_strdup_printf("required by %s %s",
rsc->id, action->task);
trigger_unfencing(NULL, node, reason, NULL, data_set);
free(reason);
}
}
}
gboolean
StartRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
action_t *start = NULL;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "%s on %s %d %d", rsc->id, next ? next->details->uname : "N/A", optional, next ? next->weight : 0);
start = start_action(rsc, next, TRUE);
order_after_unfencing(rsc, next, start, pe_order_implies_then, data_set);
if (is_set(start->flags, pe_action_runnable) && optional == FALSE) {
update_action_flags(start, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__);
}
return TRUE;
}
gboolean
PromoteRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
gboolean runnable = TRUE;
GListPtr action_list = NULL;
CRM_ASSERT(rsc);
CRM_CHECK(next != NULL, return FALSE);
pe_rsc_trace(rsc, "%s on %s", rsc->id, next->details->uname);
action_list = pe__resource_actions(rsc, next, RSC_START, TRUE);
for (gIter = action_list; gIter != NULL; gIter = gIter->next) {
action_t *start = (action_t *) gIter->data;
if (is_set(start->flags, pe_action_runnable) == FALSE) {
runnable = FALSE;
}
}
g_list_free(action_list);
if (runnable) {
promote_action(rsc, next, optional);
return TRUE;
}
pe_rsc_debug(rsc, "%s\tPromote %s (canceled)", next->details->uname, rsc->id);
action_list = pe__resource_actions(rsc, next, RSC_PROMOTE, TRUE);
for (gIter = action_list; gIter != NULL; gIter = gIter->next) {
action_t *promote = (action_t *) gIter->data;
update_action_flags(promote, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__);
}
g_list_free(action_list);
return TRUE;
}
gboolean
DemoteRsc(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "%s", rsc->id);
/* CRM_CHECK(rsc->next_role == RSC_ROLE_SLAVE, return FALSE); */
for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) {
node_t *current = (node_t *) gIter->data;
pe_rsc_trace(rsc, "%s on %s", rsc->id, next ? next->details->uname : "N/A");
demote_action(rsc, current, optional);
}
return TRUE;
}
gboolean
RoleError(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
CRM_ASSERT(rsc);
crm_err("%s on %s", rsc->id, next ? next->details->uname : "N/A");
CRM_CHECK(FALSE, return FALSE);
return FALSE;
}
gboolean
NullOp(resource_t * rsc, node_t * next, gboolean optional, pe_working_set_t * data_set)
{
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "%s", rsc->id);
return FALSE;
}
gboolean
DeleteRsc(resource_t * rsc, node_t * node, gboolean optional, pe_working_set_t * data_set)
{
if (is_set(rsc->flags, pe_rsc_failed)) {
pe_rsc_trace(rsc, "Resource %s not deleted from %s: failed", rsc->id, node->details->uname);
return FALSE;
} else if (node == NULL) {
pe_rsc_trace(rsc, "Resource %s not deleted: NULL node", rsc->id);
return FALSE;
} else if (node->details->unclean || node->details->online == FALSE) {
pe_rsc_trace(rsc, "Resource %s not deleted from %s: unrunnable", rsc->id,
node->details->uname);
return FALSE;
}
crm_notice("Removing %s from %s", rsc->id, node->details->uname);
delete_action(rsc, node, optional);
new_rsc_order(rsc, RSC_STOP, rsc, RSC_DELETE,
optional ? pe_order_implies_then : pe_order_optional, data_set);
new_rsc_order(rsc, RSC_DELETE, rsc, RSC_START,
optional ? pe_order_implies_then : pe_order_optional, data_set);
return TRUE;
}
gboolean
native_create_probe(resource_t * rsc, node_t * node, action_t * complete,
gboolean force, pe_working_set_t * data_set)
{
enum pe_ordering flags = pe_order_optional;
char *key = NULL;
action_t *probe = NULL;
node_t *running = NULL;
node_t *allowed = NULL;
resource_t *top = uber_parent(rsc);
static const char *rc_master = NULL;
static const char *rc_inactive = NULL;
if (rc_inactive == NULL) {
rc_inactive = crm_itoa(PCMK_OCF_NOT_RUNNING);
rc_master = crm_itoa(PCMK_OCF_RUNNING_MASTER);
}
CRM_CHECK(node != NULL, return FALSE);
if (force == FALSE && is_not_set(data_set->flags, pe_flag_startup_probes)) {
pe_rsc_trace(rsc, "Skipping active resource detection for %s", rsc->id);
return FALSE;
}
if (pe__is_guest_or_remote_node(node)) {
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
if (safe_str_eq(class, PCMK_RESOURCE_CLASS_STONITH)) {
pe_rsc_trace(rsc,
"Skipping probe for %s on %s because Pacemaker Remote nodes cannot run stonith agents",
rsc->id, node->details->id);
return FALSE;
} else if (pe__is_guest_node(node)
&& pe__resource_contains_guest_node(data_set, rsc)) {
pe_rsc_trace(rsc,
"Skipping probe for %s on %s because guest nodes cannot run resources containing guest nodes",
rsc->id, node->details->id);
return FALSE;
} else if (rsc->is_remote_node) {
pe_rsc_trace(rsc,
"Skipping probe for %s on %s because Pacemaker Remote nodes cannot host remote connections",
rsc->id, node->details->id);
return FALSE;
}
}
if (rsc->children) {
GListPtr gIter = NULL;
gboolean any_created = FALSE;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
any_created = child_rsc->cmds->create_probe(child_rsc, node, complete, force, data_set)
|| any_created;
}
return any_created;
} else if ((rsc->container) && (!rsc->is_remote_node)) {
pe_rsc_trace(rsc, "Skipping %s: it is within container %s", rsc->id, rsc->container->id);
return FALSE;
}
if (is_set(rsc->flags, pe_rsc_orphan)) {
pe_rsc_trace(rsc, "Skipping orphan: %s", rsc->id);
return FALSE;
}
// Check whether resource is already known on node
if (!force && g_hash_table_lookup(rsc->known_on, node->details->id)) {
pe_rsc_trace(rsc, "Skipping known: %s on %s", rsc->id, node->details->uname);
return FALSE;
}
allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (rsc->exclusive_discover || top->exclusive_discover) {
if (allowed == NULL) {
/* exclusive discover is enabled and this node is not in the allowed list. */
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, A", rsc->id, node->details->id);
return FALSE;
} else if (allowed->rsc_discover_mode != pe_discover_exclusive) {
/* exclusive discover is enabled and this node is not marked
* as a node this resource should be discovered on */
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, B", rsc->id, node->details->id);
return FALSE;
}
}
if(allowed == NULL && node->rsc_discover_mode == pe_discover_never) {
/* If this node was allowed to host this resource it would
* have been explicitly added to the 'allowed_nodes' list.
* However it wasn't and the node has discovery disabled, so
* no need to probe for this resource.
*/
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, C", rsc->id, node->details->id);
return FALSE;
}
if (allowed && allowed->rsc_discover_mode == pe_discover_never) {
/* this resource is marked as not needing to be discovered on this node */
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, discovery mode", rsc->id, node->details->id);
return FALSE;
}
if (pe__is_guest_node(node)) {
resource_t *remote = node->details->remote_rsc->container;
if(remote->role == RSC_ROLE_STOPPED) {
/* If the container is stopped, then we know anything that
* might have been inside it is also stopped and there is
* no need to probe.
*
* If we don't know the container's state on the target
* either:
*
* - the container is running, the transition will abort
* and we'll end up in a different case next time, or
*
* - the container is stopped
*
* Either way there is no need to probe.
*
*/
if(remote->allocated_to
&& g_hash_table_lookup(remote->known_on, remote->allocated_to->details->id) == NULL) {
/* For safety, we order the 'rsc' start after 'remote'
* has been probed.
*
* Using 'top' helps for groups, but we may need to
* follow the start's ordering chain backwards.
*/
custom_action_order(remote, generate_op_key(remote->id, RSC_STATUS, 0), NULL,
top, generate_op_key(top->id, RSC_START, 0), NULL,
pe_order_optional, data_set);
}
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, %s is stopped",
rsc->id, node->details->id, remote->id);
return FALSE;
/* Here we really we want to check if remote->stop is required,
* but that information doesn't exist yet
*/
} else if(node->details->remote_requires_reset
|| node->details->unclean
|| is_set(remote->flags, pe_rsc_failed)
|| remote->next_role == RSC_ROLE_STOPPED
|| (remote->allocated_to
&& pe_find_node(remote->running_on, remote->allocated_to->details->uname) == NULL)
) {
/* The container is stopping or restarting, don't start
* 'rsc' until 'remote' stops as this also implies that
* 'rsc' is stopped - avoiding the need to probe
*/
custom_action_order(remote, generate_op_key(remote->id, RSC_STOP, 0), NULL,
top, generate_op_key(top->id, RSC_START, 0), NULL,
pe_order_optional, data_set);
pe_rsc_trace(rsc, "Skipping probe for %s on node %s, %s is stopping, restarting or moving",
rsc->id, node->details->id, remote->id);
return FALSE;
/* } else {
* The container is running so there is no problem probing it
*/
}
}
key = generate_op_key(rsc->id, RSC_STATUS, 0);
probe = custom_action(rsc, key, RSC_STATUS, node, FALSE, TRUE, data_set);
update_action_flags(probe, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__);
order_after_unfencing(rsc, node, probe, pe_order_optional, data_set);
/*
* We need to know if it's running_on (not just known_on) this node
* to correctly determine the target rc.
*/
running = pe_find_node_id(rsc->running_on, node->details->id);
if (running == NULL) {
add_hash_param(probe->meta, XML_ATTR_TE_TARGET_RC, rc_inactive);
} else if (rsc->role == RSC_ROLE_MASTER) {
add_hash_param(probe->meta, XML_ATTR_TE_TARGET_RC, rc_master);
}
crm_debug("Probing %s on %s (%s) %d %p", rsc->id, node->details->uname, role2text(rsc->role),
is_set(probe->flags, pe_action_runnable), rsc->running_on);
if (is_unfence_device(rsc, data_set) || !pe_rsc_is_clone(top)) {
top = rsc;
} else {
crm_trace("Probing %s on %s (%s) as %s", rsc->id, node->details->uname, role2text(rsc->role), top->id);
}
if(is_not_set(probe->flags, pe_action_runnable) && rsc->running_on == NULL) {
/* Prevent the start from occurring if rsc isn't active, but
* don't cause it to stop if it was active already
*/
flags |= pe_order_runnable_left;
}
custom_action_order(rsc, NULL, probe,
top, generate_op_key(top->id, RSC_START, 0), NULL,
flags, data_set);
/* Before any reloads, if they exist */
custom_action_order(rsc, NULL, probe,
top, reload_key(rsc), NULL,
pe_order_optional, data_set);
#if 0
// complete is always null currently
if (!is_unfence_device(rsc, data_set)) {
/* Normally rsc.start depends on probe complete which depends
* on rsc.probe. But this can't be the case for fence devices
* with unfencing, as it would create graph loops.
*
* So instead we explicitly order 'rsc.probe then rsc.start'
*/
order_actions(probe, complete, pe_order_implies_then);
}
#endif
return TRUE;
}
/*!
* \internal
* \brief Check whether a resource is known on a particular node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return TRUE if resource (or parent if an anonymous clone) is known
*/
static bool
rsc_is_known_on(pe_resource_t *rsc, const pe_node_t *node)
{
if (pe_hash_table_lookup(rsc->known_on, node->details->id)) {
return TRUE;
} else if ((rsc->variant == pe_native)
&& pe_rsc_is_anon_clone(rsc->parent)
&& pe_hash_table_lookup(rsc->parent->known_on, node->details->id)) {
/* We check only the parent, not the uber-parent, because we cannot
* assume that the resource is known if it is in an anonymously cloned
* group (which may be only partially known).
*/
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Order a resource's start and promote actions relative to fencing
*
* \param[in] rsc Resource to be ordered
* \param[in] stonith_op Fence action
* \param[in] data_set Cluster information
*/
static void
native_start_constraints(resource_t * rsc, action_t * stonith_op, pe_working_set_t * data_set)
{
node_t *target;
GListPtr gIter = NULL;
CRM_CHECK(stonith_op && stonith_op->node, return);
target = stonith_op->node;
for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
switch (action->needs) {
case rsc_req_nothing:
// Anything other than start or promote requires nothing
break;
case rsc_req_stonith:
order_actions(stonith_op, action, pe_order_optional);
break;
case rsc_req_quorum:
if (safe_str_eq(action->task, RSC_START)
&& pe_hash_table_lookup(rsc->allowed_nodes, target->details->id)
&& !rsc_is_known_on(rsc, target)) {
/* If we don't know the status of the resource on the node
* we're about to shoot, we have to assume it may be active
* there. Order the resource start after the fencing. This
* is analogous to waiting for all the probes for a resource
* to complete before starting it.
*
* The most likely explanation is that the DC died and took
* its status with it.
*/
pe_rsc_debug(rsc, "Ordering %s after %s recovery", action->uuid,
target->details->uname);
order_actions(stonith_op, action,
pe_order_optional | pe_order_runnable_left);
}
break;
}
}
}
static void
native_stop_constraints(resource_t * rsc, action_t * stonith_op, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
GListPtr action_list = NULL;
bool order_implicit = false;
resource_t *top = uber_parent(rsc);
pe_action_t *parent_stop = NULL;
node_t *target;
CRM_CHECK(stonith_op && stonith_op->node, return);
target = stonith_op->node;
/* Get a list of stop actions potentially implied by the fencing */
action_list = pe__resource_actions(rsc, target, RSC_STOP, FALSE);
/* If resource requires fencing, implicit actions must occur after fencing.
*
* Implied stops and demotes of resources running on guest nodes are always
* ordered after fencing, even if the resource does not require fencing,
* because guest node "fencing" is actually just a resource stop.
*/
if (is_set(rsc->flags, pe_rsc_needs_fencing) || pe__is_guest_node(target)) {
order_implicit = true;
}
if (action_list && order_implicit) {
parent_stop = find_first_action(top->actions, NULL, RSC_STOP, NULL);
}
for (gIter = action_list; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
// The stop would never complete, so convert it into a pseudo-action.
update_action_flags(action, pe_action_pseudo|pe_action_runnable,
__FUNCTION__, __LINE__);
if (order_implicit) {
update_action_flags(action, pe_action_implied_by_stonith,
__FUNCTION__, __LINE__);
/* Order the stonith before the parent stop (if any).
*
* Also order the stonith before the resource stop, unless the
* resource is inside a bundle -- that would cause a graph loop.
* We can rely on the parent stop's ordering instead.
*
* User constraints must not order a resource in a guest node
* relative to the guest node container resource. The
* pe_order_preserve flag marks constraints as generated by the
* cluster and thus immune to that check (and is irrelevant if
* target is not a guest).
*/
if (!pe_rsc_is_bundled(rsc)) {
order_actions(stonith_op, action, pe_order_preserve);
}
order_actions(stonith_op, parent_stop, pe_order_preserve);
}
if (is_set(rsc->flags, pe_rsc_failed)) {
crm_notice("Stop of failed resource %s is implicit %s %s is fenced",
rsc->id, (order_implicit? "after" : "because"),
target->details->uname);
} else {
crm_info("%s is implicit %s %s is fenced",
action->uuid, (order_implicit? "after" : "because"),
target->details->uname);
}
if (is_set(rsc->flags, pe_rsc_notify)) {
/* Create a second notification that will be delivered
* immediately after the node is fenced
*
* Basic problem:
* - C is a clone active on the node to be shot and stopping on another
* - R is a resource that depends on C
*
* + C.stop depends on R.stop
* + C.stopped depends on STONITH
* + C.notify depends on C.stopped
* + C.healthy depends on C.notify
* + R.stop depends on C.healthy
*
* The extra notification here changes
* + C.healthy depends on C.notify
* into:
* + C.healthy depends on C.notify'
* + C.notify' depends on STONITH'
* thus breaking the loop
*/
create_secondary_notification(action, rsc, stonith_op, data_set);
}
/* From Bug #1601, successful fencing must be an input to a failed resources stop action.
However given group(rA, rB) running on nodeX and B.stop has failed,
A := stop healthy resource (rA.stop)
B := stop failed resource (pseudo operation B.stop)
C := stonith nodeX
A requires B, B requires C, C requires A
This loop would prevent the cluster from making progress.
This block creates the "C requires A" dependency and therefore must (at least
for now) be disabled.
Instead, run the block above and treat all resources on nodeX as B would be
(marked as a pseudo op depending on the STONITH).
TODO: Break the "A requires B" dependency in update_action() and re-enable this block
} else if(is_stonith == FALSE) {
crm_info("Moving healthy resource %s"
" off %s before fencing",
rsc->id, node->details->uname);
* stop healthy resources before the
* stonith op
*
custom_action_order(
rsc, stop_key(rsc), NULL,
NULL,strdup(CRM_OP_FENCE),stonith_op,
pe_order_optional, data_set);
*/
}
g_list_free(action_list);
/* Get a list of demote actions potentially implied by the fencing */
action_list = pe__resource_actions(rsc, target, RSC_DEMOTE, FALSE);
for (gIter = action_list; gIter != NULL; gIter = gIter->next) {
action_t *action = (action_t *) gIter->data;
if (action->node->details->online == FALSE || action->node->details->unclean == TRUE
|| is_set(rsc->flags, pe_rsc_failed)) {
if (is_set(rsc->flags, pe_rsc_failed)) {
pe_rsc_info(rsc,
"Demote of failed resource %s is implicit after %s is fenced",
rsc->id, target->details->uname);
} else {
pe_rsc_info(rsc, "%s is implicit after %s is fenced",
action->uuid, target->details->uname);
}
/* The demote would never complete and is now implied by the
* fencing, so convert it into a pseudo-action.
*/
update_action_flags(action, pe_action_pseudo|pe_action_runnable,
__FUNCTION__, __LINE__);
if (pe_rsc_is_bundled(rsc)) {
/* Do nothing, let the recovery be ordered after the parent's implied stop */
} else if (order_implicit) {
order_actions(stonith_op, action, pe_order_preserve|pe_order_optional);
}
}
}
g_list_free(action_list);
}
void
rsc_stonith_ordering(resource_t * rsc, action_t * stonith_op, pe_working_set_t * data_set)
{
if (rsc->children) {
GListPtr gIter = NULL;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
rsc_stonith_ordering(child_rsc, stonith_op, data_set);
}
} else if (is_not_set(rsc->flags, pe_rsc_managed)) {
pe_rsc_trace(rsc, "Skipping fencing constraints for unmanaged resource: %s", rsc->id);
} else {
native_start_constraints(rsc, stonith_op, data_set);
native_stop_constraints(rsc, stonith_op, data_set);
}
}
void
ReloadRsc(resource_t * rsc, node_t *node, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
action_t *reload = NULL;
if (rsc->children) {
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
ReloadRsc(child_rsc, node, data_set);
}
return;
} else if (rsc->variant > pe_native) {
/* Complex resource with no children */
return;
} else if (is_not_set(rsc->flags, pe_rsc_managed)) {
pe_rsc_trace(rsc, "%s: unmanaged", rsc->id);
return;
} else if (is_set(rsc->flags, pe_rsc_failed) || is_set(rsc->flags, pe_rsc_start_pending)) {
pe_rsc_trace(rsc, "%s: general resource state: flags=0x%.16llx", rsc->id, rsc->flags);
stop_action(rsc, node, FALSE); /* Force a full restart, overkill? */
return;
} else if (node == NULL) {
pe_rsc_trace(rsc, "%s: not active", rsc->id);
return;
}
pe_rsc_trace(rsc, "Processing %s", rsc->id);
set_bit(rsc->flags, pe_rsc_reload);
reload = custom_action(
rsc, reload_key(rsc), CRMD_ACTION_RELOAD, node, FALSE, TRUE, data_set);
pe_action_set_reason(reload, "resource definition change", FALSE);
custom_action_order(NULL, NULL, reload, rsc, stop_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
data_set);
custom_action_order(NULL, NULL, reload, rsc, demote_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
data_set);
}
void
native_append_meta(resource_t * rsc, xmlNode * xml)
{
char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION);
resource_t *parent;
if (value) {
char *name = NULL;
name = crm_meta_name(XML_RSC_ATTR_INCARNATION);
crm_xml_add(xml, name, value);
free(name);
}
value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE);
if (value) {
char *name = NULL;
name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE);
crm_xml_add(xml, name, value);
free(name);
}
for (parent = rsc; parent != NULL; parent = parent->parent) {
if (parent->container) {
crm_xml_add(xml, CRM_META"_"XML_RSC_ATTR_CONTAINER, parent->container->id);
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_transition.c b/lib/pacemaker/pcmk_sched_transition.c
index c415b75940..1698c85dd0 100644
--- a/lib/pacemaker/pcmk_sched_transition.c
+++ b/lib/pacemaker/pcmk_sched_transition.c
@@ -1,867 +1,862 @@
/*
* Copyright 2009-2019 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include // lrmd_event_data_t, lrmd_free_event()
#include
#include
#include
#include
#include
static bool fake_quiet = FALSE;
static cib_t *fake_cib = NULL;
static GListPtr fake_resource_list = NULL;
static GListPtr fake_op_fail_list = NULL;
gboolean bringing_nodes_online = FALSE;
#define STATUS_PATH_MAX 512
#define quiet_log(fmt, args...) do { \
if(fake_quiet) { \
crm_trace(fmt, ##args); \
} else { \
printf(fmt , ##args); \
} \
} while(0)
#define NEW_NODE_TEMPLATE "//"XML_CIB_TAG_NODE"[@uname='%s']"
#define NODE_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']"
#define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']"
static void
inject_transient_attr(xmlNode * cib_node, const char *name, const char *value)
{
xmlNode *attrs = NULL;
xmlNode *instance_attrs = NULL;
xmlChar *node_path;
const char *node_uuid = ID(cib_node);
node_path = xmlGetNodePath(cib_node);
quiet_log(" + Injecting attribute %s=%s into %s '%s'\n",
name, value, node_path, ID(cib_node));
free(node_path);
attrs = first_named_child(cib_node, XML_TAG_TRANSIENT_NODEATTRS);
if (attrs == NULL) {
attrs = create_xml_node(cib_node, XML_TAG_TRANSIENT_NODEATTRS);
crm_xml_add(attrs, XML_ATTR_ID, node_uuid);
}
instance_attrs = first_named_child(attrs, XML_TAG_ATTR_SETS);
if (instance_attrs == NULL) {
instance_attrs = create_xml_node(attrs, XML_TAG_ATTR_SETS);
crm_xml_add(instance_attrs, XML_ATTR_ID, node_uuid);
}
crm_create_nvpair_xml(instance_attrs, NULL, name, value);
}
static void
update_failcounts(xmlNode * cib_node, const char *resource, const char *task,
guint interval_ms, int rc)
{
if (rc == 0) {
return;
} else if ((rc == 7) && (interval_ms == 0)) {
return;
} else {
char *name = NULL;
char *now = crm_ttoa(time(NULL));
name = crm_failcount_name(resource, task, interval_ms);
inject_transient_attr(cib_node, name, "value++");
free(name);
name = crm_lastfailure_name(resource, task, interval_ms);
inject_transient_attr(cib_node, name, now);
free(name);
free(now);
}
}
static void
create_node_entry(cib_t * cib_conn, const char *node)
{
int rc = pcmk_ok;
char *xpath = crm_strdup_printf(NEW_NODE_TEMPLATE, node);
rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local);
if (rc == -ENXIO) {
xmlNode *cib_object = create_xml_node(NULL, XML_CIB_TAG_NODE);
crm_xml_add(cib_object, XML_ATTR_ID, node); // Use node name as ID
crm_xml_add(cib_object, XML_ATTR_UNAME, node);
cib_conn->cmds->create(cib_conn, XML_CIB_TAG_NODES, cib_object,
cib_sync_call | cib_scope_local);
/* Not bothering with subsequent query to see if it exists,
we'll bomb out later in the call to query_node_uuid()... */
free_xml(cib_object);
}
free(xpath);
}
static lrmd_event_data_t *
create_op(xmlNode *cib_resource, const char *task, guint interval_ms,
int outcome)
{
lrmd_event_data_t *op = NULL;
xmlNode *xop = NULL;
- op = calloc(1, sizeof(lrmd_event_data_t));
-
- op->rsc_id = strdup(ID(cib_resource));
- op->interval_ms = interval_ms;
- op->op_type = strdup(task);
-
+ op = lrmd_new_event(ID(cib_resource), task, interval_ms);
op->rc = outcome;
op->op_status = 0;
op->params = NULL; /* TODO: Fill me in */
op->t_run = (unsigned int) time(NULL);
op->t_rcchange = op->t_run;
op->call_id = 0;
for (xop = __xml_first_child_element(cib_resource); xop != NULL;
xop = __xml_next_element(xop)) {
int tmp = 0;
crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp);
if (tmp > op->call_id) {
op->call_id = tmp;
}
}
op->call_id++;
return op;
}
static xmlNode *
inject_op(xmlNode * cib_resource, lrmd_event_data_t * op, int target_rc)
{
return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET,
target_rc, NULL, crm_system_name,
LOG_TRACE);
}
static xmlNode *
inject_node_state(cib_t * cib_conn, const char *node, const char *uuid)
{
int rc = pcmk_ok;
xmlNode *cib_object = NULL;
char *xpath = crm_strdup_printf(NODE_TEMPLATE, node);
if (bringing_nodes_online) {
create_node_entry(cib_conn, node);
}
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath | cib_sync_call | cib_scope_local);
if (cib_object && ID(cib_object) == NULL) {
crm_err("Detected multiple node_state entries for xpath=%s, bailing", xpath);
crm_log_xml_warn(cib_object, "Duplicates");
free(xpath);
crm_exit(CRM_EX_SOFTWARE);
return NULL; // not reached, but makes static analysis happy
}
if (rc == -ENXIO) {
char *found_uuid = NULL;
if (uuid == NULL) {
query_node_uuid(cib_conn, node, &found_uuid, NULL);
} else {
found_uuid = strdup(uuid);
}
cib_object = create_xml_node(NULL, XML_CIB_TAG_STATE);
crm_xml_add(cib_object, XML_ATTR_UUID, found_uuid);
crm_xml_add(cib_object, XML_ATTR_UNAME, node);
cib_conn->cmds->create(cib_conn, XML_CIB_TAG_STATUS, cib_object,
cib_sync_call | cib_scope_local);
free_xml(cib_object);
free(found_uuid);
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath | cib_sync_call | cib_scope_local);
crm_trace("injecting node state for %s. rc is %d", node, rc);
}
free(xpath);
CRM_ASSERT(rc == pcmk_ok);
return cib_object;
}
static xmlNode *
modify_node(cib_t * cib_conn, char *node, gboolean up)
{
xmlNode *cib_node = inject_node_state(cib_conn, node, NULL);
if (up) {
crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_YES);
crm_xml_add(cib_node, XML_NODE_IS_PEER, ONLINESTATUS);
crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_MEMBER);
crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_MEMBER);
} else {
crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO);
crm_xml_add(cib_node, XML_NODE_IS_PEER, OFFLINESTATUS);
crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_DOWN);
crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_DOWN);
}
crm_xml_add(cib_node, XML_ATTR_ORIGIN, crm_system_name);
return cib_node;
}
static xmlNode *
find_resource_xml(xmlNode * cib_node, const char *resource)
{
xmlNode *match = NULL;
const char *node = crm_element_value(cib_node, XML_ATTR_UNAME);
char *xpath = crm_strdup_printf(RSC_TEMPLATE, node, resource);
match = get_xpath_object(xpath, cib_node, LOG_TRACE);
free(xpath);
return match;
}
static xmlNode *
inject_resource(xmlNode * cib_node, const char *resource, const char *lrm_name,
const char *rclass, const char *rtype, const char *rprovider)
{
xmlNode *lrm = NULL;
xmlNode *container = NULL;
xmlNode *cib_resource = NULL;
char *xpath = NULL;
cib_resource = find_resource_xml(cib_node, resource);
if (cib_resource != NULL) {
/* If an existing LRM history entry uses the resource name,
* continue using it, even if lrm_name is different.
*/
return cib_resource;
}
// Check for history entry under preferred name
if (strcmp(resource, lrm_name)) {
cib_resource = find_resource_xml(cib_node, lrm_name);
if (cib_resource != NULL) {
return cib_resource;
}
}
/* One day, add query for class, provider, type */
if (rclass == NULL || rtype == NULL) {
fprintf(stderr, "Resource %s not found in the status section of %s."
" Please supply the class and type to continue\n", resource, ID(cib_node));
return NULL;
} else if (safe_str_neq(rclass, PCMK_RESOURCE_CLASS_OCF)
&& safe_str_neq(rclass, PCMK_RESOURCE_CLASS_STONITH)
&& safe_str_neq(rclass, PCMK_RESOURCE_CLASS_SERVICE)
&& safe_str_neq(rclass, PCMK_RESOURCE_CLASS_UPSTART)
&& safe_str_neq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD)
&& safe_str_neq(rclass, PCMK_RESOURCE_CLASS_LSB)) {
fprintf(stderr, "Invalid class for %s: %s\n", resource, rclass);
return NULL;
} else if (is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider)
&& (rprovider == NULL)) {
fprintf(stderr, "Please specify the provider for resource %s\n", resource);
return NULL;
}
xpath = (char *)xmlGetNodePath(cib_node);
crm_info("Injecting new resource %s into %s '%s'", lrm_name, xpath, ID(cib_node));
free(xpath);
lrm = first_named_child(cib_node, XML_CIB_TAG_LRM);
if (lrm == NULL) {
const char *node_uuid = ID(cib_node);
lrm = create_xml_node(cib_node, XML_CIB_TAG_LRM);
crm_xml_add(lrm, XML_ATTR_ID, node_uuid);
}
container = first_named_child(lrm, XML_LRM_TAG_RESOURCES);
if (container == NULL) {
container = create_xml_node(lrm, XML_LRM_TAG_RESOURCES);
}
cib_resource = create_xml_node(container, XML_LRM_TAG_RESOURCE);
// If we're creating a new entry, use the preferred name
crm_xml_add(cib_resource, XML_ATTR_ID, lrm_name);
crm_xml_add(cib_resource, XML_AGENT_ATTR_CLASS, rclass);
crm_xml_add(cib_resource, XML_AGENT_ATTR_PROVIDER, rprovider);
crm_xml_add(cib_resource, XML_ATTR_TYPE, rtype);
return cib_resource;
}
#define XPATH_MAX 1024
static int
find_ticket_state(cib_t * the_cib, const char *ticket_id, xmlNode ** ticket_state_xml)
{
int offset = 0;
int rc = pcmk_ok;
xmlNode *xml_search = NULL;
char *xpath_string = NULL;
CRM_ASSERT(ticket_state_xml != NULL);
*ticket_state_xml = NULL;
xpath_string = calloc(1, XPATH_MAX);
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "%s", "/cib/status/tickets");
if (ticket_id) {
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "/%s[@id=\"%s\"]",
XML_CIB_TAG_TICKET_STATE, ticket_id);
}
CRM_LOG_ASSERT(offset > 0);
rc = the_cib->cmds->query(the_cib, xpath_string, &xml_search,
cib_sync_call | cib_scope_local | cib_xpath);
if (rc != pcmk_ok) {
goto bail;
}
crm_log_xml_debug(xml_search, "Match");
if (xml_has_children(xml_search)) {
if (ticket_id) {
fprintf(stdout, "Multiple ticket_states match ticket_id=%s\n", ticket_id);
}
*ticket_state_xml = xml_search;
} else {
*ticket_state_xml = xml_search;
}
bail:
free(xpath_string);
return rc;
}
static int
set_ticket_state_attr(const char *ticket_id, const char *attr_name,
const char *attr_value, cib_t * cib, int cib_options)
{
int rc = pcmk_ok;
xmlNode *xml_top = NULL;
xmlNode *ticket_state_xml = NULL;
rc = find_ticket_state(cib, ticket_id, &ticket_state_xml);
if (rc == pcmk_ok) {
crm_debug("Found a match state for ticket: id=%s", ticket_id);
xml_top = ticket_state_xml;
} else if (rc != -ENXIO) {
return rc;
} else {
xmlNode *xml_obj = NULL;
xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS);
xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS);
ticket_state_xml = create_xml_node(xml_obj, XML_CIB_TAG_TICKET_STATE);
crm_xml_add(ticket_state_xml, XML_ATTR_ID, ticket_id);
}
crm_xml_add(ticket_state_xml, attr_name, attr_value);
crm_log_xml_debug(xml_top, "Update");
rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, xml_top, cib_options);
free_xml(xml_top);
return rc;
}
void
modify_configuration(pe_working_set_t * data_set, cib_t *cib,
const char *quorum, const char *watchdog, GListPtr node_up, GListPtr node_down, GListPtr node_fail,
GListPtr op_inject, GListPtr ticket_grant, GListPtr ticket_revoke,
GListPtr ticket_standby, GListPtr ticket_activate)
{
int rc = pcmk_ok;
GListPtr gIter = NULL;
xmlNode *cib_op = NULL;
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
lrmd_event_data_t *op = NULL;
if (quorum) {
xmlNode *top = create_xml_node(NULL, XML_TAG_CIB);
quiet_log(" + Setting quorum: %s\n", quorum);
/* crm_xml_add(top, XML_ATTR_DC_UUID, dc_uuid); */
crm_xml_add(top, XML_ATTR_HAVE_QUORUM, quorum);
rc = cib->cmds->modify(cib, NULL, top, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
if (watchdog) {
quiet_log(" + Setting watchdog: %s\n", watchdog);
rc = update_attr_delegate(cib, cib_sync_call | cib_scope_local,
XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL,
XML_ATTR_HAVE_WATCHDOG, watchdog, FALSE, NULL, NULL);
CRM_ASSERT(rc == pcmk_ok);
}
for (gIter = node_up; gIter != NULL; gIter = gIter->next) {
char *node = (char *)gIter->data;
quiet_log(" + Bringing node %s online\n", node);
cib_node = modify_node(cib, node, TRUE);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
}
for (gIter = node_down; gIter != NULL; gIter = gIter->next) {
char xpath[STATUS_PATH_MAX];
char *node = (char *)gIter->data;
quiet_log(" + Taking node %s offline\n", node);
cib_node = modify_node(cib, node, FALSE);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_CIB_TAG_LRM);
cib->cmds->remove(cib, xpath, NULL,
cib_xpath | cib_sync_call | cib_scope_local);
snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node,
XML_TAG_TRANSIENT_NODEATTRS);
cib->cmds->remove(cib, xpath, NULL,
cib_xpath | cib_sync_call | cib_scope_local);
}
for (gIter = node_fail; gIter != NULL; gIter = gIter->next) {
char *node = (char *)gIter->data;
quiet_log(" + Failing node %s\n", node);
cib_node = modify_node(cib, node, TRUE);
crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
}
for (gIter = ticket_grant; gIter != NULL; gIter = gIter->next) {
char *ticket_id = (char *)gIter->data;
quiet_log(" + Granting ticket %s\n", ticket_id);
rc = set_ticket_state_attr(ticket_id, "granted", "true",
cib, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
for (gIter = ticket_revoke; gIter != NULL; gIter = gIter->next) {
char *ticket_id = (char *)gIter->data;
quiet_log(" + Revoking ticket %s\n", ticket_id);
rc = set_ticket_state_attr(ticket_id, "granted", "false",
cib, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
for (gIter = ticket_standby; gIter != NULL; gIter = gIter->next) {
char *ticket_id = (char *)gIter->data;
quiet_log(" + Making ticket %s standby\n", ticket_id);
rc = set_ticket_state_attr(ticket_id, "standby", "true",
cib, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
for (gIter = ticket_activate; gIter != NULL; gIter = gIter->next) {
char *ticket_id = (char *)gIter->data;
quiet_log(" + Activating ticket %s\n", ticket_id);
rc = set_ticket_state_attr(ticket_id, "standby", "false",
cib, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
for (gIter = op_inject; gIter != NULL; gIter = gIter->next) {
char *spec = (char *)gIter->data;
int rc = 0;
int outcome = 0;
guint interval_ms = 0;
char *key = NULL;
char *node = NULL;
char *task = NULL;
char *resource = NULL;
const char *rtype = NULL;
const char *rclass = NULL;
const char *rprovider = NULL;
resource_t *rsc = NULL;
quiet_log(" + Injecting %s into the configuration\n", spec);
key = calloc(1, strlen(spec) + 1);
node = calloc(1, strlen(spec) + 1);
rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome);
if (rc != 3) {
fprintf(stderr, "Invalid operation spec: %s. Only found %d fields\n", spec, rc);
free(key);
free(node);
continue;
}
parse_op_key(key, &resource, &task, &interval_ms);
rsc = pe_find_resource(data_set->resources, resource);
if (rsc == NULL) {
fprintf(stderr, " - Invalid resource name: %s\n", resource);
} else {
rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE);
rprovider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
cib_node = inject_node_state(cib, node, NULL);
CRM_ASSERT(cib_node != NULL);
update_failcounts(cib_node, resource, task, interval_ms, outcome);
cib_resource = inject_resource(cib_node, resource, resource,
rclass, rtype, rprovider);
CRM_ASSERT(cib_resource != NULL);
op = create_op(cib_resource, task, interval_ms, outcome);
CRM_ASSERT(op != NULL);
cib_op = inject_op(cib_resource, op, 0);
CRM_ASSERT(cib_op != NULL);
lrmd_free_event(op);
rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
free(task);
free(node);
free(key);
}
}
static gboolean
exec_pseudo_action(crm_graph_t * graph, crm_action_t * action)
{
const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
action->confirmed = TRUE;
quiet_log(" * Pseudo action: %s%s%s\n", task, node ? " on " : "", node ? node : "");
update_graph(graph, action);
return TRUE;
}
static gboolean
exec_rsc_action(crm_graph_t * graph, crm_action_t * action)
{
int rc = 0;
GListPtr gIter = NULL;
lrmd_event_data_t *op = NULL;
int target_outcome = 0;
const char *rtype = NULL;
const char *rclass = NULL;
const char *resource = NULL;
const char *rprovider = NULL;
const char *lrm_name = NULL;
const char *operation = crm_element_value(action->xml, "operation");
const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC);
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET);
char *uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID);
const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE);
if (safe_str_eq(operation, CRM_OP_PROBED)
|| safe_str_eq(operation, CRM_OP_REPROBE)) {
crm_info("Skipping %s op for %s", operation, node);
goto done;
}
if (action_rsc == NULL) {
crm_log_xml_err(action->xml, "Bad");
free(node); free(uuid);
return FALSE;
}
/* Look for the preferred name
* If not found, try the expected 'local' name
* If not found use the preferred name anyway
*/
resource = crm_element_value(action_rsc, XML_ATTR_ID);
CRM_ASSERT(resource != NULL); // makes static analysis happy
lrm_name = resource; // Preferred name when writing history
if (pe_find_resource(fake_resource_list, resource) == NULL) {
const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG);
if (longname && pe_find_resource(fake_resource_list, longname)) {
resource = longname;
}
}
if (safe_str_eq(operation, "delete") || safe_str_eq(operation, RSC_METADATA)) {
quiet_log(" * Resource action: %-15s %s on %s\n", resource, operation, node);
goto done;
}
rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS);
rtype = crm_element_value(action_rsc, XML_ATTR_TYPE);
rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER);
if (target_rc_s != NULL) {
target_outcome = crm_parse_int(target_rc_s, "0");
}
CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call | cib_scope_local) ==
pcmk_ok);
cib_node = inject_node_state(fake_cib, node, (router_node? node : uuid));
CRM_ASSERT(cib_node != NULL);
cib_resource = inject_resource(cib_node, resource, lrm_name,
rclass, rtype, rprovider);
if (cib_resource == NULL) {
crm_err("invalid resource in transition");
free(node); free(uuid);
free_xml(cib_node);
return FALSE;
}
op = convert_graph_action(cib_resource, action, 0, target_outcome);
if (op->interval_ms) {
quiet_log(" * Resource action: %-15s %s=%u on %s\n",
resource, op->op_type, op->interval_ms, node);
} else {
quiet_log(" * Resource action: %-15s %s on %s\n", resource, op->op_type, node);
}
for (gIter = fake_op_fail_list; gIter != NULL; gIter = gIter->next) {
char *spec = (char *)gIter->data;
char *key = NULL;
const char *match_name = NULL;
// Allow user to specify anonymous clone with or without instance number
key = crm_strdup_printf(CRM_OP_FMT "@%s=", resource, op->op_type,
op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = resource;
}
free(key);
if ((match_name == NULL) && strcmp(resource, lrm_name)) {
key = crm_strdup_printf(CRM_OP_FMT "@%s=", lrm_name, op->op_type,
op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = lrm_name;
}
free(key);
}
if (match_name != NULL) {
rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc);
// ${match_name}_${task}_${interval_in_ms}@${node}=${rc}
if (rc != 1) {
fprintf(stderr,
"Invalid failed operation spec: %s. Result code must be integer\n",
spec);
continue;
}
action->failed = TRUE;
graph->abort_priority = INFINITY;
printf("\tPretending action %d failed with rc=%d\n", action->id, op->rc);
update_failcounts(cib_node, match_name, op->op_type,
op->interval_ms, op->rc);
break;
}
}
inject_op(cib_resource, op, target_outcome);
lrmd_free_event(op);
rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
done:
free(node); free(uuid);
free_xml(cib_node);
action->confirmed = TRUE;
update_graph(graph, action);
return TRUE;
}
static gboolean
exec_crmd_action(crm_graph_t * graph, crm_action_t * action)
{
const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
action->confirmed = TRUE;
if(rsc) {
quiet_log(" * Cluster action: %s for %s on %s\n", task, ID(rsc), node);
} else {
quiet_log(" * Cluster action: %s on %s\n", task, node);
}
update_graph(graph, action);
return TRUE;
}
static gboolean
exec_stonith_action(crm_graph_t * graph, crm_action_t * action)
{
const char *op = crm_meta_value(action->params, "stonith_action");
char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET);
quiet_log(" * Fencing %s (%s)\n", target, op);
if(safe_str_neq(op, "on")) {
int rc = 0;
char xpath[STATUS_PATH_MAX];
xmlNode *cib_node = modify_node(fake_cib, target, FALSE);
crm_xml_add(cib_node, XML_ATTR_ORIGIN, __FUNCTION__);
CRM_ASSERT(cib_node != NULL);
rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_CIB_TAG_LRM);
fake_cib->cmds->remove(fake_cib, xpath, NULL,
cib_xpath | cib_sync_call | cib_scope_local);
snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target,
XML_TAG_TRANSIENT_NODEATTRS);
fake_cib->cmds->remove(fake_cib, xpath, NULL,
cib_xpath | cib_sync_call | cib_scope_local);
free_xml(cib_node);
}
action->confirmed = TRUE;
update_graph(graph, action);
free(target);
return TRUE;
}
int
run_simulation(pe_working_set_t * data_set, cib_t *cib, GListPtr op_fail_list, bool quiet)
{
crm_graph_t *transition = NULL;
enum transition_status graph_rc = -1;
crm_graph_functions_t exec_fns = {
exec_pseudo_action,
exec_rsc_action,
exec_crmd_action,
exec_stonith_action,
};
fake_cib = cib;
fake_quiet = quiet;
fake_op_fail_list = op_fail_list;
quiet_log("\nExecuting cluster transition:\n");
set_graph_functions(&exec_fns);
transition = unpack_graph(data_set->graph, crm_system_name);
print_graph(LOG_DEBUG, transition);
fake_resource_list = data_set->resources;
do {
graph_rc = run_graph(transition);
} while (graph_rc == transition_active);
fake_resource_list = NULL;
if (graph_rc != transition_complete) {
fprintf(stdout, "Transition failed: %s\n", transition_status(graph_rc));
print_graph(LOG_ERR, transition);
}
destroy_graph(transition);
if (graph_rc != transition_complete) {
fprintf(stdout, "An invalid transition was produced\n");
}
if (quiet == FALSE) {
xmlNode *cib_object = NULL;
int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call | cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
pe_reset_working_set(data_set);
data_set->input = cib_object;
}
if (graph_rc != transition_complete) {
return graph_rc;
}
return 0;
}
diff --git a/lib/pacemaker/pcmk_trans_unpack.c b/lib/pacemaker/pcmk_trans_unpack.c
index e57f3867aa..3e53289b2c 100644
--- a/lib/pacemaker/pcmk_trans_unpack.c
+++ b/lib/pacemaker/pcmk_trans_unpack.c
@@ -1,331 +1,328 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
static crm_action_t *
unpack_action(synapse_t * parent, xmlNode * xml_action)
{
crm_action_t *action = NULL;
const char *value = crm_element_value(xml_action, XML_ATTR_ID);
if (value == NULL) {
crm_err("Actions must have an id!");
crm_log_xml_trace(xml_action, "Action with missing id");
return NULL;
}
action = calloc(1, sizeof(crm_action_t));
if (action == NULL) {
crm_perror(LOG_CRIT, "Cannot unpack action");
crm_log_xml_trace(xml_action, "Lost action");
return NULL;
}
action->id = crm_parse_int(value, NULL);
action->type = action_type_rsc;
action->xml = copy_xml(xml_action);
action->synapse = parent;
if (safe_str_eq(crm_element_name(action->xml), XML_GRAPH_TAG_RSC_OP)) {
action->type = action_type_rsc;
} else if (safe_str_eq(crm_element_name(action->xml), XML_GRAPH_TAG_PSEUDO_EVENT)) {
action->type = action_type_pseudo;
} else if (safe_str_eq(crm_element_name(action->xml), XML_GRAPH_TAG_CRM_EVENT)) {
action->type = action_type_crm;
}
action->params = xml2list(action->xml);
value = g_hash_table_lookup(action->params, "CRM_meta_timeout");
if (value != NULL) {
action->timeout = crm_parse_int(value, NULL);
}
/* Take start-delay into account for the timeout of the action timer */
value = g_hash_table_lookup(action->params, "CRM_meta_start_delay");
if (value != NULL) {
action->timeout += crm_parse_int(value, NULL);
}
value = g_hash_table_lookup(action->params, "CRM_meta_interval");
if (value != NULL) {
action->interval_ms = crm_parse_ms(value);
}
value = g_hash_table_lookup(action->params, "CRM_meta_can_fail");
if (value != NULL) {
crm_str_to_boolean(value, &(action->can_fail));
}
crm_trace("Action %d has timer set to %dms", action->id, action->timeout);
return action;
}
static synapse_t *
unpack_synapse(crm_graph_t * new_graph, xmlNode * xml_synapse)
{
const char *value = NULL;
xmlNode *inputs = NULL;
xmlNode *action_set = NULL;
synapse_t *new_synapse = NULL;
CRM_CHECK(xml_synapse != NULL, return NULL);
crm_trace("looking in synapse %s", ID(xml_synapse));
new_synapse = calloc(1, sizeof(synapse_t));
new_synapse->id = crm_parse_int(ID(xml_synapse), NULL);
value = crm_element_value(xml_synapse, XML_CIB_ATTR_PRIORITY);
if (value != NULL) {
new_synapse->priority = crm_parse_int(value, NULL);
}
CRM_CHECK(new_synapse->id >= 0, free(new_synapse);
return NULL);
new_graph->num_synapses++;
crm_trace("look for actions in synapse %s", crm_element_value(xml_synapse, XML_ATTR_ID));
for (action_set = __xml_first_child(xml_synapse); action_set != NULL;
action_set = __xml_next(action_set)) {
if (crm_str_eq((const char *)action_set->name, "action_set", TRUE)) {
xmlNode *action = NULL;
for (action = __xml_first_child(action_set); action != NULL;
action = __xml_next(action)) {
crm_action_t *new_action = unpack_action(new_synapse, action);
if (new_action == NULL) {
continue;
}
new_graph->num_actions++;
crm_trace("Adding action %d to synapse %d", new_action->id, new_synapse->id);
new_synapse->actions = g_list_append(new_synapse->actions, new_action);
}
}
}
crm_trace("look for inputs in synapse %s", ID(xml_synapse));
for (inputs = __xml_first_child(xml_synapse); inputs != NULL; inputs = __xml_next(inputs)) {
if (crm_str_eq((const char *)inputs->name, "inputs", TRUE)) {
xmlNode *trigger = NULL;
for (trigger = __xml_first_child(inputs); trigger != NULL;
trigger = __xml_next(trigger)) {
xmlNode *input = NULL;
for (input = __xml_first_child(trigger); input != NULL; input = __xml_next(input)) {
crm_action_t *new_input = unpack_action(new_synapse, input);
if (new_input == NULL) {
continue;
}
crm_trace("Adding input %d to synapse %d", new_input->id, new_synapse->id);
new_synapse->inputs = g_list_append(new_synapse->inputs, new_input);
}
}
}
}
return new_synapse;
}
static void destroy_action(crm_action_t * action);
crm_graph_t *
unpack_graph(xmlNode * xml_graph, const char *reference)
{
/*
id = -1;
new_graph->abort_priority = 0;
new_graph->network_delay = 0;
new_graph->stonith_timeout = 0;
new_graph->completion_action = tg_done;
if (reference) {
new_graph->source = strdup(reference);
} else {
new_graph->source = strdup("unknown");
}
if (xml_graph != NULL) {
t_id = crm_element_value(xml_graph, "transition_id");
CRM_CHECK(t_id != NULL, free(new_graph);
return NULL);
new_graph->id = crm_parse_int(t_id, "-1");
time = crm_element_value(xml_graph, "cluster-delay");
CRM_CHECK(time != NULL, free(new_graph);
return NULL);
new_graph->network_delay = crm_parse_interval_spec(time);
time = crm_element_value(xml_graph, "stonith-timeout");
if (time == NULL) {
new_graph->stonith_timeout = new_graph->network_delay;
} else {
new_graph->stonith_timeout = crm_parse_interval_spec(time);
}
t_id = crm_element_value(xml_graph, "batch-limit");
new_graph->batch_limit = crm_parse_int(t_id, "0");
t_id = crm_element_value(xml_graph, "migration-limit");
new_graph->migration_limit = crm_parse_int(t_id, "-1");
}
for (synapse = __xml_first_child(xml_graph); synapse != NULL; synapse = __xml_next(synapse)) {
if (crm_str_eq((const char *)synapse->name, "synapse", TRUE)) {
synapse_t *new_synapse = unpack_synapse(new_graph, synapse);
if (new_synapse != NULL) {
new_graph->synapses = g_list_append(new_graph->synapses, new_synapse);
}
}
}
crm_debug("Unpacked transition %d: %d actions in %d synapses",
new_graph->id, new_graph->num_actions, new_graph->num_synapses);
return new_graph;
}
static void
destroy_action(crm_action_t * action)
{
if (action->timer && action->timer->source_id != 0) {
crm_warn("Cancelling timer for action %d (src=%d)", action->id, action->timer->source_id);
g_source_remove(action->timer->source_id);
}
if (action->params) {
g_hash_table_destroy(action->params);
}
free_xml(action->xml);
free(action->timer);
free(action);
}
static void
destroy_synapse(synapse_t * synapse)
{
while (synapse->actions != NULL) {
crm_action_t *action = g_list_nth_data(synapse->actions, 0);
synapse->actions = g_list_remove(synapse->actions, action);
destroy_action(action);
}
while (synapse->inputs != NULL) {
crm_action_t *action = g_list_nth_data(synapse->inputs, 0);
synapse->inputs = g_list_remove(synapse->inputs, action);
destroy_action(action);
}
free(synapse);
}
void
destroy_graph(crm_graph_t * graph)
{
if (graph == NULL) {
return;
}
while (graph->synapses != NULL) {
synapse_t *synapse = g_list_nth_data(graph->synapses, 0);
graph->synapses = g_list_remove(graph->synapses, synapse);
destroy_synapse(synapse);
}
free(graph->source);
free(graph);
}
lrmd_event_data_t *
convert_graph_action(xmlNode * resource, crm_action_t * action, int status, int rc)
{
xmlNode *xop = NULL;
lrmd_event_data_t *op = NULL;
GHashTableIter iter;
const char *name = NULL;
const char *value = NULL;
xmlNode *action_resource = NULL;
CRM_CHECK(action != NULL, return NULL);
CRM_CHECK(action->type == action_type_rsc, return NULL);
action_resource = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
CRM_CHECK(action_resource != NULL, crm_log_xml_warn(action->xml, "Bad");
return NULL);
- op = calloc(1, sizeof(lrmd_event_data_t));
-
- op->rsc_id = strdup(ID(action_resource));
- op->interval_ms = action->interval_ms;
- op->op_type = strdup(crm_element_value(action->xml, XML_LRM_ATTR_TASK));
-
+ op = lrmd_new_event(ID(action_resource),
+ crm_element_value(action->xml, XML_LRM_ATTR_TASK),
+ action->interval_ms);
op->rc = rc;
op->op_status = status;
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
op->params = g_hash_table_new_full(crm_str_hash, g_str_equal, free, free);
g_hash_table_iter_init(&iter, action->params);
while (g_hash_table_iter_next(&iter, (void **)&name, (void **)&value)) {
g_hash_table_insert(op->params, strdup(name), strdup(value));
}
for (xop = __xml_first_child(resource); xop != NULL; xop = __xml_next(xop)) {
int tmp = 0;
crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp);
crm_debug("Got call_id=%d for %s", tmp, ID(resource));
if (tmp > op->call_id) {
op->call_id = tmp;
}
}
op->call_id++;
return op;
}
diff --git a/lib/pengine/common.c b/lib/pengine/common.c
index da39c99c67..e72a033dd7 100644
--- a/lib/pengine/common.c
+++ b/lib/pengine/common.c
@@ -1,474 +1,496 @@
/*
- * Copyright 2004-2018 Andrew Beekhof
+ * Copyright 2004-2020 the Pacemaker project contributors
+ *
+ * The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
gboolean was_processing_error = FALSE;
gboolean was_processing_warning = FALSE;
static gboolean
check_health(const char *value)
{
if (safe_str_eq(value, "none")) {
return TRUE;
} else if (safe_str_eq(value, "custom")) {
return TRUE;
} else if (safe_str_eq(value, "only-green")) {
return TRUE;
} else if (safe_str_eq(value, "progressive")) {
return TRUE;
} else if (safe_str_eq(value, "migrate-on-red")) {
return TRUE;
}
return FALSE;
}
static gboolean
check_stonith_action(const char *value)
{
if (safe_str_eq(value, "reboot")) {
return TRUE;
} else if (safe_str_eq(value, "poweroff")) {
return TRUE;
} else if (safe_str_eq(value, "off")) {
return TRUE;
}
return FALSE;
}
static gboolean
check_placement_strategy(const char *value)
{
if (safe_str_eq(value, "default")) {
return TRUE;
} else if (safe_str_eq(value, "utilization")) {
return TRUE;
} else if (safe_str_eq(value, "minimal")) {
return TRUE;
} else if (safe_str_eq(value, "balanced")) {
return TRUE;
}
return FALSE;
}
/* *INDENT-OFF* */
static pe_cluster_option pe_opts[] = {
/* name, old-name, validate, default, description */
{ "no-quorum-policy", NULL, "enum", "stop, freeze, ignore, suicide", "stop", &check_quorum,
"What to do when the cluster does not have quorum", NULL },
{ "symmetric-cluster", NULL, "boolean", NULL, "true", &check_boolean,
"All resources can run anywhere by default", NULL },
{ "maintenance-mode", NULL, "boolean", NULL, "false", &check_boolean,
"Should the cluster monitor resources and start/stop them as required", NULL },
{ "start-failure-is-fatal", NULL, "boolean", NULL, "true", &check_boolean, "Always treat start failures as fatal",
"When set to TRUE, the cluster will immediately ban a resource from a node if it fails to start there. When FALSE, the cluster will instead check the resource's fail count against its migration-threshold." },
{ "enable-startup-probes", NULL, "boolean", NULL, "true", &check_boolean,
"Should the cluster check for active resources during startup", NULL },
+ {
+ XML_CONFIG_ATTR_SHUTDOWN_LOCK,
+ NULL, "boolean", NULL, "false", &check_boolean,
+ "Whether to lock resources to a cleanly shut down node",
+ "When true, resources active on a node when it is cleanly shut down "
+ "are kept \"locked\" to that node (not allowed to run elsewhere) "
+ "until they start again on that node after it rejoins (or for at "
+ "most shutdown-lock-limit, if set). Stonith resources and "
+ "Pacemaker Remote connections are never locked. Clone and bundle "
+ "instances and the master role of promotable clones are currently "
+ "never locked, though support could be added in a future release."
+ },
+ {
+ XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT,
+ NULL, "time", NULL, "0", &check_timer,
+ "Do not lock resources to a cleanly shut down node longer than this",
+ "If shutdown-lock is true and this is set to a nonzero time duration, "
+ "shutdown locks will expire after this much time has passed since "
+ "the shutdown was initiated, even if the node has not rejoined."
+ },
/* Stonith Options */
{ "stonith-enabled", NULL, "boolean", NULL, "true", &check_boolean,
"Failed nodes are STONITH'd", NULL },
{ "stonith-action", NULL, "enum", "reboot, off, poweroff", "reboot", &check_stonith_action,
"Action to send to STONITH device ('poweroff' is a deprecated alias for 'off')", NULL },
{ "stonith-timeout", NULL, "time", NULL, "60s", &check_timer,
"How long to wait for the STONITH action (reboot,on,off) to complete", NULL },
{ XML_ATTR_HAVE_WATCHDOG, NULL, "boolean", NULL, "false", &check_boolean,
"Enable watchdog integration", "Set automatically by the cluster if SBD is detected. User configured values are ignored." },
{ "concurrent-fencing", NULL, "boolean", NULL,
#ifdef DEFAULT_CONCURRENT_FENCING_TRUE
"true",
#else
"false",
#endif
&check_boolean,
"Allow performing fencing operations in parallel", NULL },
{ "startup-fencing", NULL, "boolean", NULL, "true", &check_boolean,
"STONITH unseen nodes", "Advanced Use Only! Not using the default is very unsafe!" },
/* Timeouts etc */
{ "cluster-delay", NULL, "time", NULL, "60s", &check_time,
"Round trip delay over the network (excluding action execution)",
"The \"correct\" value will depend on the speed and load of your network and cluster nodes." },
{ "batch-limit", NULL, "integer", NULL, "0", &check_number,
"The number of jobs that the TE is allowed to execute in parallel",
"The \"correct\" value will depend on the speed and load of your network and cluster nodes." },
{ "migration-limit", NULL, "integer", NULL, "-1", &check_number,
"The number of migration jobs that the TE is allowed to execute in parallel on a node"},
/* Orphans and stopping */
{ "stop-all-resources", NULL, "boolean", NULL, "false", &check_boolean,
"Should the cluster stop all active resources", NULL },
{ "stop-orphan-resources", NULL, "boolean", NULL, "true", &check_boolean,
"Should deleted resources be stopped", NULL },
{ "stop-orphan-actions", NULL, "boolean", NULL, "true", &check_boolean,
"Should deleted actions be cancelled", NULL },
{ "remove-after-stop", NULL, "boolean", NULL, "false", &check_boolean,
"Remove resources from the executor after they are stopped",
"Always set this to false. Other values are, at best, poorly tested and potentially dangerous." },
/* { "", "", , "0", "", NULL }, */
/* Storing inputs */
{
"pe-error-series-max", NULL, "integer", NULL, "-1", &check_number,
"The number of scheduler inputs resulting in ERRORs to save",
"Zero to disable, -1 to store unlimited"
},
{
"pe-warn-series-max", NULL, "integer", NULL, "5000", &check_number,
"The number of scheduler inputs resulting in WARNINGs to save",
"Zero to disable, -1 to store unlimited"
},
{
"pe-input-series-max", NULL, "integer", NULL, "4000", &check_number,
"The number of other scheduler inputs to save",
"Zero to disable, -1 to store unlimited"
},
/* Node health */
{ "node-health-strategy", NULL, "enum", "none, migrate-on-red, only-green, progressive, custom", "none", &check_health,
"The strategy combining node attributes to determine overall node health.",
"Requires external entities to create node attributes (named with the prefix '#health') with values: 'red', 'yellow' or 'green'."},
{ "node-health-base", NULL, "integer", NULL, "0", &check_number,
"The base score assigned to a node",
"Only used when node-health-strategy is set to progressive." },
{ "node-health-green", NULL, "integer", NULL, "0", &check_number,
"The score 'green' translates to in rsc_location constraints",
"Only used when node-health-strategy is set to custom or progressive." },
{ "node-health-yellow", NULL, "integer", NULL, "0", &check_number,
"The score 'yellow' translates to in rsc_location constraints",
"Only used when node-health-strategy is set to custom or progressive." },
{ "node-health-red", NULL, "integer", NULL, "-INFINITY", &check_number,
"The score 'red' translates to in rsc_location constraints",
"Only used when node-health-strategy is set to custom or progressive." },
/*Placement Strategy*/
{ "placement-strategy", NULL, "enum", "default, utilization, minimal, balanced", "default", &check_placement_strategy,
"The strategy to determine resource placement", NULL},
};
/* *INDENT-ON* */
void
pe_metadata(void)
{
config_metadata("pacemaker-schedulerd", "1.0", "scheduler properties",
"Cluster properties used by Pacemaker's scheduler,"
" formerly known as pengine",
pe_opts, DIMOF(pe_opts));
}
void
verify_pe_options(GHashTable * options)
{
verify_all_options(options, pe_opts, DIMOF(pe_opts));
}
const char *
pe_pref(GHashTable * options, const char *name)
{
return get_cluster_pref(options, pe_opts, DIMOF(pe_opts), name);
}
const char *
fail2text(enum action_fail_response fail)
{
const char *result = "";
switch (fail) {
case action_fail_ignore:
result = "ignore";
break;
case action_fail_block:
result = "block";
break;
case action_fail_recover:
result = "recover";
break;
case action_fail_migrate:
result = "migrate";
break;
case action_fail_stop:
result = "stop";
break;
case action_fail_fence:
result = "fence";
break;
case action_fail_standby:
result = "standby";
break;
case action_fail_restart_container:
result = "restart-container";
break;
case action_fail_reset_remote:
result = "reset-remote";
break;
}
return result;
}
enum action_tasks
text2task(const char *task)
{
if (safe_str_eq(task, CRMD_ACTION_STOP)) {
return stop_rsc;
} else if (safe_str_eq(task, CRMD_ACTION_STOPPED)) {
return stopped_rsc;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
return start_rsc;
} else if (safe_str_eq(task, CRMD_ACTION_STARTED)) {
return started_rsc;
} else if (safe_str_eq(task, CRM_OP_SHUTDOWN)) {
return shutdown_crm;
} else if (safe_str_eq(task, CRM_OP_FENCE)) {
return stonith_node;
} else if (safe_str_eq(task, CRMD_ACTION_STATUS)) {
return monitor_rsc;
} else if (safe_str_eq(task, CRMD_ACTION_NOTIFY)) {
return action_notify;
} else if (safe_str_eq(task, CRMD_ACTION_NOTIFIED)) {
return action_notified;
} else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) {
return action_promote;
} else if (safe_str_eq(task, CRMD_ACTION_DEMOTE)) {
return action_demote;
} else if (safe_str_eq(task, CRMD_ACTION_PROMOTED)) {
return action_promoted;
} else if (safe_str_eq(task, CRMD_ACTION_DEMOTED)) {
return action_demoted;
}
#if SUPPORT_TRACING
if (safe_str_eq(task, CRMD_ACTION_CANCEL)) {
return no_action;
} else if (safe_str_eq(task, CRMD_ACTION_DELETE)) {
return no_action;
} else if (safe_str_eq(task, CRMD_ACTION_STATUS)) {
return no_action;
} else if (safe_str_eq(task, CRM_OP_PROBED)) {
return no_action;
} else if (safe_str_eq(task, CRM_OP_LRM_REFRESH)) {
return no_action;
} else if (safe_str_eq(task, CRMD_ACTION_MIGRATE)) {
return no_action;
} else if (safe_str_eq(task, CRMD_ACTION_MIGRATED)) {
return no_action;
}
crm_trace("Unsupported action: %s", task);
#endif
return no_action;
}
const char *
task2text(enum action_tasks task)
{
const char *result = "";
switch (task) {
case no_action:
result = "no_action";
break;
case stop_rsc:
result = CRMD_ACTION_STOP;
break;
case stopped_rsc:
result = CRMD_ACTION_STOPPED;
break;
case start_rsc:
result = CRMD_ACTION_START;
break;
case started_rsc:
result = CRMD_ACTION_STARTED;
break;
case shutdown_crm:
result = CRM_OP_SHUTDOWN;
break;
case stonith_node:
result = CRM_OP_FENCE;
break;
case monitor_rsc:
result = CRMD_ACTION_STATUS;
break;
case action_notify:
result = CRMD_ACTION_NOTIFY;
break;
case action_notified:
result = CRMD_ACTION_NOTIFIED;
break;
case action_promote:
result = CRMD_ACTION_PROMOTE;
break;
case action_promoted:
result = CRMD_ACTION_PROMOTED;
break;
case action_demote:
result = CRMD_ACTION_DEMOTE;
break;
case action_demoted:
result = CRMD_ACTION_DEMOTED;
break;
}
return result;
}
const char *
role2text(enum rsc_role_e role)
{
switch (role) {
case RSC_ROLE_UNKNOWN:
return RSC_ROLE_UNKNOWN_S;
case RSC_ROLE_STOPPED:
return RSC_ROLE_STOPPED_S;
case RSC_ROLE_STARTED:
return RSC_ROLE_STARTED_S;
case RSC_ROLE_SLAVE:
return RSC_ROLE_SLAVE_S;
case RSC_ROLE_MASTER:
return RSC_ROLE_MASTER_S;
}
CRM_CHECK(role >= RSC_ROLE_UNKNOWN, return RSC_ROLE_UNKNOWN_S);
CRM_CHECK(role < RSC_ROLE_MAX, return RSC_ROLE_UNKNOWN_S);
// coverity[dead_error_line]
return RSC_ROLE_UNKNOWN_S;
}
enum rsc_role_e
text2role(const char *role)
{
CRM_ASSERT(role != NULL);
if (safe_str_eq(role, RSC_ROLE_STOPPED_S)) {
return RSC_ROLE_STOPPED;
} else if (safe_str_eq(role, RSC_ROLE_STARTED_S)) {
return RSC_ROLE_STARTED;
} else if (safe_str_eq(role, RSC_ROLE_SLAVE_S)) {
return RSC_ROLE_SLAVE;
} else if (safe_str_eq(role, RSC_ROLE_MASTER_S)) {
return RSC_ROLE_MASTER;
} else if (safe_str_eq(role, RSC_ROLE_UNKNOWN_S)) {
return RSC_ROLE_UNKNOWN;
}
crm_err("Unknown role: %s", role);
return RSC_ROLE_UNKNOWN;
}
int
merge_weights(int w1, int w2)
{
int result = w1 + w2;
if (w1 <= -INFINITY || w2 <= -INFINITY) {
if (w1 >= INFINITY || w2 >= INFINITY) {
crm_trace("-INFINITY + INFINITY == -INFINITY");
}
return -INFINITY;
} else if (w1 >= INFINITY || w2 >= INFINITY) {
return INFINITY;
}
/* detect wrap-around */
if (result > 0) {
if (w1 <= 0 && w2 < 0) {
result = -INFINITY;
}
} else if (w1 > 0 && w2 > 0) {
result = INFINITY;
}
/* detect +/- INFINITY */
if (result >= INFINITY) {
result = INFINITY;
} else if (result <= -INFINITY) {
result = -INFINITY;
}
crm_trace("%d + %d = %d", w1, w2, result);
return result;
}
void
add_hash_param(GHashTable * hash, const char *name, const char *value)
{
CRM_CHECK(hash != NULL, return);
crm_trace("adding: name=%s value=%s", crm_str(name), crm_str(value));
if (name == NULL || value == NULL) {
return;
} else if (safe_str_eq(value, "#default")) {
return;
} else if (g_hash_table_lookup(hash, name) == NULL) {
g_hash_table_insert(hash, strdup(name), strdup(value));
}
}
const char *
pe_node_attribute_calculated(const pe_node_t *node, const char *name,
const resource_t *rsc)
{
const char *source;
if(node == NULL) {
return NULL;
} else if(rsc == NULL) {
return g_hash_table_lookup(node->details->attrs, name);
}
source = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET);
if(source == NULL || safe_str_eq("host", source) == FALSE) {
return g_hash_table_lookup(node->details->attrs, name);
}
/* Use attributes set for the containers location
* instead of for the container itself
*
* Useful when the container is using the host's local
* storage
*/
CRM_ASSERT(node->details->remote_rsc);
CRM_ASSERT(node->details->remote_rsc->container);
if(node->details->remote_rsc->container->running_on) {
pe_node_t *host = node->details->remote_rsc->container->running_on->data;
pe_rsc_trace(rsc, "%s: Looking for %s on the container host %s", rsc->id, name, host->details->uname);
return g_hash_table_lookup(host->details->attrs, name);
}
pe_rsc_trace(rsc, "%s: Not looking for %s on the container host: %s is inactive",
rsc->id, name, node->details->remote_rsc->container->id);
return NULL;
}
const char *
pe_node_attribute_raw(pe_node_t *node, const char *name)
{
if(node == NULL) {
return NULL;
}
return g_hash_table_lookup(node->details->attrs, name);
}
diff --git a/lib/pengine/native.c b/lib/pengine/native.c
index b0641154dc..5a6fd807e4 100644
--- a/lib/pengine/native.c
+++ b/lib/pengine/native.c
@@ -1,1383 +1,1389 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
#include
#include
#include
#define VARIANT_NATIVE 1
#include "./variant.h"
/*!
* \internal
* \brief Check whether a resource is active on multiple nodes
*/
static bool
is_multiply_active(pe_resource_t *rsc)
{
unsigned int count = 0;
if (rsc->variant == pe_native) {
pe__find_active_requires(rsc, &count);
}
return count > 1;
}
void
native_add_running(resource_t * rsc, node_t * node, pe_working_set_t * data_set)
{
GListPtr gIter = rsc->running_on;
CRM_CHECK(node != NULL, return);
for (; gIter != NULL; gIter = gIter->next) {
node_t *a_node = (node_t *) gIter->data;
CRM_CHECK(a_node != NULL, return);
if (safe_str_eq(a_node->details->id, node->details->id)) {
return;
}
}
pe_rsc_trace(rsc, "Adding %s to %s %s", rsc->id, node->details->uname,
is_set(rsc->flags, pe_rsc_managed)?"":"(unmanaged)");
rsc->running_on = g_list_append(rsc->running_on, node);
if (rsc->variant == pe_native) {
node->details->running_rsc = g_list_append(node->details->running_rsc, rsc);
}
if (rsc->variant == pe_native && node->details->maintenance) {
clear_bit(rsc->flags, pe_rsc_managed);
}
if (is_not_set(rsc->flags, pe_rsc_managed)) {
resource_t *p = rsc->parent;
pe_rsc_info(rsc, "resource %s isn't managed", rsc->id);
resource_location(rsc, node, INFINITY, "not_managed_default", data_set);
while(p && node->details->online) {
/* add without the additional location constraint */
p->running_on = g_list_append(p->running_on, node);
p = p->parent;
}
return;
}
if (is_multiply_active(rsc)) {
switch (rsc->recovery_type) {
case recovery_stop_only:
{
GHashTableIter gIter;
node_t *local_node = NULL;
/* make sure it doesn't come up again */
if (rsc->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->allowed_nodes);
}
rsc->allowed_nodes = node_hash_from_list(data_set->nodes);
g_hash_table_iter_init(&gIter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) {
local_node->weight = -INFINITY;
}
}
break;
case recovery_stop_start:
break;
case recovery_block:
clear_bit(rsc->flags, pe_rsc_managed);
set_bit(rsc->flags, pe_rsc_block);
/* If the resource belongs to a group or bundle configured with
* multiple-active=block, block the entire entity.
*/
if (rsc->parent
&& (rsc->parent->variant == pe_group || rsc->parent->variant == pe_container)
&& rsc->parent->recovery_type == recovery_block) {
GListPtr gIter = rsc->parent->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
clear_bit(child->flags, pe_rsc_managed);
set_bit(child->flags, pe_rsc_block);
}
}
break;
}
crm_debug("%s is active on multiple nodes including %s: %s",
rsc->id, node->details->uname,
recovery2text(rsc->recovery_type));
} else {
pe_rsc_trace(rsc, "Resource %s is active on: %s", rsc->id, node->details->uname);
}
if (rsc->parent != NULL) {
native_add_running(rsc->parent, node, data_set);
}
}
static void
recursive_clear_unique(pe_resource_t *rsc)
{
clear_bit(rsc->flags, pe_rsc_unique);
add_hash_param(rsc->meta, XML_RSC_ATTR_UNIQUE, XML_BOOLEAN_FALSE);
for (GList *child = rsc->children; child != NULL; child = child->next) {
recursive_clear_unique((pe_resource_t *) child->data);
}
}
gboolean
native_unpack(resource_t * rsc, pe_working_set_t * data_set)
{
resource_t *parent = uber_parent(rsc);
native_variant_data_t *native_data = NULL;
const char *standard = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
uint32_t ra_caps = pcmk_get_ra_caps(standard);
pe_rsc_trace(rsc, "Processing resource %s...", rsc->id);
native_data = calloc(1, sizeof(native_variant_data_t));
rsc->variant_opaque = native_data;
// Only some agent standards support unique and promotable clones
if (is_not_set(ra_caps, pcmk_ra_cap_unique)
&& is_set(rsc->flags, pe_rsc_unique) && pe_rsc_is_clone(parent)) {
/* @COMPAT We should probably reject this situation as an error (as we
* do for promotable below) rather than warn and convert, but that would
* be a backward-incompatible change that we should probably do with a
* transform at a schema major version bump.
*/
pe__force_anon(standard, parent, rsc->id, data_set);
/* Clear globally-unique on the parent and all its descendents unpacked
* so far (clearing the parent should make any future children unpacking
* correct). We have to clear this resource explicitly because it isn't
* hooked into the parent's children yet.
*/
recursive_clear_unique(parent);
recursive_clear_unique(rsc);
}
if (is_not_set(ra_caps, pcmk_ra_cap_promotable)
&& is_set(parent->flags, pe_rsc_promotable)) {
pe_err("Resource %s is of type %s and therefore "
"cannot be used as a promotable clone resource",
rsc->id, standard);
return FALSE;
}
return TRUE;
}
static bool
rsc_is_on_node(resource_t *rsc, const node_t *node, int flags)
{
pe_rsc_trace(rsc, "Checking whether %s is on %s",
rsc->id, node->details->uname);
if (is_set(flags, pe_find_current) && rsc->running_on) {
for (GListPtr iter = rsc->running_on; iter; iter = iter->next) {
node_t *loc = (node_t *) iter->data;
if (loc->details == node->details) {
return TRUE;
}
}
} else if (is_set(flags, pe_find_inactive) && (rsc->running_on == NULL)) {
return TRUE;
} else if (is_not_set(flags, pe_find_current) && rsc->allocated_to
&& (rsc->allocated_to->details == node->details)) {
return TRUE;
}
return FALSE;
}
resource_t *
native_find_rsc(resource_t * rsc, const char *id, const node_t *on_node,
int flags)
{
bool match = FALSE;
resource_t *result = NULL;
CRM_CHECK(id && rsc && rsc->id, return NULL);
if (flags & pe_find_clone) {
const char *rid = ID(rsc->xml);
if (!pe_rsc_is_clone(uber_parent(rsc))) {
match = FALSE;
} else if (!strcmp(id, rsc->id) || safe_str_eq(id, rid)) {
match = TRUE;
}
} else if (!strcmp(id, rsc->id)) {
match = TRUE;
} else if (is_set(flags, pe_find_renamed)
&& rsc->clone_name && strcmp(rsc->clone_name, id) == 0) {
match = TRUE;
} else if (is_set(flags, pe_find_any)
|| (is_set(flags, pe_find_anon)
&& is_not_set(rsc->flags, pe_rsc_unique))) {
match = pe_base_name_eq(rsc, id);
}
if (match && on_node) {
bool match_node = rsc_is_on_node(rsc, on_node, flags);
if (match_node == FALSE) {
match = FALSE;
}
}
if (match) {
return rsc;
}
for (GListPtr gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
result = rsc->fns->find_rsc(child, id, on_node, flags);
if (result) {
return result;
}
}
return NULL;
}
char *
native_parameter(resource_t * rsc, node_t * node, gboolean create, const char *name,
pe_working_set_t * data_set)
{
char *value_copy = NULL;
const char *value = NULL;
GHashTable *hash = NULL;
GHashTable *local_hash = NULL;
CRM_CHECK(rsc != NULL, return NULL);
CRM_CHECK(name != NULL && strlen(name) != 0, return NULL);
pe_rsc_trace(rsc, "Looking up %s in %s", name, rsc->id);
if (create || g_hash_table_size(rsc->parameters) == 0) {
if (node != NULL) {
pe_rsc_trace(rsc, "Creating hash with node %s", node->details->uname);
} else {
pe_rsc_trace(rsc, "Creating default hash");
}
local_hash = crm_str_table_new();
get_rsc_attributes(local_hash, rsc, node, data_set);
hash = local_hash;
} else {
hash = rsc->parameters;
}
value = g_hash_table_lookup(hash, name);
if (value == NULL) {
/* try meta attributes instead */
value = g_hash_table_lookup(rsc->meta, name);
}
if (value != NULL) {
value_copy = strdup(value);
}
if (local_hash != NULL) {
g_hash_table_destroy(local_hash);
}
return value_copy;
}
gboolean
native_active(resource_t * rsc, gboolean all)
{
GListPtr gIter = rsc->running_on;
for (; gIter != NULL; gIter = gIter->next) {
node_t *a_node = (node_t *) gIter->data;
if (a_node->details->unclean) {
crm_debug("Resource %s: node %s is unclean", rsc->id, a_node->details->uname);
return TRUE;
} else if (a_node->details->online == FALSE) {
crm_debug("Resource %s: node %s is offline", rsc->id, a_node->details->uname);
} else {
crm_debug("Resource %s active on %s", rsc->id, a_node->details->uname);
return TRUE;
}
}
return FALSE;
}
struct print_data_s {
long options;
void *print_data;
};
static void
native_print_attr(gpointer key, gpointer value, gpointer user_data)
{
long options = ((struct print_data_s *)user_data)->options;
void *print_data = ((struct print_data_s *)user_data)->print_data;
status_print("Option: %s = %s\n", (char *)key, (char *)value);
}
static const char *
native_pending_state(resource_t * rsc)
{
const char *pending_state = NULL;
if (safe_str_eq(rsc->pending_task, CRMD_ACTION_START)) {
pending_state = "Starting";
} else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_STOP)) {
pending_state = "Stopping";
} else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_MIGRATE)) {
pending_state = "Migrating";
} else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_MIGRATED)) {
/* Work might be done in here. */
pending_state = "Migrating";
} else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_PROMOTE)) {
pending_state = "Promoting";
} else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_DEMOTE)) {
pending_state = "Demoting";
}
return pending_state;
}
static const char *
native_pending_task(resource_t * rsc)
{
const char *pending_task = NULL;
if (safe_str_eq(rsc->pending_task, CRMD_ACTION_STATUS)) {
pending_task = "Monitoring";
/* Pending probes are not printed, even if pending
* operations are requested. If someone ever requests that
* behavior, uncomment this and the corresponding part of
* unpack.c:unpack_rsc_op().
*/
/*
} else if (safe_str_eq(rsc->pending_task, "probe")) {
pending_task = "Checking";
*/
}
return pending_task;
}
static enum rsc_role_e
native_displayable_role(resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
if ((role == RSC_ROLE_STARTED)
&& is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) {
role = RSC_ROLE_SLAVE;
}
return role;
}
static const char *
native_displayable_state(resource_t *rsc, long options)
{
const char *rsc_state = NULL;
if (options & pe_print_pending) {
rsc_state = native_pending_state(rsc);
}
if (rsc_state == NULL) {
rsc_state = role2text(native_displayable_role(rsc));
}
return rsc_state;
}
static void
native_print_xml(resource_t * rsc, const char *pre_text, long options, void *print_data)
{
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
const char *rsc_state = native_displayable_state(rsc, options);
const char *target_role = NULL;
/* resource information. */
status_print("%sxml, XML_ATTR_TYPE));
status_print("role=\"%s\" ", rsc_state);
if (rsc->meta) {
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (target_role) {
status_print("target_role=\"%s\" ", target_role);
}
status_print("active=\"%s\" ", rsc->fns->active(rsc, TRUE) ? "true" : "false");
status_print("orphaned=\"%s\" ", is_set(rsc->flags, pe_rsc_orphan) ? "true" : "false");
status_print("blocked=\"%s\" ", is_set(rsc->flags, pe_rsc_block) ? "true" : "false");
status_print("managed=\"%s\" ", is_set(rsc->flags, pe_rsc_managed) ? "true" : "false");
status_print("failed=\"%s\" ", is_set(rsc->flags, pe_rsc_failed) ? "true" : "false");
status_print("failure_ignored=\"%s\" ",
is_set(rsc->flags, pe_rsc_failure_ignored) ? "true" : "false");
status_print("nodes_running_on=\"%d\" ", g_list_length(rsc->running_on));
if (options & pe_print_pending) {
const char *pending_task = native_pending_task(rsc);
if (pending_task) {
status_print("pending=\"%s\" ", pending_task);
}
}
if (options & pe_print_dev) {
status_print("provisional=\"%s\" ",
is_set(rsc->flags, pe_rsc_provisional) ? "true" : "false");
status_print("runnable=\"%s\" ", is_set(rsc->flags, pe_rsc_runnable) ? "true" : "false");
status_print("priority=\"%f\" ", (double)rsc->priority);
status_print("variant=\"%s\" ", crm_element_name(rsc->xml));
}
/* print out the nodes this resource is running on */
if (options & pe_print_rsconly) {
status_print("/>\n");
/* do nothing */
} else if (rsc->running_on != NULL) {
GListPtr gIter = rsc->running_on;
status_print(">\n");
for (; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
status_print("%s \n", pre_text,
node->details->uname, node->details->id,
node->details->online ? "false" : "true");
}
status_print("%s\n", pre_text);
} else {
status_print("/>\n");
}
}
// Append a flag to resource description string's flags list
static bool
add_output_flag(GString *s, const char *flag_desc, bool have_flags)
{
g_string_append(s, (have_flags? ", " : " ("));
g_string_append(s, flag_desc);
return true;
}
// Append a node name to resource description string's node list
static bool
add_output_node(GString *s, const char *node, bool have_nodes)
{
g_string_append(s, (have_nodes? " " : " [ "));
g_string_append(s, node);
return true;
}
/*!
* \internal
* \brief Create a string description of a resource
*
* \param[in] rsc Resource to describe
* \param[in] name Desired identifier for the resource
* \param[in] node If not NULL, node that resource is "on"
* \param[in] options Bitmask of pe_print_*
* \param[in] target_role Resource's target role
* \param[in] show_nodes Whether to display nodes when multiply active
*
* \return Newly allocated string description of resource
* \note Caller must free the result with g_free().
*/
static gchar *
native_output_string(pe_resource_t *rsc, const char *name, pe_node_t *node,
long options, const char *target_role, bool show_nodes)
{
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *provider = NULL;
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
char *retval = NULL;
GString *outstr = NULL;
bool have_flags = false;
CRM_CHECK(name != NULL, name = "unknown");
CRM_CHECK(kind != NULL, kind = "unknown");
CRM_CHECK(class != NULL, class = "unknown");
if (is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) {
provider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
}
+ if ((node == NULL) && (rsc->lock_node != NULL)) {
+ node = rsc->lock_node;
+ }
if (is_set(options, pe_print_rsconly)
|| pcmk__list_of_multiple(rsc->running_on)) {
node = NULL;
}
// We need a string of at least this size
outstr = g_string_sized_new(strlen(name) + strlen(class) + strlen(kind)
+ (provider? (strlen(provider) + 2) : 0)
+ (node? strlen(node->details->uname) + 1 : 0)
+ 11);
// Resource name and agent
g_string_printf(outstr, "%s\t(%s%s%s:%s):\t", name, class,
/* @COMPAT This should be a single ':' (see CLBZ#5395) but
* to avoid breaking anything relying on it, we're keeping
* it like this until the next minor version bump.
*/
(provider? "::" : ""), (provider? provider : ""), kind);
// State on node
if (is_set(rsc->flags, pe_rsc_orphan)) {
g_string_append(outstr, " ORPHANED");
}
if (is_set(rsc->flags, pe_rsc_failed)) {
enum rsc_role_e role = native_displayable_role(rsc);
if (role > RSC_ROLE_SLAVE) {
g_string_append_printf(outstr, " FAILED %s", role2text(role));
} else {
g_string_append(outstr, " FAILED");
}
} else {
g_string_append(outstr, native_displayable_state(rsc, options));
}
if (node) {
g_string_append_printf(outstr, " %s", node->details->uname);
}
// Flags, as: ( [...])
if (node && !(node->details->online) && node->details->unclean) {
have_flags = add_output_flag(outstr, "UNCLEAN", have_flags);
}
+ if (node && (node == rsc->lock_node)) {
+ have_flags = add_output_flag(outstr, "LOCKED", have_flags);
+ }
if (is_set(options, pe_print_pending)) {
const char *pending_task = native_pending_task(rsc);
if (pending_task) {
have_flags = add_output_flag(outstr, pending_task, have_flags);
}
}
if (target_role) {
enum rsc_role_e target_role_e = text2role(target_role);
/* Only show target role if it limits our abilities (i.e. ignore
* Started, as it is the default anyways, and doesn't prevent the
* resource from becoming Master).
*/
if (target_role_e == RSC_ROLE_STOPPED) {
have_flags = add_output_flag(outstr, "disabled", have_flags);
} else if (is_set(uber_parent(rsc)->flags, pe_rsc_promotable)
&& target_role_e == RSC_ROLE_SLAVE) {
have_flags = add_output_flag(outstr, "target-role:", have_flags);
g_string_append(outstr, target_role);
}
}
if (is_set(rsc->flags, pe_rsc_block)) {
have_flags = add_output_flag(outstr, "blocked", have_flags);
} else if (is_not_set(rsc->flags, pe_rsc_managed)) {
have_flags = add_output_flag(outstr, "unmanaged", have_flags);
}
if (is_set(rsc->flags, pe_rsc_failure_ignored)) {
have_flags = add_output_flag(outstr, "failure ignored", have_flags);
}
if (is_set(options, pe_print_dev)) {
if (is_set(options, pe_rsc_provisional)) {
have_flags = add_output_flag(outstr, "provisional", have_flags);
}
if (is_not_set(options, pe_rsc_runnable)) {
have_flags = add_output_flag(outstr, "non-startable", have_flags);
}
have_flags = add_output_flag(outstr, "variant:", have_flags);
g_string_append_printf(outstr, "%s priority:%f",
crm_element_name(rsc->xml),
(double) (rsc->priority));
}
if (have_flags) {
g_string_append(outstr, ")");
}
// User-supplied description
if (is_set(options, pe_print_rsconly)
|| pcmk__list_of_multiple(rsc->running_on)) {
const char *desc = crm_element_value(rsc->xml, XML_ATTR_DESC);
if (desc) {
g_string_append_printf(outstr, " %s", desc);
}
}
if (show_nodes && is_not_set(options, pe_print_rsconly)
&& pcmk__list_of_multiple(rsc->running_on)) {
bool have_nodes = false;
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *n = (pe_node_t *) iter->data;
have_nodes = add_output_node(outstr, n->details->uname, have_nodes);
}
if (have_nodes) {
g_string_append(outstr, " ]");
}
}
retval = outstr->str;
g_string_free(outstr, FALSE);
return retval;
}
void
pe__common_output_html(pcmk__output_t *out, resource_t * rsc,
const char *name, node_t *node, long options)
{
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
const char *target_role = NULL;
xmlNodePtr list_node = NULL;
const char *cl = NULL;
CRM_ASSERT(rsc->variant == pe_native);
CRM_ASSERT(kind != NULL);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal) && is_not_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (is_not_set(rsc->flags, pe_rsc_managed)) {
cl = "rsc-managed";
} else if (is_set(rsc->flags, pe_rsc_failed)) {
cl = "rsc-failed";
} else if (rsc->variant == pe_native && (rsc->running_on == NULL)) {
cl = "rsc-failed";
} else if (pcmk__list_of_multiple(rsc->running_on)) {
cl = "rsc-multiple";
} else if (is_set(rsc->flags, pe_rsc_failure_ignored)) {
cl = "rsc-failure-ignored";
} else {
cl = "rsc-ok";
}
{
gchar *s = native_output_string(rsc, name, node, options, target_role,
true);
list_node = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL);
pcmk_create_html_node(list_node, "span", NULL, cl, s);
g_free(s);
}
if (is_set(options, pe_print_details)) {
GHashTableIter iter;
gpointer key, value;
out->begin_list(out, NULL, NULL, "Options");
g_hash_table_iter_init(&iter, rsc->parameters);
while (g_hash_table_iter_next(&iter, &key, &value)) {
out->list_item(out, NULL, "Option: %s = %s", (char *) key, (char *) value);
}
out->end_list(out);
}
if (is_set(options, pe_print_dev)) {
GHashTableIter iter;
node_t *n = NULL;
out->begin_list(out, NULL, NULL, "Allowed Nodes");
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
out->list_item(out, NULL, "%s %d", n->details->uname, n->weight);
}
out->end_list(out);
}
if (is_set(options, pe_print_max_details)) {
GHashTableIter iter;
node_t *n = NULL;
out->begin_list(out, NULL, NULL, "=== Allowed Nodes");
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
pe__output_node(n, FALSE, out);
}
out->end_list(out);
}
}
void
pe__common_output_text(pcmk__output_t *out, resource_t * rsc,
const char *name, node_t *node, long options)
{
const char *target_role = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal) && is_not_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
{
gchar *s = native_output_string(rsc, name, node, options, target_role,
true);
out->list_item(out, NULL, "%s", s);
g_free(s);
}
if (is_set(options, pe_print_details)) {
GHashTableIter iter;
gpointer key, value;
out->begin_list(out, NULL, NULL, "Options");
g_hash_table_iter_init(&iter, rsc->parameters);
while (g_hash_table_iter_next(&iter, &key, &value)) {
out->list_item(out, NULL, "Option: %s = %s", (char *) key, (char *) value);
}
out->end_list(out);
}
if (is_set(options, pe_print_dev)) {
GHashTableIter iter;
node_t *n = NULL;
out->begin_list(out, NULL, NULL, "Allowed Nodes");
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
out->list_item(out, NULL, "%s %d", n->details->uname, n->weight);
}
out->end_list(out);
}
if (is_set(options, pe_print_max_details)) {
GHashTableIter iter;
node_t *n = NULL;
out->begin_list(out, NULL, NULL, "=== Allowed Nodes");
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
pe__output_node(n, FALSE, out);
}
out->end_list(out);
}
}
void
common_print(resource_t * rsc, const char *pre_text, const char *name, node_t *node, long options, void *print_data)
{
const char *target_role = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (rsc->meta) {
const char *is_internal = g_hash_table_lookup(rsc->meta,
XML_RSC_ATTR_INTERNAL_RSC);
if (crm_is_true(is_internal) && is_not_set(options, pe_print_implicit)) {
crm_trace("skipping print of internal resource %s", rsc->id);
return;
}
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
}
if (options & pe_print_xml) {
native_print_xml(rsc, pre_text, options, print_data);
return;
}
if ((pre_text == NULL) && (options & pe_print_printf)) {
pre_text = " ";
}
if (options & pe_print_html) {
if (is_not_set(rsc->flags, pe_rsc_managed)) {
status_print("");
} else if (is_set(rsc->flags, pe_rsc_failed)) {
status_print("");
} else if (rsc->running_on == NULL) {
status_print("");
} else if (pcmk__list_of_multiple(rsc->running_on)) {
status_print("");
} else if (is_set(rsc->flags, pe_rsc_failure_ignored)) {
status_print("");
} else {
status_print("");
}
}
{
gchar *resource_s = native_output_string(rsc, name, node, options,
target_role, false);
status_print("%s%s", (pre_text? pre_text : ""), resource_s);
g_free(resource_s);
}
#if CURSES_ENABLED
if (is_set(options, pe_print_ncurses)
&& is_not_set(options, pe_print_rsconly)
&& !pcmk__list_of_multiple(rsc->running_on)) {
/* coverity[negative_returns] False positive */
move(-1, 0);
}
#endif
if (is_set(options, pe_print_html)) {
status_print(" ");
}
if (is_not_set(options, pe_print_rsconly)
&& pcmk__list_of_multiple(rsc->running_on)) {
GListPtr gIter = rsc->running_on;
int counter = 0;
if (options & pe_print_html) {
status_print("\n");
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print("[");
}
for (; gIter != NULL; gIter = gIter->next) {
node_t *n = (node_t *) gIter->data;
counter++;
if (options & pe_print_html) {
status_print("- \n%s", n->details->uname);
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print(" %s", n->details->uname);
} else if ((options & pe_print_log)) {
status_print("\t%d : %s", counter, n->details->uname);
} else {
status_print("%s", n->details->uname);
}
if (options & pe_print_html) {
status_print("
\n");
}
}
if (options & pe_print_html) {
status_print("
\n");
} else if ((options & pe_print_printf)
|| (options & pe_print_ncurses)) {
status_print(" ]");
}
}
if (options & pe_print_html) {
status_print("
\n");
} else if (options & pe_print_suppres_nl) {
/* nothing */
} else if ((options & pe_print_printf) || (options & pe_print_ncurses)) {
status_print("\n");
}
if (options & pe_print_details) {
struct print_data_s pdata;
pdata.options = options;
pdata.print_data = print_data;
g_hash_table_foreach(rsc->parameters, native_print_attr, &pdata);
}
if (options & pe_print_dev) {
GHashTableIter iter;
node_t *n = NULL;
status_print("%s\tAllowed Nodes", pre_text);
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
status_print("%s\t * %s %d", pre_text, n->details->uname, n->weight);
}
}
if (options & pe_print_max_details) {
GHashTableIter iter;
node_t *n = NULL;
status_print("%s\t=== Allowed Nodes\n", pre_text);
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) {
print_node("\t", n, FALSE);
}
}
}
void
native_print(resource_t * rsc, const char *pre_text, long options, void *print_data)
{
node_t *node = NULL;
CRM_ASSERT(rsc->variant == pe_native);
if (options & pe_print_xml) {
native_print_xml(rsc, pre_text, options, print_data);
return;
}
node = pe__current_node(rsc);
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
common_print(rsc, pre_text, rsc_printable_id(rsc), node, options, print_data);
}
int
pe__resource_xml(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
const char *rsc_state = native_displayable_state(rsc, options);
long is_print_pending = options & pe_print_pending;
long is_print_dev = options & pe_print_dev;
char ra_name[LINE_MAX];
char *nodes_running_on = NULL;
char *priority = NULL;
int rc = 0;
CRM_ASSERT(rsc->variant == pe_native);
/* resource information. */
sprintf(ra_name, "%s%s%s:%s", class, prov ? "::" : "", prov ? prov : ""
, crm_element_value(rsc->xml, XML_ATTR_TYPE));
nodes_running_on = crm_itoa(g_list_length(rsc->running_on));
priority = crm_ftoa(rsc->priority);
rc = pe__name_and_nvpairs_xml(out, true, "resource", 16
, "id", rsc_printable_id(rsc)
, "resource_agent", ra_name
, "role", rsc_state
, "target_role", (rsc->meta ? g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE) : NULL)
, "active", BOOL2STR(rsc->fns->active(rsc, TRUE))
, "orphaned", BOOL2STR(is_set(rsc->flags, pe_rsc_orphan))
, "blocked", BOOL2STR(is_set(rsc->flags, pe_rsc_block))
, "managed", BOOL2STR(is_set(rsc->flags, pe_rsc_managed))
, "failed", BOOL2STR(is_set(rsc->flags, pe_rsc_failed))
, "failure_ignored", BOOL2STR(is_set(rsc->flags, pe_rsc_failure_ignored))
, "nodes_running_on", nodes_running_on
, "pending", (is_print_pending ? native_pending_task(rsc) : NULL)
, "provisional", (is_print_dev ? BOOL2STR(is_set(rsc->flags, pe_rsc_provisional)) : NULL)
, "runnable", (is_print_dev ? BOOL2STR(is_set(rsc->flags, pe_rsc_runnable)) : NULL)
, "priority", (is_print_dev ? priority : NULL)
, "variant", (is_print_dev ? crm_element_name(rsc->xml) : NULL));
free(priority);
free(nodes_running_on);
CRM_ASSERT(rc == 0);
if (rsc->running_on != NULL) {
GListPtr gIter = rsc->running_on;
for (; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
rc = pe__name_and_nvpairs_xml(out, false, "node", 3
, "name", node->details->uname
, "id", node->details->id
, "cached", BOOL2STR(node->details->online));
CRM_ASSERT(rc == 0);
}
}
pcmk__output_xml_pop_parent(out);
return rc;
}
int
pe__resource_html(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
node_t *node = pe__current_node(rsc);
CRM_ASSERT(rsc->variant == pe_native);
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, options);
return 0;
}
int
pe__resource_text(pcmk__output_t *out, va_list args)
{
unsigned int options = va_arg(args, unsigned int);
pe_resource_t *rsc = va_arg(args, pe_resource_t *);
node_t *node = pe__current_node(rsc);
CRM_ASSERT(rsc->variant == pe_native);
if (node == NULL) {
// This is set only if a non-probe action is pending on this node
node = rsc->pending_node;
}
pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, options);
return 0;
}
void
native_free(resource_t * rsc)
{
pe_rsc_trace(rsc, "Freeing resource action list (not the data)");
common_free(rsc);
}
enum rsc_role_e
native_resource_state(const resource_t * rsc, gboolean current)
{
enum rsc_role_e role = rsc->next_role;
if (current) {
role = rsc->role;
}
pe_rsc_trace(rsc, "%s state: %s", rsc->id, role2text(role));
return role;
}
/*!
* \internal
* \brief List nodes where a resource (or any of its children) is
*
* \param[in] rsc Resource to check
* \param[out] list List to add result to
* \param[in] current 0 = where known, 1 = running, 2 = running or pending
*
* \return If list contains only one node, that node
*/
pe_node_t *
native_location(const pe_resource_t *rsc, GList **list, int current)
{
node_t *one = NULL;
GListPtr result = NULL;
if (rsc->children) {
GListPtr gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
child->fns->location(child, &result, current);
}
} else if (current) {
if (rsc->running_on) {
result = g_list_copy(rsc->running_on);
}
if ((current == 2) && rsc->pending_node
&& !pe_find_node_id(result, rsc->pending_node->details->id)) {
result = g_list_append(result, rsc->pending_node);
}
} else if (current == FALSE && rsc->allocated_to) {
result = g_list_append(NULL, rsc->allocated_to);
}
if (result && (result->next == NULL)) {
one = result->data;
}
if (list) {
GListPtr gIter = result;
for (; gIter != NULL; gIter = gIter->next) {
node_t *node = (node_t *) gIter->data;
if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) {
*list = g_list_append(*list, node);
}
}
}
g_list_free(result);
return one;
}
static void
get_rscs_brief(GListPtr rsc_list, GHashTable * rsc_table, GHashTable * active_table)
{
GListPtr gIter = rsc_list;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *rsc = (resource_t *) gIter->data;
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
int offset = 0;
char buffer[LINE_MAX];
int *rsc_counter = NULL;
int *active_counter = NULL;
if (rsc->variant != pe_native) {
continue;
}
offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class);
if (is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) {
const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
offset += snprintf(buffer + offset, LINE_MAX - offset, "::%s", prov);
}
offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind);
CRM_LOG_ASSERT(offset > 0);
if (rsc_table) {
rsc_counter = g_hash_table_lookup(rsc_table, buffer);
if (rsc_counter == NULL) {
rsc_counter = calloc(1, sizeof(int));
*rsc_counter = 0;
g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter);
}
(*rsc_counter)++;
}
if (active_table) {
GListPtr gIter2 = rsc->running_on;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
node_t *node = (node_t *) gIter2->data;
GHashTable *node_table = NULL;
if (node->details->unclean == FALSE && node->details->online == FALSE) {
continue;
}
node_table = g_hash_table_lookup(active_table, node->details->uname);
if (node_table == NULL) {
node_table = crm_str_table_new();
g_hash_table_insert(active_table, strdup(node->details->uname), node_table);
}
active_counter = g_hash_table_lookup(node_table, buffer);
if (active_counter == NULL) {
active_counter = calloc(1, sizeof(int));
*active_counter = 0;
g_hash_table_insert(node_table, strdup(buffer), active_counter);
}
(*active_counter)++;
}
}
}
}
static void
destroy_node_table(gpointer data)
{
GHashTable *node_table = data;
if (node_table) {
g_hash_table_destroy(node_table);
}
}
void
print_rscs_brief(GListPtr rsc_list, const char *pre_text, long options,
void *print_data, gboolean print_all)
{
GHashTable *rsc_table = crm_str_table_new();
GHashTable *active_table = g_hash_table_new_full(crm_str_hash, g_str_equal,
free, destroy_node_table);
GHashTableIter hash_iter;
char *type = NULL;
int *rsc_counter = NULL;
get_rscs_brief(rsc_list, rsc_table, active_table);
g_hash_table_iter_init(&hash_iter, rsc_table);
while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) {
GHashTableIter hash_iter2;
char *node_name = NULL;
GHashTable *node_table = NULL;
int active_counter_all = 0;
g_hash_table_iter_init(&hash_iter2, active_table);
while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) {
int *active_counter = g_hash_table_lookup(node_table, type);
if (active_counter == NULL || *active_counter == 0) {
continue;
} else {
active_counter_all += *active_counter;
}
if (options & pe_print_rsconly) {
node_name = NULL;
}
if (options & pe_print_html) {
status_print("\n");
}
if (print_all) {
status_print("%s%d/%d\t(%s):\tActive %s\n", pre_text ? pre_text : "",
active_counter ? *active_counter : 0,
rsc_counter ? *rsc_counter : 0, type,
active_counter && (*active_counter > 0) && node_name ? node_name : "");
} else {
status_print("%s%d\t(%s):\tActive %s\n", pre_text ? pre_text : "",
active_counter ? *active_counter : 0, type,
active_counter && (*active_counter > 0) && node_name ? node_name : "");
}
if (options & pe_print_html) {
status_print("\n");
}
}
if (print_all && active_counter_all == 0) {
if (options & pe_print_html) {
status_print("\n");
}
status_print("%s%d/%d\t(%s):\tActive\n", pre_text ? pre_text : "",
active_counter_all,
rsc_counter ? *rsc_counter : 0, type);
if (options & pe_print_html) {
status_print("\n");
}
}
}
if (rsc_table) {
g_hash_table_destroy(rsc_table);
rsc_table = NULL;
}
if (active_table) {
g_hash_table_destroy(active_table);
active_table = NULL;
}
}
void
pe__rscs_brief_output(pcmk__output_t *out, GListPtr rsc_list, long options, gboolean print_all)
{
GHashTable *rsc_table = crm_str_table_new();
GHashTable *active_table = g_hash_table_new_full(crm_str_hash, g_str_equal,
free, destroy_node_table);
GHashTableIter hash_iter;
char *type = NULL;
int *rsc_counter = NULL;
get_rscs_brief(rsc_list, rsc_table, active_table);
g_hash_table_iter_init(&hash_iter, rsc_table);
while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) {
GHashTableIter hash_iter2;
char *node_name = NULL;
GHashTable *node_table = NULL;
int active_counter_all = 0;
g_hash_table_iter_init(&hash_iter2, active_table);
while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) {
int *active_counter = g_hash_table_lookup(node_table, type);
if (active_counter == NULL || *active_counter == 0) {
continue;
} else {
active_counter_all += *active_counter;
}
if (options & pe_print_rsconly) {
node_name = NULL;
}
if (print_all) {
out->list_item(out, NULL, " %d/%d\t(%s):\tActive %s",
*active_counter,
rsc_counter ? *rsc_counter : 0, type,
(*active_counter > 0) && node_name ? node_name : "");
} else {
out->list_item(out, NULL, " %d\t(%s):\tActive %s",
*active_counter, type,
(*active_counter > 0) && node_name ? node_name : "");
}
}
if (print_all && active_counter_all == 0) {
out->list_item(out, NULL, " %d/%d\t(%s):\tActive",
active_counter_all,
rsc_counter ? *rsc_counter : 0, type);
}
}
if (rsc_table) {
g_hash_table_destroy(rsc_table);
rsc_table = NULL;
}
if (active_table) {
g_hash_table_destroy(active_table);
active_table = NULL;
}
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index c9fc672518..87edc83b45 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,3842 +1,3892 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include
#include
#include
#include
+#include
#include
#include
#include
#include
#include
#include
#include
#include
#include