diff --git a/cts/cts-scheduler.in b/cts/cts-scheduler.in index c4ec0e9243..8fa16fb69a 100644 --- a/cts/cts-scheduler.in +++ b/cts/cts-scheduler.in @@ -1,1458 +1,1459 @@ #!@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" __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" ], ], # @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/guest-host-not-fenceable.dot b/cts/scheduler/guest-host-not-fenceable.dot new file mode 100644 index 0000000000..98833f6160 --- /dev/null +++ b/cts/scheduler/guest-host-not-fenceable.dot @@ -0,0 +1,262 @@ + digraph "g" { +"Cancel galera-bundle-0_monitor_60000 node1" [ style=bold color="green" fontcolor="black"] +"Cancel rabbitmq-bundle-0_monitor_60000 node1" [ style=bold color="green" fontcolor="black"] +"galera-bundle-0_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-0_start_0 node1" -> "galera-bundle-0_monitor_30000 node1" [ style = dashed] +"galera-bundle-0_start_0 node1" -> "galera_clear_failcount_0 galera-bundle-0" [ style = dashed] +"galera-bundle-0_start_0 node1" -> "galera_monitor_10000 galera-bundle-0" [ style = dashed] +"galera-bundle-0_start_0 node1" -> "galera_start_0 galera-bundle-0" [ style = dashed] +"galera-bundle-0_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-0_stop_0 node1" -> "Cancel galera-bundle-0_monitor_60000 node1" [ style = bold] +"galera-bundle-0_stop_0 node1" -> "galera-bundle-0_start_0 node1" [ style = dashed] +"galera-bundle-0_stop_0 node1" -> "galera-bundle-docker-0_stop_0 node1" [ style = bold] +"galera-bundle-0_stop_0 node1" [ style=bold color="green" fontcolor="black"] +"galera-bundle-1_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-1_start_0 node1" -> "galera-bundle-1_monitor_30000 node1" [ style = dashed] +"galera-bundle-1_start_0 node1" -> "galera_monitor_10000 galera-bundle-1" [ style = dashed] +"galera-bundle-1_start_0 node1" -> "galera_start_0 galera-bundle-1" [ style = dashed] +"galera-bundle-1_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-1_stop_0 node2" -> "galera-bundle-1_start_0 node1" [ style = dashed] +"galera-bundle-1_stop_0 node2" -> "galera-bundle-docker-1_stop_0 node2" [ style = dashed] +"galera-bundle-1_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-2_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-2_start_0 node1" -> "galera-bundle-2_monitor_30000 node1" [ style = dashed] +"galera-bundle-2_start_0 node1" -> "galera_monitor_20000 galera-bundle-2" [ style = dashed] +"galera-bundle-2_start_0 node1" -> "galera_monitor_30000 galera-bundle-2" [ style = dashed] +"galera-bundle-2_start_0 node1" -> "galera_start_0 galera-bundle-2" [ style = dashed] +"galera-bundle-2_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-2_stop_0 node3" -> "galera-bundle-2_start_0 node1" [ style = dashed] +"galera-bundle-2_stop_0 node3" -> "galera-bundle-docker-2_stop_0 node3" [ style = dashed] +"galera-bundle-2_stop_0 node3" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-docker-0_monitor_60000 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-docker-0_start_0 node1" -> "galera-bundle-0_start_0 node1" [ style = dashed] +"galera-bundle-docker-0_start_0 node1" -> "galera-bundle-docker-0_monitor_60000 node1" [ style = dashed] +"galera-bundle-docker-0_start_0 node1" -> "galera-bundle_running_0" [ style = dashed] +"galera-bundle-docker-0_start_0 node1" -> "galera_start_0 galera-bundle-0" [ style = dashed] +"galera-bundle-docker-0_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-docker-0_stop_0 node1" -> "galera-bundle-docker-0_start_0 node1" [ style = dashed] +"galera-bundle-docker-0_stop_0 node1" -> "galera-bundle_stopped_0" [ style = bold] +"galera-bundle-docker-0_stop_0 node1" [ style=bold color="green" fontcolor="black"] +"galera-bundle-docker-1_stop_0 node2" -> "galera-bundle_stopped_0" [ style = dashed] +"galera-bundle-docker-1_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-docker-2_stop_0 node3" -> "galera-bundle_stopped_0" [ style = dashed] +"galera-bundle-docker-2_stop_0 node3" [ style=dashed color="red" fontcolor="black"] +"galera-bundle-master_demote_0" -> "galera-bundle-master_demoted_0" [ style = bold] +"galera-bundle-master_demote_0" -> "galera_demote_0 galera-bundle-0" [ style = bold] +"galera-bundle-master_demote_0" -> "galera_demote_0 galera-bundle-1" [ style = dashed] +"galera-bundle-master_demote_0" -> "galera_demote_0 galera-bundle-2" [ style = dashed] +"galera-bundle-master_demote_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle-master_demoted_0" -> "galera-bundle-master_start_0" [ style = dashed] +"galera-bundle-master_demoted_0" -> "galera-bundle-master_stop_0" [ style = bold] +"galera-bundle-master_demoted_0" -> "galera-bundle_demoted_0" [ style = bold] +"galera-bundle-master_demoted_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle-master_running_0" -> "galera-bundle_running_0" [ style = dashed] +"galera-bundle-master_running_0" [ style=dashed color="red" fontcolor="orange"] +"galera-bundle-master_start_0" -> "galera-bundle-master_running_0" [ style = dashed] +"galera-bundle-master_start_0" -> "galera_start_0 galera-bundle-0" [ style = dashed] +"galera-bundle-master_start_0" -> "galera_start_0 galera-bundle-1" [ style = dashed] +"galera-bundle-master_start_0" -> "galera_start_0 galera-bundle-2" [ style = dashed] +"galera-bundle-master_start_0" [ style=dashed color="red" fontcolor="orange"] +"galera-bundle-master_stop_0" -> "galera-bundle-master_stopped_0" [ style = bold] +"galera-bundle-master_stop_0" -> "galera_stop_0 galera-bundle-0" [ style = bold] +"galera-bundle-master_stop_0" -> "galera_stop_0 galera-bundle-1" [ style = dashed] +"galera-bundle-master_stop_0" -> "galera_stop_0 galera-bundle-2" [ style = dashed] +"galera-bundle-master_stop_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle-master_stopped_0" -> "galera-bundle-master_start_0" [ style = dashed] +"galera-bundle-master_stopped_0" -> "galera-bundle_stopped_0" [ style = bold] +"galera-bundle-master_stopped_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle_demote_0" -> "galera-bundle-master_demote_0" [ style = bold] +"galera-bundle_demote_0" -> "galera-bundle_demoted_0" [ style = bold] +"galera-bundle_demote_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle_demoted_0" -> "galera-bundle_start_0" [ style = dashed] +"galera-bundle_demoted_0" -> "galera-bundle_stop_0" [ style = bold] +"galera-bundle_demoted_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle_running_0" [ style=dashed color="red" fontcolor="orange"] +"galera-bundle_start_0" -> "galera-bundle-docker-0_start_0 node1" [ style = dashed] +"galera-bundle_start_0" -> "galera-bundle-master_start_0" [ style = dashed] +"galera-bundle_start_0" [ style=dashed color="red" fontcolor="orange"] +"galera-bundle_stop_0" -> "galera-bundle-docker-0_stop_0 node1" [ style = bold] +"galera-bundle_stop_0" -> "galera-bundle-docker-1_stop_0 node2" [ style = dashed] +"galera-bundle_stop_0" -> "galera-bundle-docker-2_stop_0 node3" [ style = dashed] +"galera-bundle_stop_0" -> "galera-bundle-master_stop_0" [ style = bold] +"galera-bundle_stop_0" -> "galera_stop_0 galera-bundle-0" [ style = bold] +"galera-bundle_stop_0" -> "galera_stop_0 galera-bundle-1" [ style = dashed] +"galera-bundle_stop_0" -> "galera_stop_0 galera-bundle-2" [ style = dashed] +"galera-bundle_stop_0" [ style=bold color="green" fontcolor="orange"] +"galera-bundle_stopped_0" -> "galera-bundle_start_0" [ style = dashed] +"galera-bundle_stopped_0" [ style=bold color="green" fontcolor="orange"] +"galera_clear_failcount_0 galera-bundle-0" [ style=dashed color="red" fontcolor="black"] +"galera_demote_0 galera-bundle-0" -> "galera-bundle-0_stop_0 node1" [ style = bold] +"galera_demote_0 galera-bundle-0" -> "galera-bundle-master_demoted_0" [ style = bold] +"galera_demote_0 galera-bundle-0" -> "galera_stop_0 galera-bundle-0" [ style = bold] +"galera_demote_0 galera-bundle-0" [ style=bold color="green" fontcolor="black"] +"galera_demote_0 galera-bundle-1" -> "galera-bundle-master_demoted_0" [ style = dashed] +"galera_demote_0 galera-bundle-1" -> "galera_demote_0 galera-bundle-0" [ style = dashed] +"galera_demote_0 galera-bundle-1" -> "galera_stop_0 galera-bundle-1" [ style = dashed] +"galera_demote_0 galera-bundle-1" [ style=dashed color="red" fontcolor="black"] +"galera_demote_0 galera-bundle-2" -> "galera-bundle-master_demoted_0" [ style = dashed] +"galera_demote_0 galera-bundle-2" -> "galera_demote_0 galera-bundle-1" [ style = dashed] +"galera_demote_0 galera-bundle-2" -> "galera_monitor_20000 galera-bundle-2" [ style = dashed] +"galera_demote_0 galera-bundle-2" -> "galera_monitor_30000 galera-bundle-2" [ style = dashed] +"galera_demote_0 galera-bundle-2" -> "galera_stop_0 galera-bundle-2" [ style = dashed] +"galera_demote_0 galera-bundle-2" [ style=dashed color="red" fontcolor="black"] +"galera_monitor_10000 galera-bundle-0" [ style=dashed color="red" fontcolor="black"] +"galera_monitor_10000 galera-bundle-1" [ style=dashed color="red" fontcolor="black"] +"galera_monitor_20000 galera-bundle-2" [ style=dashed color="red" fontcolor="black"] +"galera_monitor_30000 galera-bundle-2" [ style=dashed color="red" fontcolor="black"] +"galera_start_0 galera-bundle-0" -> "galera-bundle-master_running_0" [ style = dashed] +"galera_start_0 galera-bundle-0" -> "galera_monitor_10000 galera-bundle-0" [ style = dashed] +"galera_start_0 galera-bundle-0" -> "galera_start_0 galera-bundle-1" [ style = dashed] +"galera_start_0 galera-bundle-0" [ style=dashed color="red" fontcolor="black"] +"galera_start_0 galera-bundle-1" -> "galera-bundle-master_running_0" [ style = dashed] +"galera_start_0 galera-bundle-1" -> "galera_monitor_10000 galera-bundle-1" [ style = dashed] +"galera_start_0 galera-bundle-1" -> "galera_start_0 galera-bundle-2" [ style = dashed] +"galera_start_0 galera-bundle-1" [ style=dashed color="red" fontcolor="black"] +"galera_start_0 galera-bundle-2" -> "galera-bundle-master_running_0" [ style = dashed] +"galera_start_0 galera-bundle-2" -> "galera_monitor_20000 galera-bundle-2" [ style = dashed] +"galera_start_0 galera-bundle-2" -> "galera_monitor_30000 galera-bundle-2" [ style = dashed] +"galera_start_0 galera-bundle-2" [ style=dashed color="red" fontcolor="black"] +"galera_stop_0 galera-bundle-0" -> "galera-bundle-0_stop_0 node1" [ style = bold] +"galera_stop_0 galera-bundle-0" -> "galera-bundle-master_stopped_0" [ style = bold] +"galera_stop_0 galera-bundle-0" -> "galera_start_0 galera-bundle-0" [ style = dashed] +"galera_stop_0 galera-bundle-0" [ style=bold color="green" fontcolor="black"] +"galera_stop_0 galera-bundle-1" -> "galera-bundle-master_stopped_0" [ style = dashed] +"galera_stop_0 galera-bundle-1" -> "galera_start_0 galera-bundle-1" [ style = dashed] +"galera_stop_0 galera-bundle-1" -> "galera_stop_0 galera-bundle-0" [ style = dashed] +"galera_stop_0 galera-bundle-1" [ style=dashed color="red" fontcolor="black"] +"galera_stop_0 galera-bundle-2" -> "galera-bundle-master_stopped_0" [ style = dashed] +"galera_stop_0 galera-bundle-2" -> "galera_start_0 galera-bundle-2" [ style = dashed] +"galera_stop_0 galera-bundle-2" -> "galera_stop_0 galera-bundle-1" [ style = dashed] +"galera_stop_0 galera-bundle-2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-0_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-0_start_0 node1" -> "rabbitmq-bundle-0_monitor_30000 node1" [ style = dashed] +"rabbitmq-bundle-0_start_0 node1" -> "rabbitmq_monitor_10000 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq-bundle-0_start_0 node1" -> "rabbitmq_start_0 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq-bundle-0_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-0_stop_0 node1" -> "Cancel rabbitmq-bundle-0_monitor_60000 node1" [ style = bold] +"rabbitmq-bundle-0_stop_0 node1" -> "rabbitmq-bundle-0_start_0 node1" [ style = dashed] +"rabbitmq-bundle-0_stop_0 node1" -> "rabbitmq-bundle-docker-0_stop_0 node1" [ style = bold] +"rabbitmq-bundle-0_stop_0 node1" [ style=bold color="green" fontcolor="black"] +"rabbitmq-bundle-1_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-1_start_0 node1" -> "rabbitmq-bundle-1_monitor_30000 node1" [ style = dashed] +"rabbitmq-bundle-1_start_0 node1" -> "rabbitmq_monitor_10000 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle-1_start_0 node1" -> "rabbitmq_start_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle-1_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-1_stop_0 node2" -> "rabbitmq-bundle-1_start_0 node1" [ style = dashed] +"rabbitmq-bundle-1_stop_0 node2" -> "rabbitmq-bundle-docker-1_stop_0 node2" [ style = dashed] +"rabbitmq-bundle-1_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-2_monitor_30000 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-2_start_0 node1" -> "rabbitmq-bundle-2_monitor_30000 node1" [ style = dashed] +"rabbitmq-bundle-2_start_0 node1" -> "rabbitmq_monitor_10000 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle-2_start_0 node1" -> "rabbitmq_start_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle-2_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-2_stop_0 node3" -> "rabbitmq-bundle-2_start_0 node1" [ style = dashed] +"rabbitmq-bundle-2_stop_0 node3" -> "rabbitmq-bundle-docker-2_stop_0 node3" [ style = dashed] +"rabbitmq-bundle-2_stop_0 node3" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-clone_confirmed-post_notify_running_0" -> "rabbitmq-bundle_running_0" [ style = dashed] +"rabbitmq-bundle-clone_confirmed-post_notify_running_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_confirmed-post_notify_stopped_0" -> "rabbitmq-bundle-clone_pre_notify_start_0" [ style = dashed] +"rabbitmq-bundle-clone_confirmed-post_notify_stopped_0" -> "rabbitmq-bundle_stopped_0" [ style = bold] +"rabbitmq-bundle-clone_confirmed-post_notify_stopped_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-clone_confirmed-pre_notify_start_0" -> "rabbitmq-bundle-clone_post_notify_running_0" [ style = dashed] +"rabbitmq-bundle-clone_confirmed-pre_notify_start_0" -> "rabbitmq-bundle-clone_start_0" [ style = dashed] +"rabbitmq-bundle-clone_confirmed-pre_notify_start_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" -> "rabbitmq-bundle-clone_post_notify_stopped_0" [ style = bold] +"rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" -> "rabbitmq-bundle-clone_stop_0" [ style = bold] +"rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-clone_post_notify_running_0" -> "rabbitmq-bundle-clone_confirmed-post_notify_running_0" [ style = dashed] +"rabbitmq-bundle-clone_post_notify_running_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_post_notify_stopped_0" -> "rabbitmq-bundle-clone_confirmed-post_notify_stopped_0" [ style = bold] +"rabbitmq-bundle-clone_post_notify_stopped_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-clone_pre_notify_start_0" -> "rabbitmq-bundle-clone_confirmed-pre_notify_start_0" [ style = dashed] +"rabbitmq-bundle-clone_pre_notify_start_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_pre_notify_stop_0" -> "rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" [ style = bold] +"rabbitmq-bundle-clone_pre_notify_stop_0" -> "rabbitmq_pre_notify_stop_0 rabbitmq-bundle-0" [ style = bold] +"rabbitmq-bundle-clone_pre_notify_stop_0" -> "rabbitmq_pre_notify_stop_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle-clone_pre_notify_stop_0" -> "rabbitmq_pre_notify_stop_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle-clone_pre_notify_stop_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-clone_running_0" -> "rabbitmq-bundle-clone_post_notify_running_0" [ style = dashed] +"rabbitmq-bundle-clone_running_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_start_0" -> "rabbitmq-bundle-clone_running_0" [ style = dashed] +"rabbitmq-bundle-clone_start_0" -> "rabbitmq_start_0 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq-bundle-clone_start_0" -> "rabbitmq_start_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle-clone_start_0" -> "rabbitmq_start_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle-clone_start_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle-clone_stop_0" -> "rabbitmq-bundle-clone_stopped_0" [ style = bold] +"rabbitmq-bundle-clone_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-0" [ style = bold] +"rabbitmq-bundle-clone_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle-clone_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle-clone_stop_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-clone_stopped_0" -> "rabbitmq-bundle-clone_post_notify_stopped_0" [ style = bold] +"rabbitmq-bundle-clone_stopped_0" -> "rabbitmq-bundle-clone_start_0" [ style = dashed] +"rabbitmq-bundle-clone_stopped_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle-docker-0_monitor_60000 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-docker-0_start_0 node1" -> "rabbitmq-bundle-0_start_0 node1" [ style = dashed] +"rabbitmq-bundle-docker-0_start_0 node1" -> "rabbitmq-bundle-docker-0_monitor_60000 node1" [ style = dashed] +"rabbitmq-bundle-docker-0_start_0 node1" -> "rabbitmq-bundle_running_0" [ style = dashed] +"rabbitmq-bundle-docker-0_start_0 node1" -> "rabbitmq_start_0 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq-bundle-docker-0_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-docker-0_stop_0 node1" -> "rabbitmq-bundle-docker-0_start_0 node1" [ style = dashed] +"rabbitmq-bundle-docker-0_stop_0 node1" -> "rabbitmq-bundle_stopped_0" [ style = bold] +"rabbitmq-bundle-docker-0_stop_0 node1" [ style=bold color="green" fontcolor="black"] +"rabbitmq-bundle-docker-1_stop_0 node2" -> "rabbitmq-bundle_stopped_0" [ style = dashed] +"rabbitmq-bundle-docker-1_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle-docker-2_stop_0 node3" -> "rabbitmq-bundle_stopped_0" [ style = dashed] +"rabbitmq-bundle-docker-2_stop_0 node3" [ style=dashed color="red" fontcolor="black"] +"rabbitmq-bundle_running_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle_start_0" -> "rabbitmq-bundle-clone_start_0" [ style = dashed] +"rabbitmq-bundle_start_0" -> "rabbitmq-bundle-docker-0_start_0 node1" [ style = dashed] +"rabbitmq-bundle_start_0" [ style=dashed color="red" fontcolor="orange"] +"rabbitmq-bundle_stop_0" -> "rabbitmq-bundle-clone_stop_0" [ style = bold] +"rabbitmq-bundle_stop_0" -> "rabbitmq-bundle-docker-0_stop_0 node1" [ style = bold] +"rabbitmq-bundle_stop_0" -> "rabbitmq-bundle-docker-1_stop_0 node2" [ style = dashed] +"rabbitmq-bundle_stop_0" -> "rabbitmq-bundle-docker-2_stop_0 node3" [ style = dashed] +"rabbitmq-bundle_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-0" [ style = bold] +"rabbitmq-bundle_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq-bundle_stop_0" -> "rabbitmq_stop_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq-bundle_stop_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq-bundle_stopped_0" [ style=bold color="green" fontcolor="orange"] +"rabbitmq_monitor_10000 rabbitmq-bundle-0" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_monitor_10000 rabbitmq-bundle-1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_monitor_10000 rabbitmq-bundle-2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-0" -> "rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" [ style = bold] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-0" [ style=bold color="green" fontcolor="black"] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-1" -> "rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" [ style = dashed] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-2" -> "rabbitmq-bundle-clone_confirmed-pre_notify_stop_0" [ style = dashed] +"rabbitmq_pre_notify_stop_0 rabbitmq-bundle-2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_start_0 rabbitmq-bundle-0" -> "rabbitmq-bundle-clone_running_0" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-0" -> "rabbitmq_monitor_10000 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-0" -> "rabbitmq_start_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-0" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_start_0 rabbitmq-bundle-1" -> "rabbitmq-bundle-clone_running_0" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-1" -> "rabbitmq_monitor_10000 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-1" -> "rabbitmq_start_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_start_0 rabbitmq-bundle-2" -> "rabbitmq-bundle-clone_running_0" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-2" -> "rabbitmq_monitor_10000 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq_start_0 rabbitmq-bundle-2" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_stop_0 rabbitmq-bundle-0" -> "rabbitmq-bundle-0_stop_0 node1" [ style = bold] +"rabbitmq_stop_0 rabbitmq-bundle-0" -> "rabbitmq-bundle-clone_stopped_0" [ style = bold] +"rabbitmq_stop_0 rabbitmq-bundle-0" -> "rabbitmq_start_0 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-0" [ style=bold color="green" fontcolor="black"] +"rabbitmq_stop_0 rabbitmq-bundle-1" -> "rabbitmq-bundle-clone_stopped_0" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-1" -> "rabbitmq_start_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-1" -> "rabbitmq_stop_0 rabbitmq-bundle-0" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-1" [ style=dashed color="red" fontcolor="black"] +"rabbitmq_stop_0 rabbitmq-bundle-2" -> "rabbitmq-bundle-clone_stopped_0" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-2" -> "rabbitmq_start_0 rabbitmq-bundle-2" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-2" -> "rabbitmq_stop_0 rabbitmq-bundle-1" [ style = dashed] +"rabbitmq_stop_0 rabbitmq-bundle-2" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node1_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node2_monitor_60000 node1" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node2_start_0 node1" -> "stonith-fence_ipmilan-node2_monitor_60000 node1" [ style = dashed] +"stonith-fence_ipmilan-node2_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node2_stop_0 node3" -> "stonith-fence_ipmilan-node2_start_0 node1" [ style = dashed] +"stonith-fence_ipmilan-node2_stop_0 node3" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node3_monitor_60000 node1" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node3_start_0 node1" -> "stonith-fence_ipmilan-node3_monitor_60000 node1" [ style = dashed] +"stonith-fence_ipmilan-node3_start_0 node1" [ style=dashed color="red" fontcolor="black"] +"stonith-fence_ipmilan-node3_stop_0 node2" -> "stonith-fence_ipmilan-node3_start_0 node1" [ style = dashed] +"stonith-fence_ipmilan-node3_stop_0 node2" [ style=dashed color="red" fontcolor="black"] +} diff --git a/cts/scheduler/guest-host-not-fenceable.exp b/cts/scheduler/guest-host-not-fenceable.exp new file mode 100644 index 0000000000..b9293d7f47 --- /dev/null +++ b/cts/scheduler/guest-host-not-fenceable.exp @@ -0,0 +1,366 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/cts/scheduler/guest-host-not-fenceable.scores b/cts/scheduler/guest-host-not-fenceable.scores new file mode 100644 index 0000000000..130026dda5 --- /dev/null +++ b/cts/scheduler/guest-host-not-fenceable.scores @@ -0,0 +1,134 @@ +Allocation scores: +Using the original execution date of: 2019-08-26 04:52:42Z +clone_color: galera-bundle-master allocation score on galera-bundle-0: 0 +clone_color: galera-bundle-master allocation score on galera-bundle-1: 0 +clone_color: galera-bundle-master allocation score on galera-bundle-2: 0 +clone_color: galera-bundle-master allocation score on node1: -INFINITY +clone_color: galera-bundle-master allocation score on node2: -INFINITY +clone_color: galera-bundle-master allocation score on node3: -INFINITY +clone_color: galera:0 allocation score on galera-bundle-0: INFINITY +clone_color: galera:1 allocation score on galera-bundle-1: INFINITY +clone_color: galera:2 allocation score on galera-bundle-2: INFINITY +clone_color: rabbitmq-bundle-clone allocation score on node1: -INFINITY +clone_color: rabbitmq-bundle-clone allocation score on node2: -INFINITY +clone_color: rabbitmq-bundle-clone allocation score on node3: -INFINITY +clone_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-0: 0 +clone_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-1: 0 +clone_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-2: 0 +clone_color: rabbitmq:0 allocation score on rabbitmq-bundle-0: INFINITY +clone_color: rabbitmq:1 allocation score on rabbitmq-bundle-1: INFINITY +clone_color: rabbitmq:2 allocation score on rabbitmq-bundle-2: INFINITY +galera:0 promotion score on galera-bundle-0: 100 +galera:1 promotion score on galera-bundle-1: 100 +galera:2 promotion score on galera-bundle-2: -1 +native_color: galera-bundle-0 allocation score on node1: INFINITY +native_color: galera-bundle-0 allocation score on node2: 0 +native_color: galera-bundle-0 allocation score on node3: 0 +native_color: galera-bundle-1 allocation score on node1: 0 +native_color: galera-bundle-1 allocation score on node2: INFINITY +native_color: galera-bundle-1 allocation score on node3: 0 +native_color: galera-bundle-2 allocation score on node1: 0 +native_color: galera-bundle-2 allocation score on node2: 0 +native_color: galera-bundle-2 allocation score on node3: INFINITY +native_color: galera-bundle-docker-0 allocation score on node1: INFINITY +native_color: galera-bundle-docker-0 allocation score on node2: -INFINITY +native_color: galera-bundle-docker-0 allocation score on node3: -INFINITY +native_color: galera-bundle-docker-1 allocation score on node1: -INFINITY +native_color: galera-bundle-docker-1 allocation score on node2: -INFINITY +native_color: galera-bundle-docker-1 allocation score on node3: -INFINITY +native_color: galera-bundle-docker-2 allocation score on node1: -INFINITY +native_color: galera-bundle-docker-2 allocation score on node2: -INFINITY +native_color: galera-bundle-docker-2 allocation score on node3: -INFINITY +native_color: galera:0 allocation score on galera-bundle-0: INFINITY +native_color: galera:1 allocation score on galera-bundle-1: INFINITY +native_color: galera:2 allocation score on galera-bundle-2: INFINITY +native_color: rabbitmq-bundle-0 allocation score on node1: INFINITY +native_color: rabbitmq-bundle-0 allocation score on node2: 0 +native_color: rabbitmq-bundle-0 allocation score on node3: 0 +native_color: rabbitmq-bundle-1 allocation score on node1: 0 +native_color: rabbitmq-bundle-1 allocation score on node2: INFINITY +native_color: rabbitmq-bundle-1 allocation score on node3: 0 +native_color: rabbitmq-bundle-2 allocation score on node1: 0 +native_color: rabbitmq-bundle-2 allocation score on node2: 0 +native_color: rabbitmq-bundle-2 allocation score on node3: INFINITY +native_color: rabbitmq-bundle-docker-0 allocation score on node1: INFINITY +native_color: rabbitmq-bundle-docker-0 allocation score on node2: -INFINITY +native_color: rabbitmq-bundle-docker-0 allocation score on node3: -INFINITY +native_color: rabbitmq-bundle-docker-1 allocation score on node1: -INFINITY +native_color: rabbitmq-bundle-docker-1 allocation score on node2: -INFINITY +native_color: rabbitmq-bundle-docker-1 allocation score on node3: -INFINITY +native_color: rabbitmq-bundle-docker-2 allocation score on node1: -INFINITY +native_color: rabbitmq-bundle-docker-2 allocation score on node2: -INFINITY +native_color: rabbitmq-bundle-docker-2 allocation score on node3: -INFINITY +native_color: rabbitmq:0 allocation score on rabbitmq-bundle-0: INFINITY +native_color: rabbitmq:1 allocation score on rabbitmq-bundle-1: INFINITY +native_color: rabbitmq:2 allocation score on rabbitmq-bundle-2: INFINITY +native_color: stonith-fence_ipmilan-node1 allocation score on node1: -INFINITY +native_color: stonith-fence_ipmilan-node1 allocation score on node2: INFINITY +native_color: stonith-fence_ipmilan-node1 allocation score on node3: 0 +native_color: stonith-fence_ipmilan-node2 allocation score on node1: 0 +native_color: stonith-fence_ipmilan-node2 allocation score on node2: -INFINITY +native_color: stonith-fence_ipmilan-node2 allocation score on node3: INFINITY +native_color: stonith-fence_ipmilan-node3 allocation score on node1: 0 +native_color: stonith-fence_ipmilan-node3 allocation score on node2: INFINITY +native_color: stonith-fence_ipmilan-node3 allocation score on node3: -INFINITY +pcmk__bundle_color: galera-bundle allocation score on node1: 0 +pcmk__bundle_color: galera-bundle allocation score on node2: 0 +pcmk__bundle_color: galera-bundle allocation score on node3: 0 +pcmk__bundle_color: galera-bundle-0 allocation score on node1: INFINITY +pcmk__bundle_color: galera-bundle-0 allocation score on node2: 0 +pcmk__bundle_color: galera-bundle-0 allocation score on node3: 0 +pcmk__bundle_color: galera-bundle-1 allocation score on node1: 0 +pcmk__bundle_color: galera-bundle-1 allocation score on node2: INFINITY +pcmk__bundle_color: galera-bundle-1 allocation score on node3: 0 +pcmk__bundle_color: galera-bundle-2 allocation score on node1: 0 +pcmk__bundle_color: galera-bundle-2 allocation score on node2: 0 +pcmk__bundle_color: galera-bundle-2 allocation score on node3: INFINITY +pcmk__bundle_color: galera-bundle-docker-0 allocation score on node1: INFINITY +pcmk__bundle_color: galera-bundle-docker-0 allocation score on node2: 0 +pcmk__bundle_color: galera-bundle-docker-0 allocation score on node3: 0 +pcmk__bundle_color: galera-bundle-docker-1 allocation score on node1: 0 +pcmk__bundle_color: galera-bundle-docker-1 allocation score on node2: INFINITY +pcmk__bundle_color: galera-bundle-docker-1 allocation score on node3: 0 +pcmk__bundle_color: galera-bundle-docker-2 allocation score on node1: 0 +pcmk__bundle_color: galera-bundle-docker-2 allocation score on node2: 0 +pcmk__bundle_color: galera-bundle-docker-2 allocation score on node3: INFINITY +pcmk__bundle_color: galera-bundle-master allocation score on galera-bundle-0: -INFINITY +pcmk__bundle_color: galera-bundle-master allocation score on galera-bundle-1: -INFINITY +pcmk__bundle_color: galera-bundle-master allocation score on galera-bundle-2: -INFINITY +pcmk__bundle_color: galera-bundle-master allocation score on node1: 0 +pcmk__bundle_color: galera-bundle-master allocation score on node2: 0 +pcmk__bundle_color: galera-bundle-master allocation score on node3: 0 +pcmk__bundle_color: galera:0 allocation score on galera-bundle-0: INFINITY +pcmk__bundle_color: galera:1 allocation score on galera-bundle-1: INFINITY +pcmk__bundle_color: galera:2 allocation score on galera-bundle-2: INFINITY +pcmk__bundle_color: rabbitmq-bundle allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-0 allocation score on node1: INFINITY +pcmk__bundle_color: rabbitmq-bundle-0 allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle-0 allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-1 allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle-1 allocation score on node2: INFINITY +pcmk__bundle_color: rabbitmq-bundle-1 allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-2 allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle-2 allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle-2 allocation score on node3: INFINITY +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-0: -INFINITY +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-1: -INFINITY +pcmk__bundle_color: rabbitmq-bundle-clone allocation score on rabbitmq-bundle-2: -INFINITY +pcmk__bundle_color: rabbitmq-bundle-docker-0 allocation score on node1: INFINITY +pcmk__bundle_color: rabbitmq-bundle-docker-0 allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-0 allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-1 allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-1 allocation score on node2: INFINITY +pcmk__bundle_color: rabbitmq-bundle-docker-1 allocation score on node3: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-2 allocation score on node1: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-2 allocation score on node2: 0 +pcmk__bundle_color: rabbitmq-bundle-docker-2 allocation score on node3: INFINITY +pcmk__bundle_color: rabbitmq:0 allocation score on rabbitmq-bundle-0: INFINITY +pcmk__bundle_color: rabbitmq:1 allocation score on rabbitmq-bundle-1: INFINITY +pcmk__bundle_color: rabbitmq:2 allocation score on rabbitmq-bundle-2: INFINITY diff --git a/cts/scheduler/guest-host-not-fenceable.summary b/cts/scheduler/guest-host-not-fenceable.summary new file mode 100644 index 0000000000..034dfb8085 --- /dev/null +++ b/cts/scheduler/guest-host-not-fenceable.summary @@ -0,0 +1,89 @@ +Using the original execution date of: 2019-08-26 04:52:42Z + +Current cluster status: +Node node2 (2): UNCLEAN (offline) +Node node3 (3): UNCLEAN (offline) +Online: [ node1 ] +GuestOnline: [ galera-bundle-0:galera-bundle-docker-0 rabbitmq-bundle-0:rabbitmq-bundle-docker-0 ] + + Container bundle set: rabbitmq-bundle [192.168.122.139:8787/rhosp13/openstack-rabbitmq:pcmklatest] + rabbitmq-bundle-0 (ocf::heartbeat:rabbitmq-cluster): Started node1 + rabbitmq-bundle-1 (ocf::heartbeat:rabbitmq-cluster): FAILED node2 (UNCLEAN) + rabbitmq-bundle-2 (ocf::heartbeat:rabbitmq-cluster): FAILED node3 (UNCLEAN) + Container bundle set: galera-bundle [192.168.122.139:8787/rhosp13/openstack-mariadb:pcmklatest] + galera-bundle-0 (ocf::heartbeat:galera): FAILED Master node1 + galera-bundle-1 (ocf::heartbeat:galera): FAILED Master node2 (UNCLEAN) + galera-bundle-2 (ocf::heartbeat:galera): FAILED Master node3 (UNCLEAN) + stonith-fence_ipmilan-node1 (stonith:fence_ipmilan): Started node2 (UNCLEAN) + stonith-fence_ipmilan-node3 (stonith:fence_ipmilan): Started node2 (UNCLEAN) + stonith-fence_ipmilan-node2 (stonith:fence_ipmilan): Started node3 (UNCLEAN) + +Transition Summary: + * Stop rabbitmq-bundle-docker-0 ( node1 ) due to no quorum + * Stop rabbitmq-bundle-0 ( node1 ) due to no quorum + * Stop rabbitmq:0 ( rabbitmq-bundle-0 ) due to no quorum + * Stop rabbitmq-bundle-docker-1 ( node2 ) due to node availability (blocked) + * Stop rabbitmq-bundle-1 ( node2 ) due to no quorum (blocked) + * Stop rabbitmq:1 ( rabbitmq-bundle-1 ) due to no quorum (blocked) + * Stop rabbitmq-bundle-docker-2 ( node3 ) due to node availability (blocked) + * Stop rabbitmq-bundle-2 ( node3 ) due to no quorum (blocked) + * Stop rabbitmq:2 ( rabbitmq-bundle-2 ) due to no quorum (blocked) + * Stop galera-bundle-docker-0 ( node1 ) due to no quorum + * Stop galera-bundle-0 ( node1 ) due to no quorum + * Stop galera:0 ( Master galera-bundle-0 ) due to no quorum + * Stop galera-bundle-docker-1 ( node2 ) due to node availability (blocked) + * Stop galera-bundle-1 ( node2 ) due to no quorum (blocked) + * Stop galera:1 ( Master galera-bundle-1 ) due to no quorum (blocked) + * Stop galera-bundle-docker-2 ( node3 ) due to node availability (blocked) + * Stop galera-bundle-2 ( node3 ) due to no quorum (blocked) + * Stop galera:2 ( Master galera-bundle-2 ) due to no quorum (blocked) + * Stop stonith-fence_ipmilan-node1 ( node2 ) due to node availability (blocked) + * Stop stonith-fence_ipmilan-node3 ( node2 ) due to no quorum (blocked) + * Stop stonith-fence_ipmilan-node2 ( node3 ) due to no quorum (blocked) + +Executing cluster transition: + * Pseudo action: rabbitmq-bundle-clone_pre_notify_stop_0 + * Pseudo action: galera-bundle_demote_0 + * Pseudo action: rabbitmq-bundle_stop_0 + * Resource action: rabbitmq notify on rabbitmq-bundle-0 + * Pseudo action: rabbitmq-bundle-clone_confirmed-pre_notify_stop_0 + * Pseudo action: rabbitmq-bundle-clone_stop_0 + * Pseudo action: galera-bundle-master_demote_0 + * Resource action: rabbitmq stop on rabbitmq-bundle-0 + * Pseudo action: rabbitmq-bundle-clone_stopped_0 + * Resource action: rabbitmq-bundle-0 stop on node1 + * Resource action: rabbitmq-bundle-0 cancel=60000 on node1 + * Resource action: galera demote on galera-bundle-0 + * Pseudo action: galera-bundle-master_demoted_0 + * Pseudo action: galera-bundle_demoted_0 + * Pseudo action: galera-bundle_stop_0 + * Pseudo action: rabbitmq-bundle-clone_post_notify_stopped_0 + * Resource action: rabbitmq-bundle-docker-0 stop on node1 + * Pseudo action: galera-bundle-master_stop_0 + * Pseudo action: rabbitmq-bundle-clone_confirmed-post_notify_stopped_0 + * Resource action: galera stop on galera-bundle-0 + * Pseudo action: galera-bundle-master_stopped_0 + * Resource action: galera-bundle-0 stop on node1 + * Resource action: galera-bundle-0 cancel=60000 on node1 + * Pseudo action: rabbitmq-bundle_stopped_0 + * Resource action: galera-bundle-docker-0 stop on node1 + * Pseudo action: galera-bundle_stopped_0 +Using the original execution date of: 2019-08-26 04:52:42Z + +Revised cluster status: +Node node2 (2): UNCLEAN (offline) +Node node3 (3): UNCLEAN (offline) +Online: [ node1 ] + + Container bundle set: rabbitmq-bundle [192.168.122.139:8787/rhosp13/openstack-rabbitmq:pcmklatest] + rabbitmq-bundle-0 (ocf::heartbeat:rabbitmq-cluster): Stopped + rabbitmq-bundle-1 (ocf::heartbeat:rabbitmq-cluster): FAILED node2 (UNCLEAN) + rabbitmq-bundle-2 (ocf::heartbeat:rabbitmq-cluster): FAILED node3 (UNCLEAN) + Container bundle set: galera-bundle [192.168.122.139:8787/rhosp13/openstack-mariadb:pcmklatest] + galera-bundle-0 (ocf::heartbeat:galera): Stopped + galera-bundle-1 (ocf::heartbeat:galera): FAILED Master node2 (UNCLEAN) + galera-bundle-2 (ocf::heartbeat:galera): FAILED Master node3 (UNCLEAN) + stonith-fence_ipmilan-node1 (stonith:fence_ipmilan): Started node2 (UNCLEAN) + stonith-fence_ipmilan-node3 (stonith:fence_ipmilan): Started node2 (UNCLEAN) + stonith-fence_ipmilan-node2 (stonith:fence_ipmilan): Started node3 (UNCLEAN) + diff --git a/cts/scheduler/guest-host-not-fenceable.xml b/cts/scheduler/guest-host-not-fenceable.xml new file mode 100755 index 0000000000..a1ccdc84ea --- /dev/null +++ b/cts/scheduler/guest-host-not-fenceable.xml @@ -0,0 +1,413 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c index c5a6f4cbeb..1a46cc5007 100644 --- a/lib/pacemaker/pcmk_sched_allocate.c +++ b/lib/pacemaker/pcmk_sched_allocate.c @@ -1,2976 +1,2977 @@ /* * 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 #include #include #include #include CRM_TRACE_INIT_DATA(pe_allocate); 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 gboolean apply_placement_constraints(pe_working_set_t * data_set) { GListPtr gIter = NULL; crm_trace("Applying constraints..."); for (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); } return TRUE; } 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 && g_list_length(rsc->running_on) == 1) { 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; } } /* * 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; crm_trace("Applying placement constraints"); gIter = data_set->nodes; for (; 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; 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(show_utilization ? 0 : utilization_log_level, "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(show_utilization ? 0 : utilization_log_level, "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 */ stonith_constraints(node, 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) { + 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); stonith_constraints(node, 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 (g_list_length(remote_rsc->running_on) > 1 && 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_native.c b/lib/pacemaker/pcmk_sched_native.c index bf14758f36..71d4486b39 100644 --- a/lib/pacemaker/pcmk_sched_native.c +++ b/lib/pacemaker/pcmk_sched_native.c @@ -1,3331 +1,3329 @@ /* * 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 // 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); print_resource(alloc_details, "Allocating: ", rsc, FALSE); 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); } print_resource(LOG_TRACE, "Allocating: ", rsc, FALSE); 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 ? 0 : 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); print_resource(LOG_TRACE, "Allocated ", rsc, TRUE); 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); } // 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 && g_list_length(then_rsc->running_on) == 1 && 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; + 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 - if (is_set(rsc->flags, pe_rsc_needs_fencing)) { - order_implicit = TRUE; - } - - /* Implied stops and demotes of resources running on guest nodes are always + /* 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 (pe__is_guest_node(target)) { - order_implicit = TRUE; + 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, __FUNCTION__, __LINE__); - update_action_flags(action, pe_action_runnable, __FUNCTION__, __LINE__); + update_action_flags(action, pe_action_pseudo|pe_action_runnable, + __FUNCTION__, __LINE__); if (order_implicit) { - enum pe_ordering flags = pe_order_optional; - action_t *parent_stop = find_first_action(top->actions, NULL, RSC_STOP, NULL); - - if (is_set(rsc->flags, pe_rsc_failed)) { - crm_notice("Stop of failed resource %s is implicit after %s is fenced", - rsc->id, target->details->uname); - } else { - crm_info("%s is implicit after %s is fenced", - action->uuid, target->details->uname); - } update_action_flags(action, pe_action_implied_by_stonith, __FUNCTION__, __LINE__); - if (target->details->remote_rsc) { - /* User constraints must not order a resource in a guest node - * relative to the guest node container resource. This flag - * marks constraints as generated by the cluster and thus - * immune to that check. - */ - flags |= pe_order_preserve; - } - if (pe_rsc_is_bundled(rsc) == FALSE) { - order_actions(stonith_op, action, flags); + /* 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, flags); + 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 { - if (is_set(rsc->flags, pe_rsc_failed)) { - crm_notice("Stop of failed resource %s is implicit because %s will be fenced", - rsc->id, target->details->uname); - } else { - crm_info("%s is implicit because %s will be fenced", - action->uuid, target->details->uname); - } + 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, __FUNCTION__, __LINE__); - update_action_flags(action, pe_action_runnable, __FUNCTION__, __LINE__); + 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_trans_utils.c b/lib/pacemaker/pcmk_trans_utils.c index 878ee82634..7e840b9bdf 100644 --- a/lib/pacemaker/pcmk_trans_utils.c +++ b/lib/pacemaker/pcmk_trans_utils.c @@ -1,279 +1,303 @@ /* * 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 Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include extern crm_graph_functions_t *graph_fns; static gboolean pseudo_action_dummy(crm_graph_t * graph, crm_action_t * action) { static int fail = -1; if (fail < 0) { char *fail_s = getenv("PE_fail"); if (fail_s) { fail = crm_int_helper(fail_s, NULL); } else { fail = 0; } } crm_trace("Dummy event handler: action %d executed", action->id); if (action->id == fail) { crm_err("Dummy event handler: pretending action %d failed", action->id); action->failed = TRUE; graph->abort_priority = INFINITY; } action->confirmed = TRUE; update_graph(graph, action); return TRUE; } crm_graph_functions_t default_fns = { pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy }; void set_default_graph_functions(void) { graph_fns = &default_fns; } void set_graph_functions(crm_graph_functions_t * fns) { crm_info("Setting custom graph functions"); graph_fns = fns; CRM_ASSERT(graph_fns != NULL); CRM_ASSERT(graph_fns->rsc != NULL); CRM_ASSERT(graph_fns->crmd != NULL); CRM_ASSERT(graph_fns->pseudo != NULL); CRM_ASSERT(graph_fns->stonith != NULL); } const char * transition_status(enum transition_status state) { switch (state) { case transition_active: return "active"; case transition_pending: return "pending"; case transition_complete: return "complete"; case transition_stopped: return "stopped"; case transition_terminated: return "terminated"; case transition_action_failed: return "failed (action)"; case transition_failed: return "failed"; } return "unknown"; } const char * actiontype2text(action_type_e type) { switch (type) { case action_type_pseudo: return "pseudo"; case action_type_rsc: - return "rsc"; + return "resource"; case action_type_crm: - return "crm"; - + return "cluster"; } - return ""; + return "invalid"; } static crm_action_t * find_action(crm_graph_t * graph, int id) { GListPtr sIter = NULL; if (graph == NULL) { return NULL; } for (sIter = graph->synapses; sIter != NULL; sIter = sIter->next) { GListPtr aIter = NULL; synapse_t *synapse = (synapse_t *) sIter->data; for (aIter = synapse->actions; aIter != NULL; aIter = aIter->next) { crm_action_t *action = (crm_action_t *) aIter->data; if (action->id == id) { return action; } } } return NULL; } -static void -print_synapse(unsigned int log_level, crm_graph_t * graph, synapse_t * synapse) +static const char * +synapse_state_str(synapse_t *synapse) { - GListPtr lpc = NULL; - char *pending = NULL; - const char *state = "Pending"; - if (synapse->failed) { - state = "Failed"; + return "Failed"; } else if (synapse->confirmed) { - state = "Completed"; + return "Completed"; } else if (synapse->executed) { - state = "In-flight"; + return "In-flight"; } else if (synapse->ready) { - state = "Ready"; + return "Ready"; } + return "Pending"; +} - if (synapse->executed == FALSE) { - for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { - crm_action_t *input = (crm_action_t *) lpc->data; - const char *id_string = crm_element_value(input->xml, XML_ATTR_ID); +// List action IDs of inputs in graph that haven't completed successfully +static char * +synapse_pending_inputs(crm_graph_t *graph, synapse_t *synapse) +{ + char *pending = NULL; - if (input->failed) { - pending = add_list_element(pending, id_string); + for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { + crm_action_t *input = (crm_action_t *) lpc->data; - } else if (input->confirmed) { - /* Confirmed, skip */ + if (input->failed) { + pending = add_list_element(pending, ID(input->xml)); - } else if (find_action(graph, input->id)) { - /* In-flight or pending */ - pending = add_list_element(pending, id_string); - } + } else if (input->confirmed) { + // Confirmed successful inputs are not pending + + } else if (find_action(graph, input->id) != NULL) { + // In-flight or pending + pending = add_list_element(pending, ID(input->xml)); } } - - for (lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { - crm_action_t *action = (crm_action_t *) lpc->data; - const char *key = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); - const char *host = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); - char *desc = crm_strdup_printf("%s %s op %s", state, actiontype2text(action->type), key); - - do_crm_log(log_level, - "[Action %4d]: %-50s on %s (priority: %d, waiting: %s)", - action->id, desc, host ? host : "N/A", - synapse->priority, pending ? pending : "none"); - - free(desc); + if (pending == NULL) { + pending = strdup("none"); } + return pending; +} - if (synapse->executed == FALSE) { - for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { - crm_action_t *input = (crm_action_t *) lpc->data; - const char *key = crm_element_value(input->xml, XML_LRM_ATTR_TASK_KEY); - const char *host = crm_element_value(input->xml, XML_LRM_ATTR_TARGET); - - if (find_action(graph, input->id) == NULL) { - if (host == NULL) { - do_crm_log(log_level, " * [Input %2d]: Unresolved dependency %s op %s", - input->id, actiontype2text(input->type), key); - } else { - do_crm_log(log_level, " * [Input %2d]: Unresolved dependency %s op %s on %s", - input->id, actiontype2text(input->type), key, host); - } - } +// Log synapse inputs that aren't in graph +static void +log_unresolved_inputs(unsigned int log_level, crm_graph_t *graph, + synapse_t *synapse) +{ + for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { + crm_action_t *input = (crm_action_t *) lpc->data; + const char *key = crm_element_value(input->xml, XML_LRM_ATTR_TASK_KEY); + const char *host = crm_element_value(input->xml, XML_LRM_ATTR_TARGET); + + if (find_action(graph, input->id) == NULL) { + do_crm_log(log_level, + " * [Input %2d]: Unresolved dependency %s op %s%s%s", + input->id, actiontype2text(input->type), key, + (host? " on " : ""), (host? host : "")); } } +} + +static void +log_synapse_action(unsigned int log_level, synapse_t *synapse, + crm_action_t *action, const char *pending_inputs) +{ + const char *key = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); + const char *host = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); + char *desc = crm_strdup_printf("%s %s op %s", + synapse_state_str(synapse), + actiontype2text(action->type), key); + + do_crm_log(log_level, + "[Action %4d]: %-50s%s%s (priority: %d, waiting: %s)", + action->id, desc, (host? " on " : ""), (host? host : ""), + synapse->priority, pending_inputs); + free(desc); +} +static void +print_synapse(unsigned int log_level, crm_graph_t * graph, synapse_t * synapse) +{ + char *pending = NULL; + + if (!synapse->executed) { + pending = synapse_pending_inputs(graph, synapse); + } + for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { + log_synapse_action(log_level, synapse, (crm_action_t *) lpc->data, + pending); + } free(pending); + if (!synapse->executed) { + log_unresolved_inputs(log_level, graph, synapse); + } } void print_action(int log_level, const char *prefix, crm_action_t * action) { print_synapse(log_level, NULL, action->synapse); } void print_graph(unsigned int log_level, crm_graph_t * graph) { GListPtr lpc = NULL; if (graph == NULL || graph->num_actions == 0) { if (log_level > LOG_DEBUG) { crm_debug("Empty transition graph"); } return; } do_crm_log(log_level, "Graph %d with %d actions:" " batch-limit=%d jobs, network-delay=%ums", graph->id, graph->num_actions, graph->batch_limit, graph->network_delay); for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { synapse_t *synapse = (synapse_t *) lpc->data; print_synapse(log_level, graph, synapse); } } static const char * abort2text(enum transition_action abort_action) { switch (abort_action) { case tg_done: return "done"; case tg_stop: return "stop"; case tg_restart: return "restart"; case tg_shutdown: return "shutdown"; } return "unknown"; } bool update_abort_priority(crm_graph_t * graph, int priority, enum transition_action action, const char *abort_reason) { bool change = FALSE; if (graph == NULL) { return change; } if (graph->abort_priority < priority) { crm_debug("Abort priority upgraded from %d to %d", graph->abort_priority, priority); graph->abort_priority = priority; if (graph->abort_reason != NULL) { crm_debug("'%s' abort superseded by %s", graph->abort_reason, abort_reason); } graph->abort_reason = abort_reason; change = TRUE; } if (graph->completion_action < action) { crm_debug("Abort action %s superseded by %s: %s", abort2text(graph->completion_action), abort2text(action), abort_reason); graph->completion_action = action; change = TRUE; } return change; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index c5a5072375..0058ebe707 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,2561 +1,2586 @@ /* * 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 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 extern xmlNode *get_object_root(const char *object_type, xmlNode * the_root); void print_str_str(gpointer key, gpointer value, gpointer user_data); gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); void unpack_operation(action_t * action, xmlNode * xml_obj, resource_t * container, pe_working_set_t * data_set); static xmlNode *find_rsc_op_entry_helper(resource_t * rsc, const char *key, gboolean include_disabled); #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t * pe_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; CRM_CHECK(action != NULL, return NULL); if (action->action_details == NULL) { action->action_details = calloc(1, sizeof(pe_rsc_action_details_t)); CRM_CHECK(action->action_details != NULL, return NULL); } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters == NULL) { details->versioned_parameters = create_xml_node(NULL, XML_TAG_OP_VER_ATTRS); } if (details->versioned_meta == NULL) { details->versioned_meta = create_xml_node(NULL, XML_TAG_OP_VER_META); } return details; } static void pe_free_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; if ((action == NULL) || (action->action_details == NULL)) { return; } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters) { free_xml(details->versioned_parameters); } if (details->versioned_meta) { free_xml(details->versioned_meta); } action->action_details = NULL; } #endif /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] data_set Working set for cluster * \param[in] node Name of node to check * - * \return TRUE if node can be fenced, FALSE otherwise - * - * \note This function should only be called for cluster nodes and - * remote nodes; guest nodes are fenced by stopping their container - * resource, so fence execution requirements do not apply to them. + * \return true if node can be fenced, false otherwise */ -bool pe_can_fence(pe_working_set_t * data_set, node_t *node) +bool +pe_can_fence(pe_working_set_t *data_set, pe_node_t *node) { - if(is_not_set(data_set->flags, pe_flag_stonith_enabled)) { - return FALSE; /* Turned off */ + if (pe__is_guest_node(node)) { + /* Guest nodes are fenced by stopping their container resource. We can + * do that if the container's host is either online or fenceable. + */ + pe_resource_t *rsc = node->details->remote_rsc->container; + + for (GList *n = rsc->running_on; n != NULL; n = n->next) { + pe_node_t *container_node = n->data; + + if (!container_node->details->online + && !pe_can_fence(data_set, container_node)) { + return false; + } + } + return true; + + } else if(is_not_set(data_set->flags, pe_flag_stonith_enabled)) { + return false; /* Turned off */ } else if (is_not_set(data_set->flags, pe_flag_have_stonith_resource)) { - return FALSE; /* No devices */ + return false; /* No devices */ } else if (is_set(data_set->flags, pe_flag_have_quorum)) { - return TRUE; + return true; } else if (data_set->no_quorum_policy == no_quorum_ignore) { - return TRUE; + return true; } else if(node == NULL) { - return FALSE; + return false; } else if(node->details->online) { crm_notice("We can fence %s without quorum because they're in our membership", node->details->uname); - return TRUE; + return true; } crm_trace("Cannot fence %s", node->details->uname); - return FALSE; + return false; } node_t * node_copy(const node_t *this_node) { node_t *new_node = NULL; CRM_CHECK(this_node != NULL, return NULL); new_node = calloc(1, sizeof(node_t)); CRM_ASSERT(new_node != NULL); crm_trace("Copying %p (%s) to %p", this_node, this_node->details->uname, new_node); new_node->rsc_discover_mode = this_node->rsc_discover_mode; new_node->weight = this_node->weight; new_node->fixed = this_node->fixed; new_node->details = this_node->details; return new_node; } /* any node in list1 or list2 and not in the other gets a score of -INFINITY */ void node_list_exclude(GHashTable * hash, GListPtr list, gboolean merge_scores) { GHashTable *result = hash; node_t *other_node = NULL; GListPtr gIter = list; GHashTableIter iter; node_t *node = NULL; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { other_node = pe_find_node_id(list, node->details->id); if (other_node == NULL) { node->weight = -INFINITY; } else if (merge_scores) { node->weight = merge_weights(node->weight, other_node->weight); } } for (; gIter != NULL; gIter = gIter->next) { node_t *node = (node_t *) gIter->data; other_node = pe_hash_table_lookup(result, node->details->id); if (other_node == NULL) { node_t *new_node = node_copy(node); new_node->weight = -INFINITY; g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } } } GHashTable * node_hash_from_list(GListPtr list) { GListPtr gIter = list; GHashTable *result = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free); for (; gIter != NULL; gIter = gIter->next) { node_t *node = (node_t *) gIter->data; node_t *n = node_copy(node); g_hash_table_insert(result, (gpointer) n->details->id, n); } return result; } GListPtr node_list_dup(GListPtr list1, gboolean reset, gboolean filter) { GListPtr result = NULL; GListPtr gIter = list1; for (; gIter != NULL; gIter = gIter->next) { node_t *new_node = NULL; node_t *this_node = (node_t *) gIter->data; if (filter && this_node->weight < 0) { continue; } new_node = node_copy(this_node); if (reset) { new_node->weight = 0; } if (new_node != NULL) { result = g_list_prepend(result, new_node); } } return result; } gint sort_node_uname(gconstpointer a, gconstpointer b) { const char *name_a = ((const node_t *) a)->details->uname; const char *name_b = ((const node_t *) b)->details->uname; while (*name_a && *name_b) { if (isdigit(*name_a) && isdigit(*name_b)) { // If node names contain a number, sort numerically char *end_a = NULL; char *end_b = NULL; long num_a = strtol(name_a, &end_a, 10); long num_b = strtol(name_b, &end_b, 10); // allow ordering e.g. 007 > 7 size_t len_a = end_a - name_a; size_t len_b = end_b - name_b; if (num_a < num_b) { return -1; } else if (num_a > num_b) { return 1; } else if (len_a < len_b) { return -1; } else if (len_a > len_b) { return 1; } name_a = end_a; name_b = end_b; } else { // Compare non-digits case-insensitively int lower_a = tolower(*name_a); int lower_b = tolower(*name_b); if (lower_a < lower_b) { return -1; } else if (lower_a > lower_b) { return 1; } ++name_a; ++name_b; } } if (!*name_a && *name_b) { return -1; } else if (*name_a && !*name_b) { return 1; } return 0; } void dump_node_scores_worker(int level, const char *file, const char *function, int line, resource_t * rsc, const char *comment, GHashTable * nodes) { GHashTable *hash = nodes; GHashTableIter iter; node_t *node = NULL; if (rsc) { hash = rsc->allowed_nodes; } if (rsc && is_set(rsc->flags, pe_rsc_orphan)) { /* Don't show the allocation scores for orphans */ return; } if (level == 0) { char score[128]; int len = sizeof(score); /* For now we want this in sorted order to keep the regression tests happy */ GListPtr gIter = NULL; GListPtr list = g_hash_table_get_values(hash); list = g_list_sort(list, sort_node_uname); gIter = list; for (; gIter != NULL; gIter = gIter->next) { node_t *node = (node_t *) gIter->data; /* This function is called a whole lot, use stack allocated score */ score2char_stack(node->weight, score, len); if (rsc) { printf("%s: %s allocation score on %s: %s\n", comment, rsc->id, node->details->uname, score); } else { printf("%s: %s = %s\n", comment, node->details->uname, score); } } g_list_free(list); } else if (hash) { char score[128]; int len = sizeof(score); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { /* This function is called a whole lot, use stack allocated score */ score2char_stack(node->weight, score, len); if (rsc) { do_crm_log_alias(LOG_TRACE, file, function, line, "%s: %s allocation score on %s: %s", comment, rsc->id, node->details->uname, score); } else { do_crm_log_alias(LOG_TRACE, file, function, line + 1, "%s: %s = %s", comment, node->details->uname, score); } } } if (rsc && rsc->children) { GListPtr gIter = NULL; gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child = (resource_t *) gIter->data; dump_node_scores_worker(level, file, function, line, child, comment, nodes); } } } static void append_dump_text(gpointer key, gpointer value, gpointer user_data) { char **dump_text = user_data; char *new_text = crm_strdup_printf("%s %s=%s", *dump_text, (char *)key, (char *)value); free(*dump_text); *dump_text = new_text; } void dump_node_capacity(int level, const char *comment, node_t * node) { char *dump_text = crm_strdup_printf("%s: %s capacity:", comment, node->details->uname); g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text); if (level == 0) { fprintf(stdout, "%s\n", dump_text); } else { crm_trace("%s", dump_text); } free(dump_text); } void dump_rsc_utilization(int level, const char *comment, resource_t * rsc, node_t * node) { char *dump_text = crm_strdup_printf("%s: %s utilization on %s:", comment, rsc->id, node->details->uname); g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text); if (level == 0) { fprintf(stdout, "%s\n", dump_text); } else { crm_trace("%s", dump_text); } free(dump_text); } gint sort_rsc_index(gconstpointer a, gconstpointer b) { const resource_t *resource1 = (const resource_t *)a; const resource_t *resource2 = (const resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->sort_index > resource2->sort_index) { return -1; } if (resource1->sort_index < resource2->sort_index) { return 1; } return 0; } gint sort_rsc_priority(gconstpointer a, gconstpointer b) { const resource_t *resource1 = (const resource_t *)a; const resource_t *resource2 = (const resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->priority > resource2->priority) { return -1; } if (resource1->priority < resource2->priority) { return 1; } return 0; } action_t * custom_action(resource_t * rsc, char *key, const char *task, node_t * on_node, gboolean optional, gboolean save_action, pe_working_set_t * data_set) { action_t *action = NULL; GListPtr possible_matches = NULL; CRM_CHECK(key != NULL, return NULL); CRM_CHECK(task != NULL, free(key); return NULL); if (save_action && rsc != NULL) { possible_matches = find_actions(rsc->actions, key, on_node); } else if(save_action) { #if 0 action = g_hash_table_lookup(data_set->singletons, key); #else /* More expensive but takes 'node' into account */ possible_matches = find_actions(data_set->actions, key, on_node); #endif } if(data_set->singletons == NULL) { data_set->singletons = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, NULL); } if (possible_matches != NULL) { if (g_list_length(possible_matches) > 1) { pe_warn("Action %s for %s on %s exists %d times", task, rsc ? rsc->id : "", on_node ? on_node->details->uname : "", g_list_length(possible_matches)); } action = g_list_nth_data(possible_matches, 0); pe_rsc_trace(rsc, "Found existing action %d (%s) for %s (%s) on %s", action->id, action->uuid, (rsc? rsc->id : "no resource"), task, (on_node? on_node->details->uname : "no node")); g_list_free(possible_matches); } if (action == NULL) { if (save_action) { pe_rsc_trace(rsc, "Creating %s action %d: %s for %s (%s) on %s", (optional? "optional" : " mandatory"), data_set->action_id, key, (rsc? rsc->id : "no resource"), task, (on_node? on_node->details->uname : "no node")); } action = calloc(1, sizeof(action_t)); if (save_action) { action->id = data_set->action_id++; } else { action->id = 0; } action->rsc = rsc; CRM_ASSERT(task != NULL); action->task = strdup(task); if (on_node) { action->node = node_copy(on_node); } action->uuid = strdup(key); pe_set_action_bit(action, pe_action_runnable); if (optional) { pe_set_action_bit(action, pe_action_optional); } else { pe_clear_action_bit(action, pe_action_optional); } action->extra = crm_str_table_new(); action->meta = crm_str_table_new(); if (save_action) { data_set->actions = g_list_prepend(data_set->actions, action); if(rsc == NULL) { g_hash_table_insert(data_set->singletons, action->uuid, action); } } if (rsc != NULL) { action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE); unpack_operation(action, action->op_entry, rsc->container, data_set); if (save_action) { rsc->actions = g_list_prepend(rsc->actions, action); } } if (save_action) { pe_rsc_trace(rsc, "Action %d created", action->id); } } if (!optional && is_set(action->flags, pe_action_optional)) { pe_rsc_trace(rsc, "Unset optional on action %d", action->id); pe_clear_action_bit(action, pe_action_optional); } if (rsc != NULL) { enum action_tasks a_task = text2task(action->task); int warn_level = LOG_TRACE; if (save_action) { warn_level = LOG_WARNING; } if (is_set(action->flags, pe_action_have_node_attrs) == FALSE && action->node != NULL && action->op_entry != NULL) { pe_set_action_bit(action, pe_action_have_node_attrs); pe__unpack_dataset_nvpairs(action->op_entry, XML_TAG_ATTR_SETS, action->node->details->attrs, action->extra, NULL, FALSE, data_set); } if (is_set(action->flags, pe_action_pseudo)) { /* leave untouched */ } else if (action->node == NULL) { pe_rsc_trace(rsc, "Unset runnable on %s", action->uuid); pe_clear_action_bit(action, pe_action_runnable); } else if (is_not_set(rsc->flags, pe_rsc_managed) && g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS) == NULL) { crm_debug("Action %s (unmanaged)", action->uuid); pe_rsc_trace(rsc, "Set optional on %s", action->uuid); pe_set_action_bit(action, pe_action_optional); /* action->runnable = FALSE; */ } else if (action->node->details->online == FALSE && (!pe__is_guest_node(action->node) || action->node->details->remote_requires_reset)) { pe_clear_action_bit(action, pe_action_runnable); do_crm_log(warn_level, "Action %s on %s is unrunnable (offline)", action->uuid, action->node->details->uname); if (is_set(action->rsc->flags, pe_rsc_managed) && save_action && a_task == stop_rsc && action->node->details->unclean == FALSE) { pe_fence_node(data_set, action->node, "resource actions are unrunnable"); } } else if (action->node->details->pending) { pe_clear_action_bit(action, pe_action_runnable); do_crm_log(warn_level, "Action %s on %s is unrunnable (pending)", action->uuid, action->node->details->uname); } else if (action->needs == rsc_req_nothing) { pe_rsc_trace(rsc, "Action %s does not require anything", action->uuid); pe_action_set_reason(action, NULL, TRUE); - pe_set_action_bit(action, pe_action_runnable); + if (pe__is_guest_node(action->node) + && !pe_can_fence(data_set, action->node)) { + /* An action that requires nothing usually does not require any + * fencing in order to be runnable. However, there is an + * exception: an action cannot be completed if it is on a guest + * node whose host is unclean and cannot be fenced. + */ + pe_clear_action_bit(action, pe_action_runnable); + crm_debug("%s\t%s (cancelled : host cannot be fenced)", + action->node->details->uname, action->uuid); + } else { + pe_set_action_bit(action, pe_action_runnable); + } #if 0 /* * No point checking this * - if we don't have quorum we can't stonith anyway */ } else if (action->needs == rsc_req_stonith) { crm_trace("Action %s requires only stonith", action->uuid); action->runnable = TRUE; #endif } else if (is_set(data_set->flags, pe_flag_have_quorum) == FALSE && data_set->no_quorum_policy == no_quorum_stop) { pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, NULL, "no quorum", pe_action_runnable, TRUE); crm_debug("%s\t%s (cancelled : quorum)", action->node->details->uname, action->uuid); } else if (is_set(data_set->flags, pe_flag_have_quorum) == FALSE && data_set->no_quorum_policy == no_quorum_freeze) { pe_rsc_trace(rsc, "Check resource is already active: %s %s %s %s", rsc->id, action->uuid, role2text(rsc->next_role), role2text(rsc->role)); if (rsc->fns->active(rsc, TRUE) == FALSE || rsc->next_role > rsc->role) { pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, NULL, "quorum freeze", pe_action_runnable, TRUE); pe_rsc_debug(rsc, "%s\t%s (cancelled : quorum freeze)", action->node->details->uname, action->uuid); } } else if(is_not_set(action->flags, pe_action_runnable)) { pe_rsc_trace(rsc, "Action %s is runnable", action->uuid); //pe_action_set_reason(action, NULL, TRUE); pe_set_action_bit(action, pe_action_runnable); } if (save_action) { switch (a_task) { case stop_rsc: set_bit(rsc->flags, pe_rsc_stopping); break; case start_rsc: clear_bit(rsc->flags, pe_rsc_starting); if (is_set(action->flags, pe_action_runnable)) { set_bit(rsc->flags, pe_rsc_starting); } break; default: break; } } } free(key); return action; } static const char * unpack_operation_on_fail(action_t * action) { const char *value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ON_FAIL); if (safe_str_eq(action->task, CRMD_ACTION_STOP) && safe_str_eq(value, "standby")) { crm_config_err("on-fail=standby is not allowed for stop actions: %s", action->rsc->id); return NULL; } else if (safe_str_eq(action->task, CRMD_ACTION_DEMOTE) && !value) { /* demote on_fail defaults to master monitor value if present */ xmlNode *operation = NULL; const char *name = NULL; const char *role = NULL; const char *on_fail = NULL; const char *interval_spec = NULL; const char *enabled = NULL; CRM_CHECK(action->rsc != NULL, return NULL); for (operation = __xml_first_child_element(action->rsc->ops_xml); operation && !value; operation = __xml_next_element(operation)) { if (!crm_str_eq((const char *)operation->name, "op", TRUE)) { continue; } name = crm_element_value(operation, "name"); role = crm_element_value(operation, "role"); on_fail = crm_element_value(operation, XML_OP_ATTR_ON_FAIL); enabled = crm_element_value(operation, "enabled"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); if (!on_fail) { continue; } else if (enabled && !crm_is_true(enabled)) { continue; } else if (safe_str_neq(name, "monitor") || safe_str_neq(role, "Master")) { continue; } else if (crm_parse_interval_spec(interval_spec) == 0) { continue; } value = on_fail; } } return value; } static xmlNode * find_min_interval_mon(resource_t * rsc, gboolean include_disabled) { guint interval_ms = 0; guint min_interval_ms = G_MAXUINT; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; xmlNode *op = NULL; 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)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } if (safe_str_neq(name, RSC_STATUS)) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); if (interval_ms && (interval_ms < min_interval_ms)) { min_interval_ms = interval_ms; op = operation; } } } return op; } static int unpack_start_delay(const char *value, GHashTable *meta) { int start_delay = 0; if (value != NULL) { start_delay = crm_get_msec(value); if (start_delay < 0) { start_delay = 0; } if (meta) { g_hash_table_replace(meta, strdup(XML_OP_ATTR_START_DELAY), crm_itoa(start_delay)); } } return start_delay; } // true if value contains valid, non-NULL interval origin for recurring op static bool unpack_interval_origin(const char *value, xmlNode *xml_obj, guint interval_ms, crm_time_t *now, long long *start_delay) { long long result = 0; guint interval_sec = interval_ms / 1000; crm_time_t *origin = NULL; // Ignore unspecified values and non-recurring operations if ((value == NULL) || (interval_ms == 0) || (now == NULL)) { return false; } // Parse interval origin from text origin = crm_time_new(value); if (origin == NULL) { crm_config_err("Operation '%s' contains invalid " XML_OP_ATTR_ORIGIN " '%s'", (ID(xml_obj)? ID(xml_obj) : "(unspecified)"), value); return false; } // Get seconds since origin (negative if origin is in the future) result = crm_time_get_seconds(now) - crm_time_get_seconds(origin); crm_time_free(origin); // Calculate seconds from closest interval to now result = result % interval_sec; // Calculate seconds remaining until next interval result = ((result <= 0)? 0 : interval_sec) - result; crm_info("Calculated a start delay of %llds for operation '%s'", result, (ID(xml_obj)? ID(xml_obj) : "(unspecified)")); if (start_delay != NULL) { *start_delay = result * 1000; // milliseconds } return true; } static int unpack_timeout(const char *value) { int timeout = crm_get_msec(value); if (timeout < 0) { timeout = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } return timeout; } int pe_get_configured_timeout(resource_t *rsc, const char *action, pe_working_set_t *data_set) { xmlNode *child = NULL; const char *timeout = NULL; int timeout_ms = 0; for (child = first_named_child(rsc->ops_xml, XML_ATTR_OP); child != NULL; child = crm_next_same_xml(child)) { if (safe_str_eq(action, crm_element_value(child, XML_NVPAIR_ATTR_NAME))) { timeout = crm_element_value(child, XML_ATTR_TIMEOUT); break; } } if (timeout == NULL && data_set->op_defaults) { GHashTable *action_meta = crm_str_table_new(); pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, NULL, action_meta, NULL, FALSE, data_set); timeout = g_hash_table_lookup(action_meta, XML_ATTR_TIMEOUT); } // @TODO check meta-attributes (including versioned meta-attributes) // @TODO maybe use min-interval monitor timeout as default for monitors timeout_ms = crm_get_msec(timeout); if (timeout_ms < 0) { timeout_ms = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } return timeout_ms; } #if ENABLE_VERSIONED_ATTRS static void unpack_versioned_meta(xmlNode *versioned_meta, xmlNode *xml_obj, guint interval_ms, crm_time_t *now) { xmlNode *attrs = NULL; xmlNode *attr = NULL; for (attrs = __xml_first_child_element(versioned_meta); attrs != NULL; attrs = __xml_next_element(attrs)) { for (attr = __xml_first_child_element(attrs); attr != NULL; attr = __xml_next_element(attr)) { const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); if (safe_str_eq(name, XML_OP_ATTR_START_DELAY)) { int start_delay = unpack_start_delay(value, NULL); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } else if (safe_str_eq(name, XML_OP_ATTR_ORIGIN)) { long long start_delay = 0; if (unpack_interval_origin(value, xml_obj, interval_ms, now, &start_delay)) { crm_xml_add(attr, XML_NVPAIR_ATTR_NAME, XML_OP_ATTR_START_DELAY); crm_xml_add_ll(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } } else if (safe_str_eq(name, XML_ATTR_TIMEOUT)) { int timeout = unpack_timeout(value); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, timeout); } } } } #endif /*! * \brief Unpack operation XML into an action structure * * Unpack an operation's meta-attributes (normalizing the interval, timeout, * and start delay values as integer milliseconds), requirements, and * failure policy. * * \param[in,out] action Action to unpack into * \param[in] xml_obj Operation XML (or NULL if all defaults) * \param[in] container Resource that contains affected resource, if any * \param[in] data_set Cluster state */ void unpack_operation(action_t * action, xmlNode * xml_obj, resource_t * container, pe_working_set_t * data_set) { guint interval_ms = 0; int timeout = 0; char *value_ms = NULL; const char *value = NULL; const char *field = NULL; char *default_timeout = NULL; #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t *rsc_details = NULL; #endif CRM_CHECK(action && action->rsc, return); // Cluster-wide pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, NULL, action->meta, NULL, FALSE, data_set); // Probe timeouts default differently, so handle timeout default later default_timeout = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT); if (default_timeout) { default_timeout = strdup(default_timeout); g_hash_table_remove(action->meta, XML_ATTR_TIMEOUT); } if (xml_obj) { xmlAttrPtr xIter = NULL; // take precedence over defaults pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_META_SETS, NULL, action->meta, NULL, TRUE, data_set); #if ENABLE_VERSIONED_ATTRS rsc_details = pe_rsc_action_details(action); pe_unpack_versioned_attributes(data_set->input, xml_obj, XML_TAG_ATTR_SETS, NULL, rsc_details->versioned_parameters, data_set->now, NULL); pe_unpack_versioned_attributes(data_set->input, xml_obj, XML_TAG_META_SETS, NULL, rsc_details->versioned_meta, data_set->now, NULL); #endif /* Anything set as an XML property has highest precedence. * This ensures we use the name and interval from the tag. */ for (xIter = xml_obj->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_obj, prop_name); g_hash_table_replace(action->meta, strdup(prop_name), strdup(prop_value)); } } g_hash_table_remove(action->meta, "id"); // Normalize interval to milliseconds field = XML_LRM_ATTR_INTERVAL; value = g_hash_table_lookup(action->meta, field); if (value != NULL) { interval_ms = crm_parse_interval_spec(value); } else if ((xml_obj == NULL) && !strcmp(action->task, RSC_STATUS)) { /* An orphaned recurring monitor will not have any XML. However, we * want the interval to be set, so the action can be properly detected * as a recurring monitor. Parse it from the key in this case. */ parse_op_key(action->uuid, NULL, NULL, &interval_ms); } if (interval_ms > 0) { value_ms = crm_strdup_printf("%u", interval_ms); g_hash_table_replace(action->meta, strdup(field), value_ms); } else if (value) { g_hash_table_remove(action->meta, field); } // Handle timeout default, now that we know the interval if (g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT)) { free(default_timeout); } else { // Probe timeouts default to minimum-interval monitor's if (safe_str_eq(action->task, RSC_STATUS) && (interval_ms == 0)) { xmlNode *min_interval_mon = find_min_interval_mon(action->rsc, FALSE); if (min_interval_mon) { value = crm_element_value(min_interval_mon, XML_ATTR_TIMEOUT); if (value) { crm_trace("\t%s defaults to minimum-interval monitor's timeout '%s'", action->uuid, value); free(default_timeout); default_timeout = strdup(value); } } } if (default_timeout) { g_hash_table_insert(action->meta, strdup(XML_ATTR_TIMEOUT), default_timeout); } } if (safe_str_neq(action->task, RSC_START) && safe_str_neq(action->task, RSC_PROMOTE)) { action->needs = rsc_req_nothing; value = "nothing (not start/promote)"; } else if (is_set(action->rsc->flags, pe_rsc_needs_fencing)) { action->needs = rsc_req_stonith; value = "fencing (resource)"; } else if (is_set(action->rsc->flags, pe_rsc_needs_quorum)) { action->needs = rsc_req_quorum; value = "quorum (resource)"; } else { action->needs = rsc_req_nothing; value = "nothing (resource)"; } pe_rsc_trace(action->rsc, "\tAction %s requires: %s", action->uuid, value); value = unpack_operation_on_fail(action); if (value == NULL) { } else if (safe_str_eq(value, "block")) { action->on_fail = action_fail_block; g_hash_table_insert(action->meta, strdup(XML_OP_ATTR_ON_FAIL), strdup("block")); value = "block"; // The above could destroy the original string } else if (safe_str_eq(value, "fence")) { action->on_fail = action_fail_fence; value = "node fencing"; if (is_set(data_set->flags, pe_flag_stonith_enabled) == FALSE) { crm_config_err("Specifying on_fail=fence and" " stonith-enabled=false makes no sense"); action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } } else if (safe_str_eq(value, "standby")) { action->on_fail = action_fail_standby; value = "node standby"; } else if (safe_str_eq(value, "ignore") || safe_str_eq(value, "nothing")) { action->on_fail = action_fail_ignore; value = "ignore"; } else if (safe_str_eq(value, "migrate")) { action->on_fail = action_fail_migrate; value = "force migration"; } else if (safe_str_eq(value, "stop")) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } else if (safe_str_eq(value, "restart")) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate)"; } else if (safe_str_eq(value, "restart-container")) { if (container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate)"; } else { value = NULL; } } else { pe_err("Resource %s: Unknown failure type (%s)", action->rsc->id, value); value = NULL; } /* defaults */ if (value == NULL && container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate) (default)"; /* For remote nodes, ensure that any failure that results in dropping an * active connection to the node results in fencing of the node. * * There are only two action failures that don't result in fencing. * 1. probes - probe failures are expected. * 2. start - a start failure indicates that an active connection does not already * exist. The user can set op on-fail=fence if they really want to fence start * failures. */ } else if (((value == NULL) || !is_set(action->rsc->flags, pe_rsc_managed)) && (pe__resource_is_remote_conn(action->rsc, data_set) && !(safe_str_eq(action->task, CRMD_ACTION_STATUS) && (interval_ms == 0)) && (safe_str_neq(action->task, CRMD_ACTION_START)))) { if (!is_set(action->rsc->flags, pe_rsc_managed)) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop unmanaged remote node (enforcing default)"; } else { if (is_set(data_set->flags, pe_flag_stonith_enabled)) { value = "fence remote node (default)"; } else { value = "recover remote node connection (default)"; } if (action->rsc->remote_reconnect_ms) { action->fail_role = RSC_ROLE_STOPPED; } action->on_fail = action_fail_reset_remote; } } else if (value == NULL && safe_str_eq(action->task, CRMD_ACTION_STOP)) { if (is_set(data_set->flags, pe_flag_stonith_enabled)) { action->on_fail = action_fail_fence; value = "resource fence (default)"; } else { action->on_fail = action_fail_block; value = "resource block (default)"; } } else if (value == NULL) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate) (default)"; } pe_rsc_trace(action->rsc, "\t%s failure handling: %s", action->task, value); value = NULL; if (xml_obj != NULL) { value = g_hash_table_lookup(action->meta, "role_after_failure"); if (value) { pe_warn_once(pe_wo_role_after, "Support for role_after_failure is deprecated and will be removed in a future release"); } } if (value != NULL && action->fail_role == RSC_ROLE_UNKNOWN) { action->fail_role = text2role(value); } /* defaults */ if (action->fail_role == RSC_ROLE_UNKNOWN) { if (safe_str_eq(action->task, CRMD_ACTION_PROMOTE)) { action->fail_role = RSC_ROLE_SLAVE; } else { action->fail_role = RSC_ROLE_STARTED; } } pe_rsc_trace(action->rsc, "\t%s failure results in: %s", action->task, role2text(action->fail_role)); value = g_hash_table_lookup(action->meta, XML_OP_ATTR_START_DELAY); if (value) { unpack_start_delay(value, action->meta); } else { long long start_delay = 0; value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ORIGIN); if (unpack_interval_origin(value, xml_obj, interval_ms, data_set->now, &start_delay)) { g_hash_table_replace(action->meta, strdup(XML_OP_ATTR_START_DELAY), crm_strdup_printf("%lld", start_delay)); } } value = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT); timeout = unpack_timeout(value); g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT), crm_itoa(timeout)); #if ENABLE_VERSIONED_ATTRS unpack_versioned_meta(rsc_details->versioned_meta, xml_obj, interval_ms, data_set->now); #endif } static xmlNode * find_rsc_op_entry_helper(resource_t * rsc, const char *key, gboolean include_disabled) { guint interval_ms = 0; gboolean do_retry = TRUE; char *local_key = NULL; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; char *match_key = NULL; xmlNode *op = NULL; xmlNode *operation = NULL; retry: 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)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); match_key = generate_op_key(rsc->id, name, interval_ms); if (safe_str_eq(key, match_key)) { op = operation; } free(match_key); if (rsc->clone_name) { match_key = generate_op_key(rsc->clone_name, name, interval_ms); if (safe_str_eq(key, match_key)) { op = operation; } free(match_key); } if (op != NULL) { free(local_key); return op; } } } free(local_key); if (do_retry == FALSE) { return NULL; } do_retry = FALSE; if (strstr(key, CRMD_ACTION_MIGRATE) || strstr(key, CRMD_ACTION_MIGRATED)) { local_key = generate_op_key(rsc->id, "migrate", 0); key = local_key; goto retry; } else if (strstr(key, "_notify_")) { local_key = generate_op_key(rsc->id, "notify", 0); key = local_key; goto retry; } return NULL; } xmlNode * find_rsc_op_entry(resource_t * rsc, const char *key) { return find_rsc_op_entry_helper(rsc, key, FALSE); } void print_node(const char *pre_text, node_t * node, gboolean details) { if (node == NULL) { crm_trace("%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } CRM_ASSERT(node->details); crm_trace("%s%s%sNode %s: (weight=%d, fixed=%s)", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": ", node->details->online ? "" : "Unavailable/Unclean ", node->details->uname, node->weight, node->fixed ? "True" : "False"); if (details) { char *pe_mutable = strdup("\t\t"); GListPtr gIter = node->details->running_rsc; crm_trace("\t\t===Node Attributes"); g_hash_table_foreach(node->details->attrs, print_str_str, pe_mutable); free(pe_mutable); crm_trace("\t\t=== Resources"); for (; gIter != NULL; gIter = gIter->next) { resource_t *rsc = (resource_t *) gIter->data; print_resource(LOG_TRACE, "\t\t", rsc, FALSE); } } } /* * Used by the HashTable for-loop */ void print_str_str(gpointer key, gpointer value, gpointer user_data) { crm_trace("%s%s %s ==> %s", user_data == NULL ? "" : (char *)user_data, user_data == NULL ? "" : ": ", (char *)key, (char *)value); } void print_resource(int log_level, const char *pre_text, resource_t * rsc, gboolean details) { long options = pe_print_log | pe_print_pending; if (rsc == NULL) { do_crm_log(log_level - 1, "%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } if (details) { options |= pe_print_details; } rsc->fns->print(rsc, pre_text, options, &log_level); } void pe_free_action(action_t * action) { if (action == NULL) { return; } g_list_free_full(action->actions_before, free); /* action_wrapper_t* */ g_list_free_full(action->actions_after, free); /* action_wrapper_t* */ if (action->extra) { g_hash_table_destroy(action->extra); } if (action->meta) { g_hash_table_destroy(action->meta); } #if ENABLE_VERSIONED_ATTRS if (action->rsc) { pe_free_rsc_action_details(action); } #endif free(action->cancel_task); free(action->reason); free(action->task); free(action->uuid); free(action->node); free(action); } GListPtr find_recurring_actions(GListPtr input, node_t * not_on_node) { const char *value = NULL; GListPtr result = NULL; GListPtr gIter = input; CRM_CHECK(input != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { action_t *action = (action_t *) gIter->data; value = g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS); if (value == NULL) { /* skip */ } else if (safe_str_eq(value, "0")) { /* skip */ } else if (safe_str_eq(CRMD_ACTION_CANCEL, action->task)) { /* skip */ } else if (not_on_node == NULL) { crm_trace("(null) Found: %s", action->uuid); result = g_list_prepend(result, action); } else if (action->node == NULL) { /* skip */ } else if (action->node->details != not_on_node->details) { crm_trace("Found: %s", action->uuid); result = g_list_prepend(result, action); } } return result; } enum action_tasks get_complex_task(resource_t * rsc, const char *name, gboolean allow_non_atomic) { enum action_tasks task = text2task(name); if (rsc == NULL) { return task; } else if (allow_non_atomic == FALSE || rsc->variant == pe_native) { switch (task) { case stopped_rsc: case started_rsc: case action_demoted: case action_promoted: crm_trace("Folding %s back into its atomic counterpart for %s", name, rsc->id); return task - 1; break; default: break; } } return task; } action_t * find_first_action(GListPtr input, const char *uuid, const char *task, node_t * on_node) { GListPtr gIter = NULL; CRM_CHECK(uuid || task, return NULL); for (gIter = input; gIter != NULL; gIter = gIter->next) { action_t *action = (action_t *) gIter->data; if (uuid != NULL && safe_str_neq(uuid, action->uuid)) { continue; } else if (task != NULL && safe_str_neq(task, action->task)) { continue; } else if (on_node == NULL) { return action; } else if (action->node == NULL) { continue; } else if (on_node->details == action->node->details) { return action; } } return NULL; } GListPtr find_actions(GListPtr input, const char *key, const node_t *on_node) { GListPtr gIter = input; GListPtr result = NULL; CRM_CHECK(key != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { action_t *action = (action_t *) gIter->data; if (safe_str_neq(key, action->uuid)) { crm_trace("%s does not match action %s", key, action->uuid); continue; } else if (on_node == NULL) { crm_trace("Action %s matches (ignoring node)", key); result = g_list_prepend(result, action); } else if (action->node == NULL) { crm_trace("Action %s matches (unallocated, assigning to %s)", key, on_node->details->uname); action->node = node_copy(on_node); result = g_list_prepend(result, action); } else if (on_node->details == action->node->details) { crm_trace("Action %s on %s matches", key, on_node->details->uname); result = g_list_prepend(result, action); } else { crm_trace("Action %s on node %s does not match requested node %s", key, action->node->details->uname, on_node->details->uname); } } return result; } GList * find_actions_exact(GList *input, const char *key, const pe_node_t *on_node) { GList *result = NULL; CRM_CHECK(key != NULL, return NULL); if (on_node == NULL) { crm_trace("Not searching for action %s because node not specified", key); return NULL; } for (GList *gIter = input; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (action->node == NULL) { crm_trace("Skipping comparison of %s vs action %s without node", key, action->uuid); } else if (safe_str_neq(key, action->uuid)) { crm_trace("Desired action %s doesn't match %s", key, action->uuid); } else if (safe_str_neq(on_node->details->id, action->node->details->id)) { crm_trace("Action %s desired node ID %s doesn't match %s", key, on_node->details->id, action->node->details->id); } else { crm_trace("Action %s matches", key); result = g_list_prepend(result, action); } } return result; } /*! * \brief Find all actions of given type for a resource * * \param[in] rsc Resource to search * \param[in] node Find only actions scheduled on this node * \param[in] task Action name to search for * \param[in] require_node If TRUE, NULL node or action node will not match * * \return List of actions found (or NULL if none) * \note If node is not NULL and require_node is FALSE, matching actions * without a node will be assigned to node. */ GList * pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node, const char *task, bool require_node) { GList *result = NULL; char *key = generate_op_key(rsc->id, task, 0); if (require_node) { result = find_actions_exact(rsc->actions, key, node); } else { result = find_actions(rsc->actions, key, node); } free(key); return result; } static void resource_node_score(resource_t * rsc, node_t * node, int score, const char *tag) { node_t *match = NULL; if ((rsc->exclusive_discover || (node->rsc_discover_mode == pe_discover_never)) && safe_str_eq(tag, "symmetric_default")) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child_rsc = (resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } pe_rsc_trace(rsc, "Setting %s for %s on %s: %d", tag, rsc->id, node->details->uname, score); match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match == NULL) { match = node_copy(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match); } match->weight = merge_weights(match->weight, score); } void resource_location(resource_t * rsc, node_t * node, int score, const char *tag, pe_working_set_t * data_set) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (data_set != NULL) { GListPtr gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { node_t *node_iter = (node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) { resource_node_score(rsc, node_iter, score, tag); } } if (node == NULL && score == -INFINITY) { if (rsc->allocated_to) { crm_info("Deallocating %s from %s", rsc->id, rsc->allocated_to->details->uname); free(rsc->allocated_to); rsc->allocated_to = NULL; } } } #define sort_return(an_int, why) do { \ free(a_uuid); \ free(b_uuid); \ crm_trace("%s (%d) %c %s (%d) : %s", \ a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \ b_xml_id, b_call_id, why); \ return an_int; \ } while(0) gint sort_op_by_callid(gconstpointer a, gconstpointer b) { int a_call_id = -1; int b_call_id = -1; char *a_uuid = NULL; char *b_uuid = NULL; const xmlNode *xml_a = a; const xmlNode *xml_b = b; const char *a_xml_id = crm_element_value(xml_a, XML_ATTR_ID); const char *b_xml_id = crm_element_value(xml_b, XML_ATTR_ID); if (safe_str_eq(a_xml_id, b_xml_id)) { /* We have duplicate lrm_rsc_op entries in the status * section which is unlikely to be a good thing * - we can handle it easily enough, but we need to get * to the bottom of why it's happening. */ pe_err("Duplicate lrm_rsc_op entries named %s", a_xml_id); sort_return(0, "duplicate"); } crm_element_value_int(xml_a, XML_LRM_ATTR_CALLID, &a_call_id); crm_element_value_int(xml_b, XML_LRM_ATTR_CALLID, &b_call_id); if (a_call_id == -1 && b_call_id == -1) { /* both are pending ops so it doesn't matter since * stops are never pending */ sort_return(0, "pending"); } else if (a_call_id >= 0 && a_call_id < b_call_id) { sort_return(-1, "call id"); } else if (b_call_id >= 0 && a_call_id > b_call_id) { sort_return(1, "call id"); } else if (b_call_id >= 0 && a_call_id == b_call_id) { /* * The op and last_failed_op are the same * Order on last-rc-change */ time_t last_a = -1; time_t last_b = -1; crm_element_value_epoch(xml_a, XML_RSC_OP_LAST_CHANGE, &last_a); crm_element_value_epoch(xml_b, XML_RSC_OP_LAST_CHANGE, &last_b); crm_trace("rc-change: %lld vs %lld", (long long) last_a, (long long) last_b); if (last_a >= 0 && last_a < last_b) { sort_return(-1, "rc-change"); } else if (last_b >= 0 && last_a > last_b) { sort_return(1, "rc-change"); } sort_return(0, "rc-change"); } else { /* One of the inputs is a pending operation * Attempt to use XML_ATTR_TRANSITION_MAGIC to determine its age relative to the other */ int a_id = -1; int b_id = -1; const char *a_magic = crm_element_value(xml_a, XML_ATTR_TRANSITION_MAGIC); const char *b_magic = crm_element_value(xml_b, XML_ATTR_TRANSITION_MAGIC); CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic")); if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic a"); } if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic b"); } /* try to determine the relative age of the operation... * some pending operations (e.g. a start) may have been superseded * by a subsequent stop * * [a|b]_id == -1 means it's a shutdown operation and _always_ comes last */ if (safe_str_neq(a_uuid, b_uuid) || a_id == b_id) { /* * some of the logic in here may be redundant... * * if the UUID from the TE doesn't match then one better * be a pending operation. * pending operations don't survive between elections and joins * because we query the LRM directly */ if (b_call_id == -1) { sort_return(-1, "transition + call"); } else if (a_call_id == -1) { sort_return(1, "transition + call"); } } else if ((a_id >= 0 && a_id < b_id) || b_id == -1) { sort_return(-1, "transition"); } else if ((b_id >= 0 && a_id > b_id) || a_id == -1) { sort_return(1, "transition"); } } /* we should never end up here */ CRM_CHECK(FALSE, sort_return(0, "default")); } time_t get_effective_time(pe_working_set_t * data_set) { if(data_set) { if (data_set->now == NULL) { crm_trace("Recording a new 'now'"); data_set->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(data_set->now); } crm_trace("Defaulting to 'now'"); return time(NULL); } gboolean get_target_role(resource_t * rsc, enum rsc_role_e * role) { enum rsc_role_e local_role = RSC_ROLE_UNKNOWN; const char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (value == NULL || safe_str_eq("started", value) || safe_str_eq("default", value)) { return FALSE; } local_role = text2role(value); if (local_role == RSC_ROLE_UNKNOWN) { crm_config_err("%s: Unknown value for %s: %s", rsc->id, XML_RSC_ATTR_TARGET_ROLE, value); return FALSE; } else if (local_role > RSC_ROLE_STARTED) { if (is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) { if (local_role > RSC_ROLE_SLAVE) { /* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */ return FALSE; } } else { crm_config_err("%s is not part of a promotable clone resource, a %s of '%s' makes no sense", rsc->id, XML_RSC_ATTR_TARGET_ROLE, value); return FALSE; } } *role = local_role; return TRUE; } gboolean order_actions(action_t * lh_action, action_t * rh_action, enum pe_ordering order) { GListPtr gIter = NULL; action_wrapper_t *wrapper = NULL; GListPtr list = NULL; if (order == pe_order_none) { return FALSE; } if (lh_action == NULL || rh_action == NULL) { return FALSE; } crm_trace("Ordering Action %s before %s", lh_action->uuid, rh_action->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ CRM_ASSERT(lh_action != rh_action); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = lh_action->actions_after; for (; gIter != NULL; gIter = gIter->next) { action_wrapper_t *after = (action_wrapper_t *) gIter->data; if (after->action == rh_action && (after->type & order)) { return FALSE; } } wrapper = calloc(1, sizeof(action_wrapper_t)); wrapper->action = rh_action; wrapper->type = order; list = lh_action->actions_after; list = g_list_prepend(list, wrapper); lh_action->actions_after = list; wrapper = NULL; /* order |= pe_order_implies_then; */ /* order ^= pe_order_implies_then; */ wrapper = calloc(1, sizeof(action_wrapper_t)); wrapper->action = lh_action; wrapper->type = order; list = rh_action->actions_before; list = g_list_prepend(list, wrapper); rh_action->actions_before = list; return TRUE; } action_t * get_pseudo_op(const char *name, pe_working_set_t * data_set) { action_t *op = NULL; if(data_set->singletons) { op = g_hash_table_lookup(data_set->singletons, name); } if (op == NULL) { op = custom_action(NULL, strdup(name), name, NULL, TRUE, TRUE, data_set); set_bit(op->flags, pe_action_pseudo); set_bit(op->flags, pe_action_runnable); } return op; } void destroy_ticket(gpointer data) { ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } ticket_t * ticket_new(const char *ticket_id, pe_working_set_t * data_set) { ticket_t *ticket = NULL; if (ticket_id == NULL || strlen(ticket_id) == 0) { return NULL; } if (data_set->tickets == NULL) { data_set->tickets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_ticket); } ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(ticket_t)); if (ticket == NULL) { crm_err("Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creaing ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->granted = FALSE; ticket->last_granted = -1; ticket->standby = FALSE; ticket->state = crm_str_table_new(); g_hash_table_insert(data_set->tickets, strdup(ticket->id), ticket); } return ticket; } static void filter_parameters(xmlNode * param_set, const char *param_string, bool need_present) { if (param_set && param_string) { xmlAttrPtr xIter = param_set->properties; while (xIter) { const char *prop_name = (const char *)xIter->name; char *name = crm_strdup_printf(" %s ", prop_name); char *match = strstr(param_string, name); free(name); // Do now, because current entry might get removed below xIter = xIter->next; if (need_present && match == NULL) { crm_trace("%s not found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } else if (need_present == FALSE && match) { crm_trace("%s found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } } } } #if ENABLE_VERSIONED_ATTRS static void append_versioned_params(xmlNode *versioned_params, const char *ra_version, xmlNode *params) { GHashTable *hash = pe_unpack_versioned_parameters(versioned_params, ra_version); char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { crm_xml_add(params, key, value); } g_hash_table_destroy(hash); } #endif /*! * \internal * \brief Calculate action digests and store in node's digest cache * * \param[in] rsc Resource that action was for * \param[in] task Name of action performed * \param[in] key Action's task key * \param[in] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] calc_secure Whether to calculate secure digest * \param[in] data_set Cluster working set * * \return Pointer to node's digest cache entry */ static op_digest_cache_t * rsc_action_digest(pe_resource_t *rsc, const char *task, const char *key, pe_node_t *node, xmlNode *xml_op, bool calc_secure, pe_working_set_t *data_set) { op_digest_cache_t *data = NULL; data = g_hash_table_lookup(node->details->digest_cache, key); if (data == NULL) { GHashTable *local_rsc_params = crm_str_table_new(); action_t *action = custom_action(rsc, strdup(key), task, node, TRUE, FALSE, data_set); #if ENABLE_VERSIONED_ATTRS xmlNode *local_versioned_params = create_xml_node(NULL, XML_TAG_RSC_VER_ATTRS); const char *ra_version = NULL; #endif const char *op_version; const char *restart_list = NULL; const char *secure_list = " passwd password "; data = calloc(1, sizeof(op_digest_cache_t)); CRM_ASSERT(data != NULL); get_rsc_attributes(local_rsc_params, rsc, node, data_set); #if ENABLE_VERSIONED_ATTRS pe_get_versioned_attributes(local_versioned_params, rsc, node, data_set); #endif data->params_all = create_xml_node(NULL, XML_TAG_PARAMS); // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside if (pe__add_bundle_remote_name(rsc, data->params_all, XML_RSC_ATTR_REMOTE_RA_ADDR)) { crm_trace("Set address for bundle connection %s (on %s)", rsc->id, node->details->uname); } g_hash_table_foreach(local_rsc_params, hash2field, data->params_all); g_hash_table_foreach(action->extra, hash2field, data->params_all); g_hash_table_foreach(rsc->parameters, hash2field, data->params_all); g_hash_table_foreach(action->meta, hash2metafield, data->params_all); if(xml_op) { secure_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_SECURE); restart_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_RESTART); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); #if ENABLE_VERSIONED_ATTRS ra_version = crm_element_value(xml_op, XML_ATTR_RA_VERSION); #endif } else { op_version = CRM_FEATURE_SET; } #if ENABLE_VERSIONED_ATTRS append_versioned_params(local_versioned_params, ra_version, data->params_all); append_versioned_params(rsc->versioned_parameters, ra_version, data->params_all); { pe_rsc_action_details_t *details = pe_rsc_action_details(action); append_versioned_params(details->versioned_parameters, ra_version, data->params_all); } #endif filter_action_parameters(data->params_all, op_version); g_hash_table_destroy(local_rsc_params); pe_free_action(action); data->digest_all_calc = calculate_operation_digest(data->params_all, op_version); if (calc_secure) { data->params_secure = copy_xml(data->params_all); if(secure_list) { filter_parameters(data->params_secure, secure_list, FALSE); } data->digest_secure_calc = calculate_operation_digest(data->params_secure, op_version); } if(xml_op && crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST) != NULL) { data->params_restart = copy_xml(data->params_all); if (restart_list) { filter_parameters(data->params_restart, restart_list, TRUE); } data->digest_restart_calc = calculate_operation_digest(data->params_restart, op_version); } g_hash_table_insert(node->details->digest_cache, strdup(key), data); } return data; } op_digest_cache_t * rsc_action_digest_cmp(resource_t * rsc, xmlNode * xml_op, node_t * node, pe_working_set_t * data_set) { op_digest_cache_t *data = NULL; char *key = NULL; guint interval_ms = 0; const char *op_version; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *interval_ms_s = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS); const char *digest_all; const char *digest_restart; CRM_ASSERT(node != NULL); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); digest_all = crm_element_value(xml_op, XML_LRM_ATTR_OP_DIGEST); digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST); interval_ms = crm_parse_ms(interval_ms_s); key = generate_op_key(rsc->id, task, interval_ms); data = rsc_action_digest(rsc, task, key, node, xml_op, is_set(data_set->flags, pe_flag_sanitized), data_set); data->rc = RSC_DIGEST_MATCH; if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (restart:%s) %s", key, node->details->uname, crm_str(digest_restart), data->digest_restart_calc, op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_RESTART; } else if (digest_all == NULL) { /* it is unknown what the previous op digest was */ data->rc = RSC_DIGEST_UNKNOWN; } else if (strcmp(digest_all, data->digest_all_calc) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (%s:%s) %s", key, node->details->uname, crm_str(digest_all), data->digest_all_calc, (interval_ms > 0)? "reschedule" : "reload", op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_ALL; } free(key); return data; } /*! * \internal * \brief Create an unfencing summary for use in special node attribute * * Create a string combining a fence device's resource ID, agent type, and * parameter digest (whether for all parameters or just non-private parameters). * This can be stored in a special node attribute, allowing us to detect changes * in either the agent type or parameters, to know whether unfencing must be * redone or can be safely skipped when the device's history is cleaned. * * \param[in] rsc_id Fence device resource ID * \param[in] agent_type Fence device agent * \param[in] param_digest Fence device parameter digest * * \return Newly allocated string with unfencing digest * \note The caller is responsible for freeing the result. */ static inline char * create_unfencing_summary(const char *rsc_id, const char *agent_type, const char *param_digest) { return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest); } /*! * \internal * \brief Check whether a node can skip unfencing * * Check whether a fence device's current definition matches a node's * stored summary of when it was last unfenced by the device. * * \param[in] rsc_id Fence device's resource ID * \param[in] agent Fence device's agent type * \param[in] digest_calc Fence device's current parameter digest * \param[in] node_summary Value of node's special unfencing node attribute * (a comma-separated list of unfencing summaries for * all devices that have unfenced this node) * * \return TRUE if digest matches, FALSE otherwise */ static bool unfencing_digest_matches(const char *rsc_id, const char *agent, const char *digest_calc, const char *node_summary) { bool matches = FALSE; if (rsc_id && agent && digest_calc && node_summary) { char *search_secure = create_unfencing_summary(rsc_id, agent, digest_calc); /* The digest was calculated including the device ID and agent, * so there is no risk of collision using strstr(). */ matches = (strstr(node_summary, search_secure) != NULL); crm_trace("Calculated unfencing digest '%s' %sfound in '%s'", search_secure, matches? "" : "not ", node_summary); free(search_secure); } return matches; } /* Magic string to use as action name for digest cache entries used for * unfencing checks. This is not a real action name (i.e. "on"), so * check_action_definition() won't confuse these entries with real actions. */ #define STONITH_DIGEST_TASK "stonith-on" /*! * \internal * \brief Calculate fence device digests and digest comparison result * * \param[in] rsc Fence device resource * \param[in] agent Fence device's agent type * \param[in] node Node with digest cache to use * \param[in] data_set Cluster working set * * \return Node's digest cache entry */ static op_digest_cache_t * fencing_action_digest_cmp(pe_resource_t *rsc, const char *agent, pe_node_t *node, pe_working_set_t *data_set) { const char *node_summary = NULL; // Calculate device's current parameter digests char *key = generate_op_key(rsc->id, STONITH_DIGEST_TASK, 0); op_digest_cache_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, key, node, NULL, TRUE, data_set); free(key); // Check whether node has special unfencing summary node attribute node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_ALL); if (node_summary == NULL) { data->rc = RSC_DIGEST_UNKNOWN; return data; } // Check whether full parameter digest matches if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; return data; } // Check whether secure parameter digest matches node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_SECURE); if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; if (is_set(data_set->flags, pe_flag_stdout)) { printf("Only 'private' parameters to %s for unfencing %s changed\n", rsc->id, node->details->uname); } return data; } // Parameters don't match data->rc = RSC_DIGEST_ALL; if (is_set(data_set->flags, (pe_flag_sanitized|pe_flag_stdout)) && data->digest_secure_calc) { char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); printf("Parameters to %s for unfencing %s changed, try '%s'\n", rsc->id, node->details->uname, digest); free(digest); } return data; } const char *rsc_printable_id(resource_t *rsc) { if (is_not_set(rsc->flags, pe_rsc_unique)) { return ID(rsc->xml); } return rsc->id; } void clear_bit_recursive(resource_t * rsc, unsigned long long flag) { GListPtr gIter = rsc->children; clear_bit(rsc->flags, flag); for (; gIter != NULL; gIter = gIter->next) { resource_t *child_rsc = (resource_t *) gIter->data; clear_bit_recursive(child_rsc, flag); } } void set_bit_recursive(resource_t * rsc, unsigned long long flag) { GListPtr gIter = rsc->children; set_bit(rsc->flags, flag); for (; gIter != NULL; gIter = gIter->next) { resource_t *child_rsc = (resource_t *) gIter->data; set_bit_recursive(child_rsc, flag); } } static GListPtr find_unfencing_devices(GListPtr candidates, GListPtr matches) { for (GListPtr gIter = candidates; gIter != NULL; gIter = gIter->next) { resource_t *candidate = gIter->data; const char *provides = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_PROVIDES); const char *requires = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_REQUIRES); if(candidate->children) { matches = find_unfencing_devices(candidate->children, matches); } else if (is_not_set(candidate->flags, pe_rsc_fence_device)) { continue; } else if (crm_str_eq(provides, "unfencing", FALSE) || crm_str_eq(requires, "unfencing", FALSE)) { matches = g_list_prepend(matches, candidate); } } return matches; } action_t * pe_fence_op(node_t * node, const char *op, bool optional, const char *reason, pe_working_set_t * data_set) { char *op_key = NULL; action_t *stonith_op = NULL; if(op == NULL) { op = data_set->stonith_action; } op_key = crm_strdup_printf("%s-%s-%s", CRM_OP_FENCE, node->details->uname, op); if(data_set->singletons) { stonith_op = g_hash_table_lookup(data_set->singletons, op_key); } if(stonith_op == NULL) { stonith_op = custom_action(NULL, op_key, CRM_OP_FENCE, node, TRUE, TRUE, data_set); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET, node->details->uname); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET_UUID, node->details->id); add_hash_param(stonith_op->meta, "stonith_action", op); if (pe__is_guest_or_remote_node(node) && is_set(data_set->flags, pe_flag_enable_unfencing)) { /* Extra work to detect device changes on remotes * * We may do this for all nodes in the future, but for now * the check_action_definition() based stuff works fine. */ long max = 1024; long digests_all_offset = 0; long digests_secure_offset = 0; - char *digests_all = malloc(max); - char *digests_secure = malloc(max); + char *digests_all = calloc(max, sizeof(char)); + char *digests_secure = calloc(max, sizeof(char)); GListPtr matches = find_unfencing_devices(data_set->resources, NULL); for (GListPtr gIter = matches; gIter != NULL; gIter = gIter->next) { resource_t *match = gIter->data; const char *agent = g_hash_table_lookup(match->meta, XML_ATTR_TYPE); op_digest_cache_t *data = NULL; data = fencing_action_digest_cmp(match, agent, node, data_set); if(data->rc == RSC_DIGEST_ALL) { optional = FALSE; crm_notice("Unfencing %s (remote): because the definition of %s changed", node->details->uname, match->id); if (is_set(data_set->flags, pe_flag_stdout)) { fprintf(stdout, " notice: Unfencing %s (remote): because the definition of %s changed\n", node->details->uname, match->id); } } digests_all_offset += snprintf( digests_all+digests_all_offset, max-digests_all_offset, "%s:%s:%s,", match->id, agent, data->digest_all_calc); digests_secure_offset += snprintf( digests_secure+digests_secure_offset, max-digests_secure_offset, "%s:%s:%s,", match->id, agent, data->digest_secure_calc); } g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_ALL), digests_all); g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_SECURE), digests_secure); } } else { free(op_key); } if(optional == FALSE && pe_can_fence(data_set, node)) { pe_action_required(stonith_op, NULL, reason); } else if(reason && stonith_op->reason == NULL) { stonith_op->reason = strdup(reason); } return stonith_op; } void trigger_unfencing( resource_t * rsc, node_t *node, const char *reason, action_t *dependency, pe_working_set_t * data_set) { if(is_not_set(data_set->flags, pe_flag_enable_unfencing)) { /* No resources require it */ return; } else if (rsc != NULL && is_not_set(rsc->flags, pe_rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { action_t *unfence = pe_fence_op(node, "on", FALSE, reason, data_set); if(dependency) { order_actions(unfence, dependency, pe_order_optional); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { trigger_unfencing(rsc, node, reason, dependency, data_set); } } } } gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref) { tag_t *tag = NULL; GListPtr gIter = NULL; gboolean is_existing = FALSE; CRM_CHECK(tags && tag_name && obj_ref, return FALSE); tag = g_hash_table_lookup(tags, tag_name); if (tag == NULL) { tag = calloc(1, sizeof(tag_t)); if (tag == NULL) { return FALSE; } tag->id = strdup(tag_name); tag->refs = NULL; g_hash_table_insert(tags, strdup(tag_name), tag); } for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) { const char *existing_ref = (const char *) gIter->data; if (crm_str_eq(existing_ref, obj_ref, TRUE)){ is_existing = TRUE; break; } } if (is_existing == FALSE) { tag->refs = g_list_append(tag->refs, strdup(obj_ref)); crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref); } return TRUE; } 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) { bool unset = FALSE; bool update = FALSE; const char *change = NULL; if(is_set(flags, pe_action_runnable)) { unset = TRUE; change = "unrunnable"; } else if(is_set(flags, pe_action_optional)) { unset = TRUE; change = "required"; } else if(is_set(flags, pe_action_migrate_runnable)) { unset = TRUE; overwrite = TRUE; change = "unrunnable"; } else if(is_set(flags, pe_action_dangle)) { change = "dangling"; } else if(is_set(flags, pe_action_requires_any)) { change = "required"; } else { crm_err("Unknown flag change to %x by %s: 0x%s", flags, action->uuid, (reason? reason->uuid : "0")); } if(unset) { if(is_set(action->flags, flags)) { action->flags = crm_clear_bit(function, line, action->uuid, action->flags, flags); update = TRUE; } } else { if(is_not_set(action->flags, flags)) { action->flags = crm_set_bit(function, line, action->uuid, action->flags, flags); update = TRUE; } } if((change && update) || text) { char *reason_text = NULL; if(reason == NULL) { pe_action_set_reason(action, text, overwrite); } else if(reason->rsc == NULL) { reason_text = crm_strdup_printf("%s %s%c %s", change, reason->task, text?':':0, text?text:""); } else { reason_text = crm_strdup_printf("%s %s %s%c %s", change, reason->rsc->id, reason->task, text?':':0, text?text:"NA"); } if(reason_text && action->rsc != reason->rsc) { pe_action_set_reason(action, reason_text, overwrite); } free(reason_text); } } void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite) { if(action->reason && overwrite) { pe_rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'", action->uuid, action->reason, reason); free(action->reason); action->reason = NULL; } if(action->reason == NULL) { if(reason) { pe_rsc_trace(action->rsc, "Set %s reason to '%s'", action->uuid, reason); action->reason = strdup(reason); } else { action->reason = NULL; } } } /*! * \internal * \brief Check whether shutdown has been requested for a node * * \param[in] node Node to check * * \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise * \note This differs from simply using node->details->shutdown in that it can * be used before that has been determined (and in fact to determine it), * and it can also be used to distinguish requested shutdown from implicit * shutdown of remote nodes by virtue of their connection stopping. */ bool pe__shutdown_requested(pe_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); return shutdown && strcmp(shutdown, "0"); } /*! * \internal * \brief Update a data set's "recheck by" time * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] data_set Current working set */ void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set) { if ((recheck > get_effective_time(data_set)) && ((data_set->recheck_by == 0) || (data_set->recheck_by > recheck))) { data_set->recheck_by = recheck; } } /*! * \internal * \brief Wrapper for pe_unpack_nvpairs() using a cluster working set */ 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) { crm_time_t *next_change = crm_time_new_undefined(); pe_unpack_nvpairs(data_set->input, xml_obj, set_name, node_hash, hash, always_first, overwrite, data_set->now, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__update_recheck_time(recheck, data_set); } crm_time_free(next_change); }