Page MenuHomeClusterLabs Projects
Files F3686814

No OneTemporary
Actions

View Options

This file is larger than 256 KB, so syntax highlighting was skipped.
diff --git a/include/crm/msg_xml.h b/include/crm/msg_xml.h
index 8fefe0ccb1..3b2b117d0f 100644
--- a/include/crm/msg_xml.h
+++ b/include/crm/msg_xml.h
@@ -1,512 +1,513 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_MSG_XML__H
# define PCMK__CRM_MSG_XML__H
# include <crm/common/xml.h>
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
#include <crm/msg_xml_compat.h>
#endif
#ifdef __cplusplus
extern "C" {
#endif
/* This file defines constants for various XML syntax (mainly element and
* attribute names).
*
* For consistency, new constants should start with "PCMK_", followed by "XE"
* for XML element names, "XA" for XML attribute names, and "META" for meta
* attribute names. Old names that don't follow this policy should eventually be
* deprecated and replaced with names that do.
*/
/*
* XML elements
*/
#define PCMK_XE_DATE_EXPRESSION "date_expression"
#define PCMK_XE_OP_EXPRESSION "op_expression"
/* This has been deprecated as a CIB element (an alias for <clone> with
* "promotable" set to "true") since 2.0.0.
*/
#define PCMK_XE_PROMOTABLE_LEGACY "master"
#define PCMK_XE_RSC_EXPRESSION "rsc_expression"
/*
* XML attributes
*/
#define PCMK_XA_FORMAT "format"
/* These have been deprecated as CIB <clone> element attributes (aliases for
* "promoted-max" and "promoted-node-max") since 2.0.0.
*/
#define PCMK_XA_PROMOTED_MAX_LEGACY "master-max"
#define PCMK_XA_PROMOTED_NODE_MAX_LEGACY "master-node-max"
/*
* Meta attributes
*/
#define PCMK_META_CLONE_MAX "clone-max"
#define PCMK_META_CLONE_MIN "clone-min"
#define PCMK_META_CLONE_NODE_MAX "clone-node-max"
#define PCMK_META_ENABLED "enabled"
#define PCMK_META_FAILURE_TIMEOUT "failure-timeout"
#define PCMK_META_MIGRATION_THRESHOLD "migration-threshold"
#define PCMK_META_PROMOTED_MAX "promoted-max"
#define PCMK_META_PROMOTED_NODE_MAX "promoted-node-max"
/*
* Cluster options
*/
#define PCMK_OPT_BATCH_LIMIT "batch-limit"
#define PCMK_OPT_CLUSTER_DELAY "cluster-delay"
#define PCMK_OPT_CLUSTER_INFRASTRUCTURE "cluster-infrastructure"
#define PCMK_OPT_CLUSTER_IPC_LIMIT "cluster-ipc-limit"
#define PCMK_OPT_CLUSTER_NAME "cluster-name"
#define PCMK_OPT_CLUSTER_RECHECK_INTERVAL "cluster-recheck-interval"
#define PCMK_OPT_CONCURRENT_FENCING "concurrent-fencing"
#define PCMK_OPT_DC_DEADTIME "dc-deadtime"
#define PCMK_OPT_DC_VERSION "dc-version"
#define PCMK_OPT_ELECTION_TIMEOUT "election-timeout"
#define PCMK_OPT_ENABLE_ACL "enable-acl"
#define PCMK_OPT_ENABLE_STARTUP_PROBES "enable-startup-probes"
#define PCMK_OPT_FENCE_REACTION "fence-reaction"
#define PCMK_OPT_HAVE_WATCHDOG "have-watchdog"
#define PCMK_OPT_JOIN_FINALIZATION_TIMEOUT "join-finalization-timeout"
#define PCMK_OPT_JOIN_INTEGRATION_TIMEOUT "join-integration-timeout"
#define PCMK_OPT_LOAD_THRESHOLD "load-threshold"
#define PCMK_OPT_MAINTENANCE_MODE "maintenance-mode"
#define PCMK_OPT_MIGRATION_LIMIT "migration-limit"
#define PCMK_OPT_NO_QUORUM_POLICY "no-quorum-policy"
#define PCMK_OPT_NODE_ACTION_LIMIT "node-action-limit"
#define PCMK_OPT_NODE_PENDING_TIMEOUT "node-pending-timeout"
#define PCMK_OPT_PE_ERROR_SERIES_MAX "pe-error-series-max"
#define PCMK_OPT_PE_INPUT_SERIES_MAX "pe-input-series-max"
#define PCMK_OPT_PE_WARN_SERIES_MAX "pe-warn-series-max"
#define PCMK_OPT_PLACEMENT_STRATEGY "placement-strategy"
#define PCMK_OPT_PRIORITY_FENCING_DELAY "priority-fencing-delay"
#define PCMK_OPT_SHUTDOWN_ESCALATION "shutdown-escalation"
#define PCMK_OPT_SHUTDOWN_LOCK "shutdown-lock"
#define PCMK_OPT_SHUTDOWN_LOCK_LIMIT "shutdown-lock-limit"
#define PCMK_OPT_START_FAILURE_IS_FATAL "start-failure-is-fatal"
#define PCMK_OPT_STARTUP_FENCING "startup-fencing"
#define PCMK_OPT_STONITH_ACTION "stonith-action"
#define PCMK_OPT_STONITH_ENABLED "stonith-enabled"
#define PCMK_OPT_STONITH_MAX_ATTEMPTS "stonith-max-attempts"
#define PCMK_OPT_STONITH_TIMEOUT "stonith-timeout"
#define PCMK_OPT_STONITH_WATCHDOG_TIMEOUT "stonith-watchdog-timeout"
+#define PCMK_OPT_STOP_ALL_RESOURCES "stop-all-resources"
/*
* Older constants that don't follow current naming
*/
# ifndef F_ORIG
# define F_ORIG "src"
# endif
# ifndef F_SEQ
# define F_SEQ "seq"
# endif
# ifndef F_SUBTYPE
# define F_SUBTYPE "subt"
# endif
# ifndef F_TYPE
# define F_TYPE "t"
# endif
# ifndef F_CLIENTNAME
# define F_CLIENTNAME "cn"
# endif
# ifndef F_XML_TAGNAME
# define F_XML_TAGNAME "__name__"
# endif
# ifndef T_CRM
# define T_CRM "crmd"
# endif
# ifndef T_ATTRD
# define T_ATTRD "attrd"
# endif
# define CIB_OPTIONS_FIRST "cib-bootstrap-options"
# define F_CRM_DATA "crm_xml"
# define F_CRM_TASK "crm_task"
# define F_CRM_HOST_TO "crm_host_to"
# define F_CRM_MSG_TYPE F_SUBTYPE
# define F_CRM_SYS_TO "crm_sys_to"
# define F_CRM_SYS_FROM "crm_sys_from"
# define F_CRM_HOST_FROM F_ORIG
# define F_CRM_REFERENCE XML_ATTR_REFERENCE
# define F_CRM_VERSION XML_ATTR_VERSION
# define F_CRM_ORIGIN "origin"
# define F_CRM_USER "crm_user"
# define F_CRM_JOIN_ID "join_id"
# define F_CRM_DC_LEAVING "dc-leaving"
# define F_CRM_ELECTION_ID "election-id"
# define F_CRM_ELECTION_AGE_S "election-age-sec"
# define F_CRM_ELECTION_AGE_US "election-age-nano-sec"
# define F_CRM_ELECTION_OWNER "election-owner"
# define F_CRM_TGRAPH "crm-tgraph-file"
# define F_CRM_TGRAPH_INPUT "crm-tgraph-in"
# define F_CRM_THROTTLE_MODE "crm-limit-mode"
# define F_CRM_THROTTLE_MAX "crm-limit-max"
/*---- Common tags/attrs */
# define XML_DIFF_MARKER "__crm_diff_marker__"
# define XML_TAG_CIB "cib"
# define XML_TAG_FAILED "failed"
# define XML_ATTR_CRM_VERSION "crm_feature_set"
# define XML_ATTR_DIGEST "digest"
# define XML_ATTR_VALIDATION "validate-with"
# define XML_ATTR_QUORUM_PANIC "no-quorum-panic"
# define XML_ATTR_HAVE_QUORUM "have-quorum"
# define XML_ATTR_GENERATION "epoch"
# define XML_ATTR_GENERATION_ADMIN "admin_epoch"
# define XML_ATTR_NUMUPDATES "num_updates"
# define XML_ATTR_TIMEOUT "timeout"
# define XML_ATTR_ORIGIN "crm-debug-origin"
# define XML_ATTR_TSTAMP "crm-timestamp"
# define XML_CIB_ATTR_WRITTEN "cib-last-written"
# define XML_ATTR_VERSION "version"
# define XML_ATTR_DESC "description"
# define XML_ATTR_ID "id"
# define XML_ATTR_NAME "name"
# define XML_ATTR_IDREF "id-ref"
# define XML_ATTR_ID_LONG "long-id"
# define XML_ATTR_TYPE "type"
# define XML_ATTR_OP "op"
# define XML_ATTR_DC_UUID "dc-uuid"
# define XML_ATTR_UPDATE_ORIG "update-origin"
# define XML_ATTR_UPDATE_CLIENT "update-client"
# define XML_ATTR_UPDATE_USER "update-user"
# define XML_BOOLEAN_TRUE "true"
# define XML_BOOLEAN_FALSE "false"
# define XML_BOOLEAN_YES XML_BOOLEAN_TRUE
# define XML_BOOLEAN_NO XML_BOOLEAN_FALSE
# define XML_TAG_OPTIONS "options"
/*---- top level tags/attrs */
# define XML_ATTR_REQUEST "request"
# define XML_ATTR_RESPONSE "response"
# define XML_ATTR_UNAME "uname"
# define XML_ATTR_REFERENCE "reference"
# define XML_CRM_TAG_PING "ping_response"
# define XML_PING_ATTR_STATUS "result"
# define XML_PING_ATTR_SYSFROM "crm_subsystem"
# define XML_PING_ATTR_CRMDSTATE "crmd_state"
# define XML_PING_ATTR_PACEMAKERDSTATE "pacemakerd_state"
# define XML_PING_ATTR_PACEMAKERDSTATE_INIT "init"
# define XML_PING_ATTR_PACEMAKERDSTATE_STARTINGDAEMONS "starting_daemons"
# define XML_PING_ATTR_PACEMAKERDSTATE_WAITPING "wait_for_ping"
# define XML_PING_ATTR_PACEMAKERDSTATE_RUNNING "running"
# define XML_PING_ATTR_PACEMAKERDSTATE_SHUTTINGDOWN "shutting_down"
# define XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE "shutdown_complete"
# define XML_PING_ATTR_PACEMAKERDSTATE_REMOTE "remote"
# define XML_FAIL_TAG_CIB "failed_update"
# define XML_FAILCIB_ATTR_ID "id"
# define XML_FAILCIB_ATTR_OBJTYPE "object_type"
# define XML_FAILCIB_ATTR_OP "operation"
# define XML_FAILCIB_ATTR_REASON "reason"
/*---- CIB specific tags/attrs */
# define XML_CIB_TAG_SECTION_ALL "all"
# define XML_CIB_TAG_CONFIGURATION "configuration"
# define XML_CIB_TAG_STATUS "status"
# define XML_CIB_TAG_RESOURCES "resources"
# define XML_CIB_TAG_NODES "nodes"
# define XML_CIB_TAG_CONSTRAINTS "constraints"
# define XML_CIB_TAG_CRMCONFIG "crm_config"
# define XML_CIB_TAG_OPCONFIG "op_defaults"
# define XML_CIB_TAG_RSCCONFIG "rsc_defaults"
# define XML_CIB_TAG_ACLS "acls"
# define XML_CIB_TAG_ALERTS "alerts"
# define XML_CIB_TAG_ALERT "alert"
# define XML_CIB_TAG_ALERT_RECIPIENT "recipient"
# define XML_CIB_TAG_ALERT_SELECT "select"
# define XML_CIB_TAG_ALERT_ATTRIBUTES "select_attributes"
# define XML_CIB_TAG_ALERT_FENCING "select_fencing"
# define XML_CIB_TAG_ALERT_NODES "select_nodes"
# define XML_CIB_TAG_ALERT_RESOURCES "select_resources"
# define XML_CIB_TAG_ALERT_ATTR "attribute"
# define XML_CIB_TAG_STATE "node_state"
# define XML_CIB_TAG_NODE "node"
# define XML_CIB_TAG_NVPAIR "nvpair"
# define XML_CIB_TAG_PROPSET "cluster_property_set"
# define XML_TAG_ATTR_SETS "instance_attributes"
# define XML_TAG_META_SETS "meta_attributes"
# define XML_TAG_ATTRS "attributes"
# define XML_TAG_PARAMS "parameters"
# define XML_TAG_PARAM "param"
# define XML_TAG_UTILIZATION "utilization"
# define XML_TAG_RESOURCE_REF "resource_ref"
# define XML_CIB_TAG_RESOURCE "primitive"
# define XML_CIB_TAG_GROUP "group"
# define XML_CIB_TAG_INCARNATION "clone"
# define XML_CIB_TAG_CONTAINER "bundle"
# define XML_CIB_TAG_RSC_TEMPLATE "template"
# define XML_RSC_ATTR_TARGET "container-attribute-target"
# define XML_RSC_ATTR_RESTART "restart-type"
# define XML_RSC_ATTR_ORDERED "ordered"
# define XML_RSC_ATTR_INTERLEAVE "interleave"
# define XML_RSC_ATTR_INCARNATION "clone"
# define XML_RSC_ATTR_PROMOTABLE "promotable"
# define XML_RSC_ATTR_MANAGED "is-managed"
# define XML_RSC_ATTR_TARGET_ROLE "target-role"
# define XML_RSC_ATTR_UNIQUE "globally-unique"
# define XML_RSC_ATTR_NOTIFY "notify"
# define XML_RSC_ATTR_STICKINESS "resource-stickiness"
# define XML_RSC_ATTR_MULTIPLE "multiple-active"
# define XML_RSC_ATTR_REQUIRES "requires"
# define XML_RSC_ATTR_CONTAINER "container"
# define XML_RSC_ATTR_INTERNAL_RSC "internal_rsc"
# define XML_RSC_ATTR_MAINTENANCE "maintenance"
# define XML_RSC_ATTR_REMOTE_NODE "remote-node"
# define XML_RSC_ATTR_CLEAR_OP "clear_failure_op"
# define XML_RSC_ATTR_CLEAR_INTERVAL "clear_failure_interval"
# define XML_RSC_ATTR_REMOTE_RA_ADDR "addr"
# define XML_RSC_ATTR_REMOTE_RA_SERVER "server"
# define XML_RSC_ATTR_REMOTE_RA_PORT "port"
# define XML_RSC_ATTR_CRITICAL "critical"
# define XML_REMOTE_ATTR_RECONNECT_INTERVAL "reconnect_interval"
# define XML_OP_ATTR_ON_FAIL "on-fail"
# define XML_OP_ATTR_START_DELAY "start-delay"
# define XML_OP_ATTR_ALLOW_MIGRATE "allow-migrate"
# define XML_OP_ATTR_ORIGIN "interval-origin"
# define XML_OP_ATTR_PENDING "record-pending"
# define XML_OP_ATTR_DIGESTS_ALL "digests-all"
# define XML_OP_ATTR_DIGESTS_SECURE "digests-secure"
# define XML_CIB_TAG_LRM "lrm"
# define XML_LRM_TAG_RESOURCES "lrm_resources"
# define XML_LRM_TAG_RESOURCE "lrm_resource"
# define XML_LRM_TAG_RSC_OP "lrm_rsc_op"
# define XML_AGENT_ATTR_CLASS "class"
# define XML_AGENT_ATTR_PROVIDER "provider"
//! \deprecated Do not use (will be removed in a future release)
# define XML_CIB_ATTR_REPLACE "replace"
# define XML_CIB_ATTR_PRIORITY "priority"
# define XML_NODE_IS_REMOTE "remote_node"
# define XML_NODE_IS_FENCED "node_fenced"
# define XML_NODE_IS_MAINTENANCE "node_in_maintenance"
# define XML_CIB_ATTR_SHUTDOWN "shutdown"
/* Aside from being an old name for the executor, LRM is a misnomer here because
* the controller and scheduler use these to track actions, which are not always
* executor operations.
*/
// XML attribute that takes interval specification (user-facing configuration)
# define XML_LRM_ATTR_INTERVAL "interval"
// XML attribute that takes interval in milliseconds (daemon APIs)
// (identical value as above, but different constant allows clearer code intent)
# define XML_LRM_ATTR_INTERVAL_MS XML_LRM_ATTR_INTERVAL
# define XML_LRM_ATTR_TASK "operation"
# define XML_LRM_ATTR_TASK_KEY "operation_key"
# define XML_LRM_ATTR_TARGET "on_node"
# define XML_LRM_ATTR_TARGET_UUID "on_node_uuid"
/*! Actions to be executed on Pacemaker Remote nodes are routed through the
* controller on the cluster node hosting the remote connection. That cluster
* node is considered the router node for the action.
*/
# define XML_LRM_ATTR_ROUTER_NODE "router_node"
# define XML_LRM_ATTR_RSCID "rsc-id"
# define XML_LRM_ATTR_OPSTATUS "op-status"
# define XML_LRM_ATTR_RC "rc-code"
# define XML_LRM_ATTR_CALLID "call-id"
# define XML_LRM_ATTR_OP_DIGEST "op-digest"
# define XML_LRM_ATTR_OP_RESTART "op-force-restart"
# define XML_LRM_ATTR_OP_SECURE "op-secure-params"
# define XML_LRM_ATTR_RESTART_DIGEST "op-restart-digest"
# define XML_LRM_ATTR_SECURE_DIGEST "op-secure-digest"
# define XML_LRM_ATTR_EXIT_REASON "exit-reason"
# define XML_RSC_OP_LAST_CHANGE "last-rc-change"
# define XML_RSC_OP_T_EXEC "exec-time"
# define XML_RSC_OP_T_QUEUE "queue-time"
# define XML_LRM_ATTR_MIGRATE_SOURCE "migrate_source"
# define XML_LRM_ATTR_MIGRATE_TARGET "migrate_target"
# define XML_TAG_GRAPH "transition_graph"
# define XML_GRAPH_TAG_RSC_OP "rsc_op"
# define XML_GRAPH_TAG_PSEUDO_EVENT "pseudo_event"
# define XML_GRAPH_TAG_CRM_EVENT "crm_event"
# define XML_GRAPH_TAG_DOWNED "downed"
# define XML_GRAPH_TAG_MAINTENANCE "maintenance"
# define XML_TAG_RULE "rule"
# define XML_RULE_ATTR_SCORE "score"
# define XML_RULE_ATTR_SCORE_ATTRIBUTE "score-attribute"
# define XML_RULE_ATTR_ROLE "role"
# define XML_RULE_ATTR_BOOLEAN_OP "boolean-op"
# define XML_TAG_EXPRESSION "expression"
# define XML_EXPR_ATTR_ATTRIBUTE "attribute"
# define XML_EXPR_ATTR_OPERATION "operation"
# define XML_EXPR_ATTR_VALUE "value"
# define XML_EXPR_ATTR_TYPE "type"
# define XML_EXPR_ATTR_VALUE_SOURCE "value-source"
# define XML_CONS_TAG_RSC_DEPEND "rsc_colocation"
# define XML_CONS_TAG_RSC_ORDER "rsc_order"
# define XML_CONS_TAG_RSC_LOCATION "rsc_location"
# define XML_CONS_TAG_RSC_TICKET "rsc_ticket"
# define XML_CONS_TAG_RSC_SET "resource_set"
# define XML_CONS_ATTR_SYMMETRICAL "symmetrical"
# define XML_LOCATION_ATTR_DISCOVERY "resource-discovery"
# define XML_COLOC_ATTR_SOURCE "rsc"
# define XML_COLOC_ATTR_SOURCE_ROLE "rsc-role"
# define XML_COLOC_ATTR_TARGET "with-rsc"
# define XML_COLOC_ATTR_TARGET_ROLE "with-rsc-role"
# define XML_COLOC_ATTR_NODE_ATTR "node-attribute"
# define XML_COLOC_ATTR_INFLUENCE "influence"
//! \deprecated Deprecated since 2.1.5
# define XML_COLOC_ATTR_SOURCE_INSTANCE "rsc-instance"
//! \deprecated Deprecated since 2.1.5
# define XML_COLOC_ATTR_TARGET_INSTANCE "with-rsc-instance"
# define XML_LOC_ATTR_SOURCE "rsc"
# define XML_LOC_ATTR_SOURCE_PATTERN "rsc-pattern"
# define XML_ORDER_ATTR_FIRST "first"
# define XML_ORDER_ATTR_THEN "then"
# define XML_ORDER_ATTR_FIRST_ACTION "first-action"
# define XML_ORDER_ATTR_THEN_ACTION "then-action"
# define XML_ORDER_ATTR_KIND "kind"
//! \deprecated Deprecated since 2.1.5
# define XML_ORDER_ATTR_FIRST_INSTANCE "first-instance"
//! \deprecated Deprecated since 2.1.5
# define XML_ORDER_ATTR_THEN_INSTANCE "then-instance"
# define XML_TICKET_ATTR_TICKET "ticket"
# define XML_TICKET_ATTR_LOSS_POLICY "loss-policy"
# define XML_NVPAIR_ATTR_NAME "name"
# define XML_NVPAIR_ATTR_VALUE "value"
# define XML_NODE_ATTR_RSC_DISCOVERY "resource-discovery-enabled"
# define XML_ALERT_ATTR_PATH "path"
# define XML_ALERT_ATTR_TIMEOUT "timeout"
# define XML_ALERT_ATTR_TSTAMP_FORMAT "timestamp-format"
# define XML_ALERT_ATTR_REC_VALUE "value"
# define XML_CIB_TAG_GENERATION_TUPPLE "generation_tuple"
# define XML_ATTR_TRANSITION_MAGIC "transition-magic"
# define XML_ATTR_TRANSITION_KEY "transition-key"
# define XML_ATTR_TE_NOWAIT "op_no_wait"
# define XML_ATTR_TE_TARGET_RC "op_target_rc"
# define XML_TAG_TRANSIENT_NODEATTRS "transient_attributes"
//! \deprecated Do not use (will be removed in a future release)
# define XML_TAG_DIFF_ADDED "diff-added"
//! \deprecated Do not use (will be removed in a future release)
# define XML_TAG_DIFF_REMOVED "diff-removed"
# define XML_ACL_TAG_USER "acl_target"
# define XML_ACL_TAG_USERv1 "acl_user"
# define XML_ACL_TAG_GROUP "acl_group"
# define XML_ACL_TAG_ROLE "acl_role"
# define XML_ACL_TAG_PERMISSION "acl_permission"
# define XML_ACL_TAG_ROLE_REF "role"
# define XML_ACL_TAG_ROLE_REFv1 "role_ref"
# define XML_ACL_ATTR_KIND "kind"
# define XML_ACL_TAG_READ "read"
# define XML_ACL_TAG_WRITE "write"
# define XML_ACL_TAG_DENY "deny"
# define XML_ACL_ATTR_REF "reference"
# define XML_ACL_ATTR_REFv1 "ref"
# define XML_ACL_ATTR_TAG "object-type"
# define XML_ACL_ATTR_TAGv1 "tag"
# define XML_ACL_ATTR_XPATH "xpath"
# define XML_ACL_ATTR_ATTRIBUTE "attribute"
# define XML_CIB_TAG_TICKETS "tickets"
# define XML_CIB_TAG_TICKET_STATE "ticket_state"
# define XML_CIB_TAG_TAGS "tags"
# define XML_CIB_TAG_TAG "tag"
# define XML_CIB_TAG_OBJ_REF "obj_ref"
# define XML_TAG_FENCING_TOPOLOGY "fencing-topology"
# define XML_TAG_FENCING_LEVEL "fencing-level"
# define XML_ATTR_STONITH_INDEX "index"
# define XML_ATTR_STONITH_TARGET "target"
# define XML_ATTR_STONITH_TARGET_VALUE "target-value"
# define XML_ATTR_STONITH_TARGET_PATTERN "target-pattern"
# define XML_ATTR_STONITH_TARGET_ATTRIBUTE "target-attribute"
# define XML_ATTR_STONITH_DEVICES "devices"
# define XML_TAG_DIFF "diff"
# define XML_DIFF_VERSION "version"
# define XML_DIFF_VSOURCE "source"
# define XML_DIFF_VTARGET "target"
# define XML_DIFF_CHANGE "change"
# define XML_DIFF_LIST "change-list"
# define XML_DIFF_ATTR "change-attr"
# define XML_DIFF_RESULT "change-result"
# define XML_DIFF_OP "operation"
# define XML_DIFF_PATH "path"
# define XML_DIFF_POSITION "position"
# define ID(x) crm_element_value(x, XML_ATTR_ID)
#ifdef __cplusplus
}
#endif
#endif
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index 8af8ad7f08..0e1520ae62 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1669 +1,1670 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <stdbool.h>
#include <stdint.h> // uint8_t, uint32_t
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
#define RSC_ROLE_MAX (pcmk_role_promoted + 1)
static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the immediate next role when transitioning from one role
* to a target role. For example, when going from Stopped to Promoted, the
* next role is Unpromoted, because the resource must be started before it
* can be promoted. The current state then becomes Started, which is fed
* into this array again, giving a next role of Promoted.
*
* Current role Immediate next role Final target role
* ------------ ------------------- -----------------
*/
/* Unknown */ { pcmk_role_unknown, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_stopped, /* Unpromoted */
pcmk_role_stopped, /* Promoted */
},
/* Stopped */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_unpromoted, /* Promoted */
},
/* Started */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Unpromoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Promoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_unpromoted, /* Stopped */
pcmk_role_unpromoted, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
};
/*!
* \internal
* \brief Function to schedule actions needed for a role change
*
* \param[in,out] rsc Resource whose role is changing
* \param[in,out] node Node where resource will be in its next role
* \param[in] optional Whether scheduled actions should be optional
*/
typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the function needed to transition directly from one role
* to another. NULL indicates that nothing is needed.
*
* Current role Transition function Next role
* ------------ ------------------- ----------
*/
/* Unknown */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
assert_role_error, /* Started */
assert_role_error, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Stopped */ { assert_role_error, /* Unknown */
NULL, /* Stopped */
start_resource, /* Started */
start_resource, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Started */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
NULL, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Unpromoted */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
stop_resource, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Promoted */ { assert_role_error, /* Unknown */
demote_resource, /* Stopped */
demote_resource, /* Started */
demote_resource, /* Unpromoted */
NULL, /* Promoted */
},
};
/*!
* \internal
* \brief Get a list of a resource's allowed nodes sorted by node score
*
* \param[in] rsc Resource to check
*
* \return List of allowed nodes sorted by node score
*/
static GList *
sorted_allowed_nodes(const pcmk_resource_t *rsc)
{
if (rsc->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->allowed_nodes);
if (nodes != NULL) {
return pcmk__sort_nodes(nodes, pe__current_node(rsc));
}
}
return NULL;
}
/*!
* \internal
* \brief Assign a resource to its best allowed node, if possible
*
* \param[in,out] rsc Resource to choose a node for
* \param[in] prefer If not \c NULL, prefer this node when all else
* equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return true if \p rsc could be assigned to a node, otherwise false
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
static bool
assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *nodes = NULL;
pcmk_node_t *chosen = NULL;
pcmk_node_t *best = NULL;
const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// We've already finished assignment of resources to nodes
return rsc->allocated_to != NULL;
}
// Sort allowed nodes by score
nodes = sorted_allowed_nodes(rsc);
if (nodes != NULL) {
best = (pcmk_node_t *) nodes->data; // First node has best score
}
if ((prefer != NULL) && (nodes != NULL)) {
// Get the allowed node version of prefer
chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id);
if (chosen == NULL) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown",
pe__node_name(prefer), rsc->id);
/* Favor the preferred node as long as its score 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 score is less than INFINITY.
*/
} else if (chosen->weight < best->weight) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
pe__node_name(chosen), rsc->id);
chosen = NULL;
} else if (!pcmk__node_available(chosen, true, false)) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pe__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pcmk__rsc_trace(rsc,
"Chose preferred node %s for %s "
"(ignoring %d candidates)",
pe__node_name(chosen), rsc->id,
g_list_length(nodes));
}
}
if ((chosen == NULL) && (best != NULL)) {
/* Either there is no preferred node, or the preferred node is not
* suitable, but another node is allowed to run the resource.
*/
chosen = best;
if (!pe_rsc_is_unique_clone(rsc->parent)
&& (chosen->weight > 0) // Zero not acceptable
&& pcmk__node_available(chosen, false, false)) {
/* 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
* pcmk__assign_instances() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unassigned instances to prefer a node that's already
* running another instance.
*/
pcmk_node_t *running = pe__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pcmk__rsc_trace(rsc,
"Current node for %s (%s) can't run resources",
rsc->id, pe__node_name(running));
} else {
int nodes_with_best_score = 1;
for (GList *iter = nodes->next; iter; iter = iter->next) {
pcmk_node_t *allowed = (pcmk_node_t *) iter->data;
if (allowed->weight != chosen->weight) {
// The nodes are sorted by score, so no more are equal
break;
}
if (pe__same_node(allowed, running)) {
// Scores are equal, so prefer the current node
chosen = allowed;
}
nodes_with_best_score++;
}
if (nodes_with_best_score > 1) {
uint8_t log_level = LOG_INFO;
if (chosen->weight >= INFINITY) {
log_level = LOG_WARNING;
}
do_crm_log(log_level,
"Chose %s for %s from %d nodes with score %s",
pe__node_name(chosen), rsc->id,
nodes_with_best_score,
pcmk_readable_score(chosen->weight));
}
}
}
pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates",
pe__node_name(chosen), rsc->id, g_list_length(nodes));
}
pcmk__assign_resource(rsc, chosen, false, stop_if_fail);
g_list_free(nodes);
return rsc->allocated_to != NULL;
}
/*!
* \internal
* \brief Apply a "this with" colocation to a node's allowed node scores
*
* \param[in,out] colocation Colocation to apply
* \param[in,out] rsc Resource being assigned
*/
static void
apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc)
{
GHashTable *archive = NULL;
pcmk_resource_t *other = colocation->primary;
// In certain cases, we will need to revert the node scores
if ((colocation->dependent_role >= pcmk_role_promoted)
|| ((colocation->score < 0) && (colocation->score > -INFINITY))) {
archive = pcmk__copy_node_table(rsc->allowed_nodes);
}
if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) {
pcmk__rsc_trace(rsc,
"%s: Assigning colocation %s primary %s first"
"(score=%d role=%s)",
rsc->id, colocation->id, other->id,
colocation->score, role2text(colocation->dependent_role));
other->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->allowed_nodes)) {
pcmk__rsc_info(rsc,
"%s: Reverting scores from colocation with %s "
"because no nodes allowed",
rsc->id, other->id);
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = archive;
archive = NULL;
}
if (archive != NULL) {
g_hash_table_destroy(archive);
}
}
/*!
* \internal
* \brief Update a Pacemaker Remote node once its connection has been assigned
*
* \param[in] connection Connection resource that has been assigned
*/
static void
remote_connection_assigned(const pcmk_resource_t *connection)
{
pcmk_node_t *remote_node = pe_find_node(connection->cluster->nodes,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->allocated_to != NULL)
&& (connection->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->details->id);
remote_node->details->online = TRUE;
if (remote_node->details->unseen) {
// Avoid unnecessary fence, since we will attempt connection
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Pacemaker Remote node %s will be shut down "
"(%sassigned connection's next role is %s)",
remote_node->details->id,
((connection->allocated_to == NULL)? "un" : ""),
role2text(connection->next_role));
remote_node->details->shutdown = TRUE;
}
}
/*!
* \internal
* \brief Assign a primitive resource to a node
*
* \param[in,out] rsc Resource to assign to a node
* \param[in] prefer Node to prefer, if all else is equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pcmk_node_t *
pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *this_with_colocations = NULL;
GList *with_this_colocations = NULL;
GList *iter = NULL;
pcmk__colocation_t *colocation = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
// Never assign a child without parent being assigned first
if ((rsc->parent != NULL)
&& !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, rsc->parent->id);
rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->allocated_to != NULL) {
node_name = pe__node_name(rsc->allocated_to);
}
pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->allocated_to;
}
// Ensure we detect assignment loops
if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pe__set_resource_flags(rsc, pcmk_rsc_assigning);
pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes,
rsc->cluster);
this_with_colocations = pcmk__this_with_colocations(rsc);
with_this_colocations = pcmk__with_this_colocations(rsc);
// Apply mandatory colocations first, to satisfy as many as possible
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->allowed_nodes, rsc->cluster);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->next_role == pcmk_role_stopped) {
pcmk__rsc_trace(rsc,
"Banning %s from all nodes because it will be stopped",
rsc->id);
resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE,
rsc->cluster);
} else if ((rsc->next_role > rsc->role)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate)
&& (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Resource %s cannot be elevated from %s to %s due to "
PCMK_OPT_NO_QUORUM_POLICY "=freeze",
rsc->id, role2text(rsc->role), role2text(rsc->next_role));
pe__set_next_role(rsc, rsc->role, PCMK_OPT_NO_QUORUM_POLICY "=freeze");
}
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->allowed_nodes, rsc->cluster);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) {
pe__clear_resource_flags(rsc, pcmk_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pcmk_node_t *assign_to = NULL;
pe__set_next_role(rsc, rsc->role, "unmanaged");
assign_to = pe__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->role == pcmk_role_promoted) {
reason = "promoted";
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id,
(assign_to? assign_to->details->uname : "no node"),
reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
- pcmk__rsc_debug(rsc, "Forcing %s to stop: stop-all-resources",
+ pcmk__rsc_debug(rsc,
+ "Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES,
rsc->id);
}
pcmk__assign_resource(rsc, NULL, true, stop_if_fail);
} else if (!assign_best_node(rsc, prefer, stop_if_fail)) {
// Assignment failed
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->running_on != NULL) && stop_if_fail) {
pcmk__rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
}
pe__clear_resource_flags(rsc, pcmk_rsc_assigning);
if (rsc->is_remote_node) {
remote_connection_assigned(rsc);
}
return rsc->allocated_to;
}
/*!
* \internal
* \brief Schedule actions to bring resource down and back to current role
*
* \param[in,out] rsc Resource to restart
* \param[in,out] current Node that resource should be brought down on
* \param[in] need_stop Whether the resource must be stopped
* \param[in] need_promote Whether the resource must be promoted
*
* \return Role that resource would have after scheduled actions are taken
*/
static void
schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current,
bool need_stop, bool need_promote)
{
enum rsc_role_e role = rsc->role;
enum rsc_role_e next_role;
rsc_transition_fn fn = NULL;
pe__set_resource_flags(rsc, pcmk_rsc_restarting);
// Bring resource down to a stop on its current node
while (role != pcmk_role_stopped) {
next_role = rsc_state_matrix[role][pcmk_role_stopped];
pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s",
(need_stop? "required" : "optional"), rsc->id,
role2text(role), role2text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, current, !need_stop);
role = next_role;
}
// Bring resource up to its next role on its next node
while ((rsc->role <= rsc->next_role) && (role != rsc->role)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s",
(required? "required" : "optional"), rsc->id,
role2text(role), role2text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, !required);
role = next_role;
}
pe__clear_resource_flags(rsc, pcmk_rsc_restarting);
}
/*!
* \internal
* \brief If a resource's next role is not explicitly specified, set a default
*
* \param[in,out] rsc Resource to set next role for
*
* \return "explicit" if next role was explicitly set, otherwise "implicit"
*/
static const char *
set_default_next_role(pcmk_resource_t *rsc)
{
if (rsc->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->allocated_to == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "assignment");
} else {
pe__set_next_role(rsc, pcmk_role_started, "assignment");
}
return "implicit";
}
/*!
* \internal
* \brief Create an action to represent an already pending start
*
* \param[in,out] rsc Resource to create start action for
*/
static void
create_pending_start(pcmk_resource_t *rsc)
{
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc,
"Creating action for %s to represent already pending start",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
pe__set_action_flags(start, pcmk_action_always_in_graph);
}
/*!
* \internal
* \brief Schedule actions needed to take a resource to its next role
*
* \param[in,out] rsc Resource to schedule actions for
*/
static void
schedule_role_transition_actions(pcmk_resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
while (role != rsc->next_role) {
enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role];
rsc_transition_fn fn = NULL;
pcmk__rsc_trace(rsc,
"Creating action to take %s from %s to %s "
"(ending at %s)",
rsc->id, role2text(role), role2text(next_role),
role2text(rsc->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, false);
role = next_role;
}
}
/*!
* \internal
* \brief Create all actions needed for a given primitive resource
*
* \param[in,out] rsc Primitive resource to create actions for
*/
void
pcmk__primitive_create_actions(pcmk_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pcmk_node_t *current = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
next_role_source = set_default_next_role(rsc);
pcmk__rsc_trace(rsc,
"Creating all actions for %s transition from %s to %s "
"(%s) on %s",
rsc->id, role2text(rsc->role), role2text(rsc->next_role),
next_role_source, pe__node_name(rsc->allocated_to));
current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active);
g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration,
rsc);
if ((current != NULL) && (rsc->allocated_to != NULL)
&& !pe__same_node(current, rsc->allocated_to)
&& (rsc->next_role >= pcmk_role_started)) {
pcmk__rsc_trace(rsc, "Moving %s from %s to %s",
rsc->id, pe__node_name(current),
pe__node_name(rsc->allocated_to));
is_moving = true;
allow_migrate = pcmk__rsc_can_migrate(rsc, current);
// This is needed even if migrating (though I'm not sure why ...)
need_stop = true;
}
// Check whether resource is partially migrated and/or multiply active
if ((rsc->partial_migration_source != NULL)
&& (rsc->partial_migration_target != NULL)
&& allow_migrate && (num_all_active == 2)
&& pe__same_node(current, rsc->partial_migration_source)
&& pe__same_node(rsc->allocated_to, rsc->partial_migration_target)) {
/* A partial migration is in progress, and the migration target remains
* the same as when the migration began.
*/
pcmk__rsc_trace(rsc,
"Partial migration of %s from %s to %s will continue",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
} else if ((rsc->partial_migration_source != NULL)
|| (rsc->partial_migration_target != NULL)) {
// A partial migration is in progress but can't be continued
if (num_all_active > 2) {
// The resource is migrating *and* multiply active!
crm_notice("Forcing recovery of %s because it is migrating "
"from %s to %s and possibly active elsewhere",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
} else {
// The migration source or target isn't available
crm_notice("Forcing recovery of %s because it can no longer "
"migrate from %s to %s",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
}
need_stop = true;
rsc->partial_migration_source = rsc->partial_migration_target = NULL;
allow_migrate = false;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
multiply_active = (num_all_active > 1);
} else {
/* 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
* PCMK_OPT_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);
}
if (multiply_active) {
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
// Resource was (possibly) incorrectly multiply active
pcmk__sched_err("%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), 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");
switch (rsc->recovery_type) {
case pcmk_multiply_active_restart:
need_stop = true;
break;
case pcmk_multiply_active_unexpected:
need_stop = true; // stop_resource() will skip expected node
pe__set_resource_flags(rsc, pcmk_rsc_stop_unexpected);
break;
default:
break;
}
} else {
pe__clear_resource_flags(rsc, pcmk_rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
create_pending_start(rsc);
}
if (is_moving) {
// Remaining tests are only for resources staying where they are
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) {
need_stop = true;
pcmk__rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id);
if (rsc->next_role == pcmk_role_promoted) {
need_promote = true;
}
}
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id);
need_stop = true;
} else if ((rsc->role > pcmk_role_started) && (current != NULL)
&& (rsc->allocated_to != NULL)) {
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
if (!pcmk_is_set(start->flags, pcmk_action_optional)) {
// Recovery of a promoted resource
pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id);
need_stop = true;
}
}
// Create any actions needed to bring resource down and back up to same role
schedule_restart_actions(rsc, current, need_stop, need_promote);
// Create any actions needed to take resource from this role to the next
schedule_role_transition_actions(rsc);
pcmk__create_recurring_actions(rsc);
if (allow_migrate) {
pcmk__create_migration_actions(rsc, current);
}
}
/*!
* \internal
* \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes
*
* \param[in] rsc Resource to check
*/
static void
rsc_avoids_remote_nodes(const pcmk_resource_t *rsc)
{
GHashTableIter iter;
pcmk_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 != NULL) {
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
*
* \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(const pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
if (rsc->allowed_nodes) {
allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
if (!pcmk__is_daemon) {
allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name);
}
return allowed_nodes;
}
/*!
* \internal
* \brief Create implicit constraints needed for a primitive resource
*
* \param[in,out] rsc Primitive resource to create implicit constraints for
*/
void
pcmk__primitive_internal_constraints(pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
bool check_unfencing = false;
bool check_utilization = false;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource "
"%s", rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags,
pcmk_sched_enable_unfencing)
&& pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& !pcmk__str_eq(rsc->cluster->placement_strategy,
"default", pcmk__str_casei);
// Order stops before starts (i.e. restart)
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL,
pcmk__ar_ordered
|pcmk__ar_first_implies_then
|pcmk__ar_intermediate_stop,
rsc->cluster);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable)
|| (rsc->role > pcmk_role_unpromoted)) {
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pcmk__ar_promoted_then_implies_first, rsc->cluster);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0),
NULL,
pcmk__ar_unrunnable_first_blocks, rsc->cluster);
}
// Don't clear resource history if probing on same node
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL,
pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first,
rsc->cluster);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization || (rsc->container != NULL)) {
allowed_nodes = allowed_nodes_as_list(rsc);
}
if (check_unfencing) {
g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc);
}
if (check_utilization) {
pcmk__create_utilization_constraints(rsc, allowed_nodes);
}
if (rsc->container != NULL) {
pcmk_resource_t *remote_rsc = NULL;
if (rsc->is_remote_node) {
// rsc is the implicit remote connection for a guest or bundle node
/* Guest resources are not allowed to run on Pacemaker Remote nodes,
* to avoid nesting remotes. However, bundles are allowed.
*/
if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
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.
*/
pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pcmk__ar_ordered);
/* 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(rsc->cluster,
rsc->container);
}
if (remote_rsc != NULL) {
/* 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) {
pcmk_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);
pcmk__new_ordering(rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_START, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
pcmk__ar_first_implies_then
|pcmk__ar_unrunnable_first_blocks,
rsc->cluster);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_STOP, 0),
NULL, pcmk__ar_then_implies_first, rsc->cluster);
if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->container, NULL, NULL,
pcmk__coloc_influence);
}
}
if (rsc->is_remote_node
|| pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
/* Remote connections and fencing devices are not allowed to run on
* Pacemaker Remote nodes
*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint, apply its score to the dependent's
* allowed node scores (if we are still placing resources) or priority (if
* we are choosing promotable clone instance roles).
*
* \param[in,out] dependent Dependent resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
* \param[in] for_dependent true if called on behalf of dependent
*/
void
pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent,
const pcmk_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
enum pcmk__coloc_affects filter_results;
CRM_ASSERT((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
// Always process on behalf of primary resource
primary->cmds->apply_coloc_score(dependent, primary, colocation, false);
return;
}
filter_results = pcmk__colocation_affects(dependent, primary, colocation,
false);
pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)",
((colocation->score > 0)? "Colocating" : "Anti-colocating"),
dependent->id, primary->id, colocation->id,
colocation->score,
filter_results);
switch (filter_results) {
case pcmk__coloc_affects_role:
pcmk__apply_coloc_to_priority(dependent, primary, colocation);
break;
case pcmk__coloc_affects_location:
pcmk__apply_coloc_to_scores(dependent, primary, colocation);
break;
default: // pcmk__coloc_affects_nothing
return;
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_with_this(list, colocation, orig_rsc);
}
}
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
}
}
/*!
* \internal
* \brief Return action flags for a given primitive resource action
*
* \param[in,out] action Action to get flags for
* \param[in] node If not NULL, limit effects to this node (ignored)
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node)
{
CRM_ASSERT(action != NULL);
return (uint32_t) action->flags;
}
/*!
* \internal
* \brief Check whether a node is a multiply active resource's expected node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p rsc is multiply active with multiple-active set to
* stop_unexpected, and \p node is the node where it will remain active
* \note This assumes that the resource's next role cannot be changed to stopped
* after this is called, which should be reasonable if status has already
* been unpacked and resources have been assigned to nodes.
*/
static bool
is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pcmk_rsc_stop_unexpected|pcmk_rsc_restarting)
&& (rsc->next_role > pcmk_role_stopped)
&& pe__same_node(rsc->allocated_to, node);
}
/*!
* \internal
* \brief Schedule actions needed to stop a resource wherever it is active
*
* \param[in,out] rsc Resource being stopped
* \param[in] node Node where resource is being stopped (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
pcmk_action_t *stop = NULL;
if (is_expected_node(rsc, current)) {
/* We are scheduling restart actions for a multiply active resource
* with multiple-active=stop_unexpected, and this is where it should
* not be stopped.
*/
pcmk__rsc_trace(rsc,
"Skipping stop of multiply active resource %s "
"on expected node %s",
rsc->id, pe__node_name(current));
continue;
}
if (rsc->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pe__same_node(current, rsc->partial_migration_target)
&& pe__same_node(current, rsc->allocated_to)) {
pcmk__rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pe__node_name(current));
continue;
} else {
pcmk__rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pe__node_name(current));
optional = false;
}
}
pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s",
rsc->id, pe__node_name(current));
stop = stop_action(rsc, current, optional);
if (rsc->allocated_to == NULL) {
pe_action_set_reason(stop, "node availability", true);
} else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting
|pcmk_rsc_stop_unexpected)) {
/* We are stopping a multiply active resource on a node that is
* not its expected node, and we are still scheduling restart
* actions, so the stop is for being multiply active.
*/
pe_action_set_reason(stop, "being multiply active", true);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pe__clear_action_flags(stop, pcmk_action_runnable);
}
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) {
pcmk__schedule_cleanup(rsc, current, optional);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) {
pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false, rsc->cluster);
order_actions(stop, unfence, pcmk__ar_then_implies_first);
if (!pcmk__node_unfenced(current)) {
pcmk__sched_err("Stopping %s until %s can be unfenced",
rsc->id, pe__node_name(current));
}
}
}
}
/*!
* \internal
* \brief Schedule actions needed to start a resource on a node
*
* \param[in,out] rsc Resource being started
* \param[in,out] node Node where resource should be started
* \param[in] optional Whether actions should be optional
*/
static void
start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
pcmk_action_t *start = NULL;
CRM_ASSERT(node != NULL);
pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)",
(optional? "optional" : "required"), rsc->id,
pe__node_name(node), node->weight);
start = start_action(rsc, node, TRUE);
pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then);
if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) {
pe__clear_action_flags(start, pcmk_action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pcmk__rsc_trace(rsc,
"Start of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pe__node_name(node));
pe__set_action_flags(start, pcmk_action_pseudo);
}
}
/*!
* \internal
* \brief Schedule actions needed to promote a resource on a node
*
* \param[in,out] rsc Resource being promoted
* \param[in] node Node where resource should be promoted
* \param[in] optional Whether actions should be optional
*/
static void
promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
GList *iter = NULL;
GList *action_list = NULL;
bool runnable = true;
CRM_ASSERT(node != NULL);
// Any start must be runnable for promotion to be runnable
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *start = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pcmk_action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pcmk_action_t *promote = promote_action(rsc, node, optional);
pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pe__node_name(node));
if (is_expected_node(rsc, node)) {
/* This could be a problem if the promote becomes necessary for
* other reasons later.
*/
pcmk__rsc_trace(rsc,
"Promotion of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pe__node_name(node));
pe__set_action_flags(promote, pcmk_action_pseudo);
}
} else {
pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable",
rsc->id, pe__node_name(node));
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE,
true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *promote = (pcmk_action_t *) iter->data;
pe__clear_action_flags(promote, pcmk_action_runnable);
}
g_list_free(action_list);
}
}
/*!
* \internal
* \brief Schedule actions needed to demote a resource wherever it is active
*
* \param[in,out] rsc Resource being demoted
* \param[in] node Node where resource should be demoted (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
/* Since this will only be called for a primitive (possibly as an instance
* of a collective resource), the resource is multiply active if it is
* running on more than one node, so we want to demote on all of them as
* part of recovery, regardless of which one is the desired node.
*/
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pcmk__rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pe__node_name(current));
} else {
pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pe__node_name(current));
demote_action(rsc, current, optional);
}
}
}
static void
assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
CRM_ASSERT(false);
}
/*!
* \internal
* \brief Schedule cleanup of a resource
*
* \param[in,out] rsc Resource to clean up
* \param[in] node Node to clean up on
* \param[in] optional Whether clean-up should be optional
*/
void
pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node,
bool optional)
{
/* If the cleanup is required, its orderings are optional, because they're
* relevant only if both actions are required. Conversely, if the cleanup is
* optional, the orderings make the then action required if the first action
* becomes required.
*/
uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed",
rsc->id, pe__node_name(node));
return;
}
if (node->details->unclean || !node->details->online) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable",
rsc->id, pe__node_name(node));
return;
}
crm_notice("Scheduling clean-up of %s on %s", rsc->id, pe__node_name(node));
delete_action(rsc, node, optional);
// stop -> clean-up -> start
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_DELETE, flag);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE,
rsc, PCMK_ACTION_START, flag);
}
/*!
* \internal
* \brief Add primitive meta-attributes relevant to graph actions to XML
*
* \param[in] rsc Primitive resource whose meta-attributes should be added
* \param[in,out] xml Transition graph action attributes XML to add to
*/
void
pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pcmk_resource_t *parent = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (xml != NULL));
/* Clone instance numbers get set internally as meta-attributes, and are
* needed in the transition graph (for example, to tell unique clone
* instances apart).
*/
value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION);
if (value != NULL) {
name = crm_meta_name(XML_RSC_ATTR_INCARNATION);
crm_xml_add(xml, name, value);
free(name);
}
// Not sure if this one is really needed ...
value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE);
if (value != NULL) {
name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE);
crm_xml_add(xml, name, value);
free(name);
}
/* The container meta-attribute can be set on the primitive itself or one of
* its parents (for example, a group inside a container resource), so check
* them all, and keep the highest one found.
*/
for (parent = rsc; parent != NULL; parent = parent->parent) {
if (parent->container != NULL) {
crm_xml_add(xml, CRM_META "_" XML_RSC_ATTR_CONTAINER,
parent->container->id);
}
}
/* Bundle replica children will get their external-ip set internally as a
* meta-attribute. The graph action needs it, but under a different naming
* convention than other meta-attributes.
*/
value = g_hash_table_lookup(rsc->meta, "external-ip");
if (value != NULL) {
crm_xml_add(xml, "pcmk_external_ip", value);
}
}
// Primitive implementation of pcmk_assignment_methods_t:add_utilization()
void
pcmk__primitive_add_utilization(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc,
GList *all_rscs, GHashTable *utilization)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (orig_rsc != NULL) && (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return;
}
pcmk__rsc_trace(orig_rsc,
"%s: Adding primitive %s as colocated utilization",
orig_rsc->id, rsc->id);
pcmk__release_node_capacity(utilization, rsc);
}
/*!
* \internal
* \brief Get epoch time of node's shutdown attribute (or now if none)
*
* \param[in,out] node Node to check
*
* \return Epoch time corresponding to shutdown attribute if set or now if not
*/
static time_t
shutdown_time(pcmk_node_t *node)
{
const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
time_t result = 0;
if (shutdown != NULL) {
long long result_ll;
if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) {
result = (time_t) result_ll;
}
}
return (result == 0)? get_effective_time(node->details->data_set) : result;
}
/*!
* \internal
* \brief Ban a resource from a node if it's not locked to the node
*
* \param[in] data Node to check
* \param[in,out] user_data Resource to check
*/
static void
ban_if_not_locked(gpointer data, gpointer user_data)
{
const pcmk_node_t *node = (const pcmk_node_t *) data;
pcmk_resource_t *rsc = (pcmk_resource_t *) user_data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
PCMK_OPT_SHUTDOWN_LOCK, rsc->cluster);
}
}
// Primitive implementation of pcmk_assignment_methods_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc)
{
const char *class = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
// Fence devices and remote connections can't be locked
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches)
|| pe__resource_is_remote_conn(rsc)) {
return;
}
if (rsc->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->running_on != NULL) {
/* The resource was started elsewhere even though it is now
* considered locked. This shouldn't be possible, but as a
* failsafe, we don't want to disturb the resource now.
*/
pcmk__rsc_info(rsc,
"Cancelling shutdown lock "
"because %s is already active", rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pcmk_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pcmk__rsc_debug(rsc,
"Not locking %s to unclean %s for shutdown",
rsc->id, pe__node_name(node));
} else {
rsc->lock_node = node;
rsc->lock_time = shutdown_time(node);
}
}
}
if (rsc->lock_node == NULL) {
// No lock needed
return;
}
if (rsc->cluster->shutdown_lock > 0) {
time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock;
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pe__node_name(rsc->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, rsc->cluster,
"shutdown lock expiration");
} else {
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pe__node_name(rsc->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pengine/common.c b/lib/pengine/common.c
index af816d0329..334b8b2007 100644
--- a/lib/pengine/common.c
+++ b/lib/pengine/common.c
@@ -1,624 +1,624 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/util.h>
#include <glib.h>
#include <crm/common/scheduler_internal.h>
#include <crm/pengine/internal.h>
static bool
check_placement_strategy(const char *value)
{
return pcmk__strcase_any_of(value, "default", "utilization", "minimal",
"balanced", NULL);
}
static pcmk__cluster_option_t pe_opts[] = {
/* name, old name, type, allowed values,
* default value, validator,
* short description,
* long description
*/
{
PCMK_OPT_NO_QUORUM_POLICY, NULL, "select",
"stop, freeze, ignore, demote, suicide",
"stop", pcmk__valid_quorum,
N_("What to do when the cluster does not have quorum"),
NULL
},
{
"symmetric-cluster", NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("Whether resources can run on any node by default"),
NULL
},
{
PCMK_OPT_MAINTENANCE_MODE, NULL, "boolean", NULL,
"false", pcmk__valid_boolean,
N_("Whether the cluster should refrain from monitoring, starting, "
"and stopping resources"),
NULL
},
{
PCMK_OPT_START_FAILURE_IS_FATAL, NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("Whether a start failure should prevent a resource from being "
"recovered on the same node"),
N_("When true, the cluster will immediately ban a resource from a node "
"if it fails to start there. When false, the cluster will instead "
"check the resource's fail count against its migration-threshold.")
},
{
PCMK_OPT_ENABLE_STARTUP_PROBES, NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("Whether the cluster should check for active resources during start-up"),
NULL
},
{
PCMK_OPT_SHUTDOWN_LOCK, NULL, "boolean", NULL,
"false", pcmk__valid_boolean,
N_("Whether to lock resources to a cleanly shut down node"),
N_("When true, resources active on a node when it is cleanly shut down "
"are kept \"locked\" to that node (not allowed to run elsewhere) "
"until they start again on that node after it rejoins (or for at "
"most shutdown-lock-limit, if set). Stonith resources and "
"Pacemaker Remote connections are never locked. Clone and bundle "
"instances and the promoted role of promotable clones are "
"currently never locked, though support could be added in a future "
"release.")
},
{
PCMK_OPT_SHUTDOWN_LOCK_LIMIT, NULL, "time", NULL,
"0", pcmk__valid_interval_spec,
N_("Do not lock resources to a cleanly shut down node longer than "
"this"),
N_("If shutdown-lock is true and this is set to a nonzero time "
"duration, shutdown locks will expire after this much time has "
"passed since the shutdown was initiated, even if the node has not "
"rejoined.")
},
// Fencing-related options
{
PCMK_OPT_STONITH_ENABLED, NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("*** Advanced Use Only *** "
"Whether nodes may be fenced as part of recovery"),
N_("If false, unresponsive nodes are immediately assumed to be harmless, "
"and resources that were active on them may be recovered "
"elsewhere. This can result in a \"split-brain\" situation, "
"potentially leading to data loss and/or service unavailability.")
},
{
PCMK_OPT_STONITH_ACTION, NULL, "select", "reboot, off, poweroff",
PCMK_ACTION_REBOOT, pcmk__is_fencing_action,
N_("Action to send to fence device when a node needs to be fenced "
"(\"poweroff\" is a deprecated alias for \"off\")"),
NULL
},
{
PCMK_OPT_STONITH_TIMEOUT, NULL, "time", NULL,
"60s", pcmk__valid_interval_spec,
N_("*** Advanced Use Only *** Unused by Pacemaker"),
N_("This value is not used by Pacemaker, but is kept for backward "
"compatibility, and certain legacy fence agents might use it.")
},
{
PCMK_OPT_HAVE_WATCHDOG, NULL, "boolean", NULL,
"false", pcmk__valid_boolean,
N_("Whether watchdog integration is enabled"),
N_("This is set automatically by the cluster according to whether SBD "
"is detected to be in use. User-configured values are ignored. "
"The value `true` is meaningful if diskless SBD is used and "
"`stonith-watchdog-timeout` is nonzero. In that case, if fencing "
"is required, watchdog-based self-fencing will be performed via "
"SBD without requiring a fencing resource explicitly configured.")
},
{
PCMK_OPT_CONCURRENT_FENCING, NULL, "boolean", NULL,
PCMK__CONCURRENT_FENCING_DEFAULT, pcmk__valid_boolean,
N_("Allow performing fencing operations in parallel"),
NULL
},
{
PCMK_OPT_STARTUP_FENCING, NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("*** Advanced Use Only *** Whether to fence unseen nodes at start-up"),
N_("Setting this to false may lead to a \"split-brain\" situation,"
"potentially leading to data loss and/or service unavailability.")
},
{
PCMK_OPT_PRIORITY_FENCING_DELAY, NULL, "time", NULL,
"0", pcmk__valid_interval_spec,
N_("Apply fencing delay targeting the lost nodes with the highest total resource priority"),
N_("Apply specified delay for the fencings that are targeting the lost "
"nodes with the highest total resource priority in case we don't "
"have the majority of the nodes in our cluster partition, so that "
"the more significant nodes potentially win any fencing match, "
"which is especially meaningful under split-brain of 2-node "
"cluster. A promoted resource instance takes the base priority + 1 "
"on calculation if the base priority is not 0. Any static/random "
"delays that are introduced by `pcmk_delay_base/max` configured "
"for the corresponding fencing resources will be added to this "
"delay. This delay should be significantly greater than, safely "
"twice, the maximum `pcmk_delay_base/max`. By default, priority "
"fencing delay is disabled.")
},
{
PCMK_OPT_NODE_PENDING_TIMEOUT, NULL, "time", NULL,
"0", pcmk__valid_interval_spec,
N_("How long to wait for a node that has joined the cluster to join "
"the controller process group"),
N_("Fence nodes that do not join the controller process group within "
"this much time after joining the cluster, to allow the cluster "
"to continue managing resources. A value of 0 means never fence "
"pending nodes. Setting the value to 2h means fence nodes after "
"2 hours.")
},
{
PCMK_OPT_CLUSTER_DELAY, NULL, "time", NULL,
"60s", pcmk__valid_interval_spec,
N_("Maximum time for node-to-node communication"),
N_("The node elected Designated Controller (DC) will consider an action "
"failed if it does not get a response from the node executing the "
"action within this time (after considering the action's own "
"timeout). The \"correct\" value will depend on the speed and "
"load of your network and cluster nodes.")
},
{
PCMK_OPT_BATCH_LIMIT, NULL, "integer", NULL,
"0", pcmk__valid_number,
N_("Maximum number of jobs that the cluster may execute in parallel "
"across all nodes"),
N_("The \"correct\" value will depend on the speed and load of your "
"network and cluster nodes. If set to 0, the cluster will "
"impose a dynamically calculated limit when any node has a "
"high load.")
},
{
PCMK_OPT_MIGRATION_LIMIT, NULL, "integer", NULL,
"-1", pcmk__valid_number,
N_("The number of live migration actions that the cluster is allowed "
"to execute in parallel on a node (-1 means no limit)")
},
/* Orphans and stopping */
{
- "stop-all-resources", NULL, "boolean", NULL,
+ PCMK_OPT_STOP_ALL_RESOURCES, NULL, "boolean", NULL,
"false", pcmk__valid_boolean,
N_("Whether the cluster should stop all active resources"),
NULL
},
{
"stop-orphan-resources", NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("Whether to stop resources that were removed from the configuration"),
NULL
},
{
"stop-orphan-actions", NULL, "boolean", NULL,
"true", pcmk__valid_boolean,
N_("Whether to cancel recurring actions removed from the configuration"),
NULL
},
{
PCMK__OPT_REMOVE_AFTER_STOP, NULL, "boolean", NULL,
"false", pcmk__valid_boolean,
N_("*** Deprecated *** Whether to remove stopped resources from "
"the executor"),
N_("Values other than default are poorly tested and potentially dangerous."
" This option will be removed in a future release.")
},
/* Storing inputs */
{
PCMK_OPT_PE_ERROR_SERIES_MAX, NULL, "integer", NULL,
"-1", pcmk__valid_number,
N_("The number of scheduler inputs resulting in errors to save"),
N_("Zero to disable, -1 to store unlimited.")
},
{
PCMK_OPT_PE_WARN_SERIES_MAX, NULL, "integer", NULL,
"5000", pcmk__valid_number,
N_("The number of scheduler inputs resulting in warnings to save"),
N_("Zero to disable, -1 to store unlimited.")
},
{
PCMK_OPT_PE_INPUT_SERIES_MAX, NULL, "integer", NULL,
"4000", pcmk__valid_number,
N_("The number of scheduler inputs without errors or warnings to save"),
N_("Zero to disable, -1 to store unlimited.")
},
/* Node health */
{
PCMK__OPT_NODE_HEALTH_STRATEGY, NULL, "select",
PCMK__VALUE_NONE ", " PCMK__VALUE_MIGRATE_ON_RED ", "
PCMK__VALUE_ONLY_GREEN ", " PCMK__VALUE_PROGRESSIVE ", "
PCMK__VALUE_CUSTOM,
PCMK__VALUE_NONE, pcmk__validate_health_strategy,
N_("How cluster should react to node health attributes"),
N_("Requires external entities to create node attributes (named with "
"the prefix \"#health\") with values \"red\", "
"\"yellow\", or \"green\".")
},
{
PCMK__OPT_NODE_HEALTH_BASE, NULL, "integer", NULL,
"0", pcmk__valid_number,
N_("Base health score assigned to a node"),
N_("Only used when \"node-health-strategy\" is set to \"progressive\".")
},
{
PCMK__OPT_NODE_HEALTH_GREEN, NULL, "integer", NULL,
"0", pcmk__valid_number,
N_("The score to use for a node health attribute whose value is \"green\""),
N_("Only used when \"node-health-strategy\" is set to \"custom\" or \"progressive\".")
},
{
PCMK__OPT_NODE_HEALTH_YELLOW, NULL, "integer", NULL,
"0", pcmk__valid_number,
N_("The score to use for a node health attribute whose value is \"yellow\""),
N_("Only used when \"node-health-strategy\" is set to \"custom\" or \"progressive\".")
},
{
PCMK__OPT_NODE_HEALTH_RED, NULL, "integer", NULL,
"-INFINITY", pcmk__valid_number,
N_("The score to use for a node health attribute whose value is \"red\""),
N_("Only used when \"node-health-strategy\" is set to \"custom\" or \"progressive\".")
},
/*Placement Strategy*/
{
PCMK_OPT_PLACEMENT_STRATEGY, NULL, "select",
"default, utilization, minimal, balanced",
"default", check_placement_strategy,
N_("How the cluster should allocate resources to nodes"),
NULL
},
};
void
pe_metadata(pcmk__output_t *out)
{
const char *desc_short = "Pacemaker scheduler options";
const char *desc_long = "Cluster options used by Pacemaker's scheduler";
gchar *s = pcmk__format_option_metadata("pacemaker-schedulerd", desc_short,
desc_long, pe_opts,
PCMK__NELEM(pe_opts));
out->output_xml(out, "metadata", s);
g_free(s);
}
void
verify_pe_options(GHashTable * options)
{
pcmk__validate_cluster_options(options, pe_opts, PCMK__NELEM(pe_opts));
}
const char *
pe_pref(GHashTable * options, const char *name)
{
return pcmk__cluster_option(options, pe_opts, PCMK__NELEM(pe_opts), name);
}
const char *
fail2text(enum action_fail_response fail)
{
const char *result = "<unknown>";
switch (fail) {
case pcmk_on_fail_ignore:
result = "ignore";
break;
case pcmk_on_fail_demote:
result = "demote";
break;
case pcmk_on_fail_block:
result = "block";
break;
case pcmk_on_fail_restart:
result = "recover";
break;
case pcmk_on_fail_ban:
result = "migrate";
break;
case pcmk_on_fail_stop:
result = "stop";
break;
case pcmk_on_fail_fence_node:
result = "fence";
break;
case pcmk_on_fail_standby_node:
result = "standby";
break;
case pcmk_on_fail_restart_container:
result = "restart-container";
break;
case pcmk_on_fail_reset_remote:
result = "reset-remote";
break;
}
return result;
}
enum action_tasks
text2task(const char *task)
{
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei)) {
return pcmk_action_stop;
} else if (pcmk__str_eq(task, PCMK_ACTION_STOPPED, pcmk__str_casei)) {
return pcmk_action_stopped;
} else if (pcmk__str_eq(task, PCMK_ACTION_START, pcmk__str_casei)) {
return pcmk_action_start;
} else if (pcmk__str_eq(task, PCMK_ACTION_RUNNING, pcmk__str_casei)) {
return pcmk_action_started;
} else if (pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_casei)) {
return pcmk_action_shutdown;
} else if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) {
return pcmk_action_fence;
} else if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
return pcmk_action_monitor;
} else if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_casei)) {
return pcmk_action_notify;
} else if (pcmk__str_eq(task, PCMK_ACTION_NOTIFIED, pcmk__str_casei)) {
return pcmk_action_notified;
} else if (pcmk__str_eq(task, PCMK_ACTION_PROMOTE, pcmk__str_casei)) {
return pcmk_action_promote;
} else if (pcmk__str_eq(task, PCMK_ACTION_DEMOTE, pcmk__str_casei)) {
return pcmk_action_demote;
} else if (pcmk__str_eq(task, PCMK_ACTION_PROMOTED, pcmk__str_casei)) {
return pcmk_action_promoted;
} else if (pcmk__str_eq(task, PCMK_ACTION_DEMOTED, pcmk__str_casei)) {
return pcmk_action_demoted;
}
return pcmk_action_unspecified;
}
const char *
task2text(enum action_tasks task)
{
const char *result = "<unknown>";
switch (task) {
case pcmk_action_unspecified:
result = "no_action";
break;
case pcmk_action_stop:
result = PCMK_ACTION_STOP;
break;
case pcmk_action_stopped:
result = PCMK_ACTION_STOPPED;
break;
case pcmk_action_start:
result = PCMK_ACTION_START;
break;
case pcmk_action_started:
result = PCMK_ACTION_RUNNING;
break;
case pcmk_action_shutdown:
result = PCMK_ACTION_DO_SHUTDOWN;
break;
case pcmk_action_fence:
result = PCMK_ACTION_STONITH;
break;
case pcmk_action_monitor:
result = PCMK_ACTION_MONITOR;
break;
case pcmk_action_notify:
result = PCMK_ACTION_NOTIFY;
break;
case pcmk_action_notified:
result = PCMK_ACTION_NOTIFIED;
break;
case pcmk_action_promote:
result = PCMK_ACTION_PROMOTE;
break;
case pcmk_action_promoted:
result = PCMK_ACTION_PROMOTED;
break;
case pcmk_action_demote:
result = PCMK_ACTION_DEMOTE;
break;
case pcmk_action_demoted:
result = PCMK_ACTION_DEMOTED;
break;
}
return result;
}
const char *
role2text(enum rsc_role_e role)
{
switch (role) {
case pcmk_role_stopped:
return PCMK__ROLE_STOPPED;
case pcmk_role_started:
return PCMK__ROLE_STARTED;
case pcmk_role_unpromoted:
#ifdef PCMK__COMPAT_2_0
return PCMK__ROLE_UNPROMOTED_LEGACY;
#else
return PCMK__ROLE_UNPROMOTED;
#endif
case pcmk_role_promoted:
#ifdef PCMK__COMPAT_2_0
return PCMK__ROLE_PROMOTED_LEGACY;
#else
return PCMK__ROLE_PROMOTED;
#endif
default: // pcmk_role_unknown
return PCMK__ROLE_UNKNOWN;
}
}
enum rsc_role_e
text2role(const char *role)
{
if (pcmk__str_eq(role, PCMK__ROLE_UNKNOWN,
pcmk__str_casei|pcmk__str_null_matches)) {
return pcmk_role_unknown;
} else if (pcmk__str_eq(role, PCMK__ROLE_STOPPED, pcmk__str_casei)) {
return pcmk_role_stopped;
} else if (pcmk__str_eq(role, PCMK__ROLE_STARTED, pcmk__str_casei)) {
return pcmk_role_started;
} else if (pcmk__strcase_any_of(role, PCMK__ROLE_UNPROMOTED,
PCMK__ROLE_UNPROMOTED_LEGACY, NULL)) {
return pcmk_role_unpromoted;
} else if (pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)) {
return pcmk_role_promoted;
}
return pcmk_role_unknown; // Invalid role given
}
void
add_hash_param(GHashTable * hash, const char *name, const char *value)
{
CRM_CHECK(hash != NULL, return);
crm_trace("Adding name='%s' value='%s' to hash table",
pcmk__s(name, "<null>"), pcmk__s(value, "<null>"));
if (name == NULL || value == NULL) {
return;
} else if (pcmk__str_eq(value, "#default", pcmk__str_casei)) {
return;
} else if (g_hash_table_lookup(hash, name) == NULL) {
g_hash_table_insert(hash, strdup(name), strdup(value));
}
}
/*!
* \internal
* \brief Look up an attribute value on the appropriate node
*
* If \p node is a guest node and either the \c XML_RSC_ATTR_TARGET meta
* attribute is set to "host" for \p rsc or \p force_host is \c true, query the
* attribute on the node's host. Otherwise, query the attribute on \p node
* itself.
*
* \param[in] node Node to query attribute value on by default
* \param[in] name Name of attribute to query
* \param[in] rsc Resource on whose behalf we're querying
* \param[in] node_type Type of resource location lookup
* \param[in] force_host Force a lookup on the guest node's host, regardless of
* the \c XML_RSC_ATTR_TARGET value
*
* \return Value of the attribute on \p node or on the host of \p node
*
* \note If \p force_host is \c true, \p node \e must be a guest node.
*/
const char *
pe__node_attribute_calculated(const pcmk_node_t *node, const char *name,
const pcmk_resource_t *rsc,
enum pcmk__rsc_node node_type,
bool force_host)
{
// @TODO: Use pe__is_guest_node() after merging libpe_{rules,status}
bool is_guest = (node != NULL)
&& (node->details->type == pcmk_node_variant_remote)
&& (node->details->remote_rsc != NULL)
&& (node->details->remote_rsc->container != NULL);
const char *source = NULL;
const char *node_type_s = NULL;
const char *reason = NULL;
const pcmk_resource_t *container = NULL;
const pcmk_node_t *host = NULL;
CRM_ASSERT((node != NULL) && (name != NULL) && (rsc != NULL)
&& (!force_host || is_guest));
/* Ignore XML_RSC_ATTR_TARGET if node is not a guest node. This represents a
* user configuration error.
*/
source = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET);
if (!force_host
&& (!is_guest || !pcmk__str_eq(source, "host", pcmk__str_casei))) {
return g_hash_table_lookup(node->details->attrs, name);
}
container = node->details->remote_rsc->container;
switch (node_type) {
case pcmk__rsc_node_assigned:
node_type_s = "assigned";
host = container->allocated_to;
if (host == NULL) {
reason = "not assigned";
}
break;
case pcmk__rsc_node_current:
node_type_s = "current";
if (container->running_on != NULL) {
host = container->running_on->data;
}
if (host == NULL) {
reason = "inactive";
}
break;
default:
// Add support for other enum pcmk__rsc_node values if needed
CRM_ASSERT(false);
break;
}
if (host != NULL) {
const char *value = g_hash_table_lookup(host->details->attrs, name);
pcmk__rsc_trace(rsc,
"%s: Value lookup for %s on %s container host %s %s%s",
rsc->id, name, node_type_s, pe__node_name(host),
((value != NULL)? "succeeded: " : "failed"),
pcmk__s(value, ""));
return value;
}
pcmk__rsc_trace(rsc,
"%s: Not looking for %s on %s container host: %s is %s",
rsc->id, name, node_type_s, container->id, reason);
return NULL;
}
const char *
pe_node_attribute_raw(const pcmk_node_t *node, const char *name)
{
if(node == NULL) {
return NULL;
}
return g_hash_table_lookup(node->details->attrs, name);
}
diff --git a/lib/pengine/pe_output.c b/lib/pengine/pe_output.c
index aaba061c54..2aa5cb38d6 100644
--- a/lib/pengine/pe_output.c
+++ b/lib/pengine/pe_output.c
@@ -1,3173 +1,3174 @@
/*
* Copyright 2019-2024 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 <crm_internal.h>
#include <stdint.h>
#include <crm/common/xml_internal.h>
#include <crm/common/output.h>
#include <crm/common/scheduler_internal.h>
#include <crm/cib/util.h>
#include <crm/msg_xml.h>
#include <crm/pengine/internal.h>
const char *
pe__resource_description(const pcmk_resource_t *rsc, uint32_t show_opts)
{
const char * desc = NULL;
// User-supplied description
if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description)) {
desc = crm_element_value(rsc->xml, XML_ATTR_DESC);
}
return desc;
}
/* Never display node attributes whose name starts with one of these prefixes */
#define FILTER_STR { PCMK__FAIL_COUNT_PREFIX, PCMK__LAST_FAILURE_PREFIX, \
"shutdown", PCMK_NODE_ATTR_TERMINATE, "standby", "#", \
NULL }
static int
compare_attribute(gconstpointer a, gconstpointer b)
{
int rc;
rc = strcmp((const char *)a, (const char *)b);
return rc;
}
/*!
* \internal
* \brief Determine whether extended information about an attribute should be added.
*
* \param[in] node Node that ran this resource
* \param[in,out] rsc_list List of resources for this node
* \param[in,out] scheduler Scheduler data
* \param[in] attrname Attribute to find
* \param[out] expected_score Expected value for this attribute
*
* \return true if extended information should be printed, false otherwise
* \note Currently, extended information is only supported for ping/pingd
* resources, for which a message will be printed if connectivity is lost
* or degraded.
*/
static bool
add_extra_info(const pcmk_node_t *node, GList *rsc_list,
pcmk_scheduler_t *scheduler, const char *attrname,
int *expected_score)
{
GList *gIter = NULL;
for (gIter = rsc_list; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
const char *type = g_hash_table_lookup(rsc->meta, "type");
const char *name = NULL;
GHashTable *params = NULL;
if (rsc->children != NULL) {
if (add_extra_info(node, rsc->children, scheduler, attrname,
expected_score)) {
return true;
}
}
if (!pcmk__strcase_any_of(type, "ping", "pingd", NULL)) {
continue;
}
params = pe_rsc_params(rsc, node, scheduler);
name = g_hash_table_lookup(params, "name");
if (name == NULL) {
name = "pingd";
}
/* To identify the resource with the attribute name. */
if (pcmk__str_eq(name, attrname, pcmk__str_casei)) {
int host_list_num = 0;
const char *hosts = g_hash_table_lookup(params, "host_list");
const char *multiplier = g_hash_table_lookup(params, "multiplier");
int multiplier_i;
if (hosts) {
char **host_list = g_strsplit(hosts, " ", 0);
host_list_num = g_strv_length(host_list);
g_strfreev(host_list);
}
if ((multiplier == NULL)
|| (pcmk__scan_min_int(multiplier, &multiplier_i,
INT_MIN) != pcmk_rc_ok)) {
/* The ocf:pacemaker:ping resource agent defaults multiplier to
* 1. The agent currently does not handle invalid text, but it
* should, and this would be a reasonable choice ...
*/
multiplier_i = 1;
}
*expected_score = host_list_num * multiplier_i;
return true;
}
}
return false;
}
static GList *
filter_attr_list(GList *attr_list, char *name)
{
int i;
const char *filt_str[] = FILTER_STR;
CRM_CHECK(name != NULL, return attr_list);
/* filtering automatic attributes */
for (i = 0; filt_str[i] != NULL; i++) {
if (g_str_has_prefix(name, filt_str[i])) {
return attr_list;
}
}
return g_list_insert_sorted(attr_list, name, compare_attribute);
}
static GList *
get_operation_list(xmlNode *rsc_entry) {
GList *op_list = NULL;
xmlNode *rsc_op = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL;
rsc_op = pcmk__xe_next(rsc_op)) {
const char *task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
const char *interval_ms_s = crm_element_value(rsc_op,
XML_LRM_ATTR_INTERVAL_MS);
const char *op_rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC);
int op_rc_i;
pcmk__scan_min_int(op_rc, &op_rc_i, 0);
/* Display 0-interval monitors as "probe" */
if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) {
task = "probe";
}
/* Ignore notifies and some probes */
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)
|| (pcmk__str_eq(task, "probe", pcmk__str_none)
&& (op_rc_i == CRM_EX_NOT_RUNNING))) {
continue;
}
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) {
op_list = g_list_append(op_list, rsc_op);
}
}
op_list = g_list_sort(op_list, sort_op_by_callid);
return op_list;
}
static void
add_dump_node(gpointer key, gpointer value, gpointer user_data)
{
xmlNodePtr node = user_data;
pcmk_create_xml_text_node(node, (const char *) key, (const char *) value);
}
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;
}
#define XPATH_STACK "//" XML_CIB_TAG_NVPAIR \
"[@" XML_NVPAIR_ATTR_NAME "='" \
PCMK_OPT_CLUSTER_INFRASTRUCTURE "']"
static const char *
get_cluster_stack(pcmk_scheduler_t *scheduler)
{
xmlNode *stack = get_xpath_object(XPATH_STACK, scheduler->input, LOG_DEBUG);
return stack? crm_element_value(stack, XML_NVPAIR_ATTR_VALUE) : "unknown";
}
static char *
last_changed_string(const char *last_written, const char *user,
const char *client, const char *origin) {
if (last_written != NULL || user != NULL || client != NULL || origin != NULL) {
return crm_strdup_printf("%s%s%s%s%s%s%s",
last_written ? last_written : "",
user ? " by " : "",
user ? user : "",
client ? " via " : "",
client ? client : "",
origin ? " on " : "",
origin ? origin : "");
} else {
return strdup("");
}
}
static char *
op_history_string(xmlNode *xml_op, const char *task, const char *interval_ms_s,
int rc, bool print_timing) {
const char *call = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
char *interval_str = NULL;
char *buf = NULL;
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *pair = pcmk__format_nvpair("interval", interval_ms_s, "ms");
interval_str = crm_strdup_printf(" %s", pair);
free(pair);
}
if (print_timing) {
char *last_change_str = NULL;
char *exec_str = NULL;
char *queue_str = NULL;
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &epoch) == pcmk_ok)
&& (epoch > 0)) {
char *epoch_str = pcmk__epoch2str(&epoch, 0);
last_change_str = crm_strdup_printf(" %s=\"%s\"",
XML_RSC_OP_LAST_CHANGE,
pcmk__s(epoch_str, ""));
free(epoch_str);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
if (value) {
char *pair = pcmk__format_nvpair(XML_RSC_OP_T_EXEC, value, "ms");
exec_str = crm_strdup_printf(" %s", pair);
free(pair);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
if (value) {
char *pair = pcmk__format_nvpair(XML_RSC_OP_T_QUEUE, value, "ms");
queue_str = crm_strdup_printf(" %s", pair);
free(pair);
}
buf = crm_strdup_printf("(%s) %s:%s%s%s%s rc=%d (%s)", call, task,
interval_str ? interval_str : "",
last_change_str ? last_change_str : "",
exec_str ? exec_str : "",
queue_str ? queue_str : "",
rc, services_ocf_exitcode_str(rc));
if (last_change_str) {
free(last_change_str);
}
if (exec_str) {
free(exec_str);
}
if (queue_str) {
free(queue_str);
}
} else {
buf = crm_strdup_printf("(%s) %s%s%s", call, task,
interval_str ? ":" : "",
interval_str ? interval_str : "");
}
if (interval_str) {
free(interval_str);
}
return buf;
}
static char *
resource_history_string(pcmk_resource_t *rsc, const char *rsc_id, bool all,
int failcount, time_t last_failure) {
char *buf = NULL;
if (rsc == NULL) {
buf = crm_strdup_printf("%s: orphan", rsc_id);
} else if (all || failcount || last_failure > 0) {
char *failcount_s = NULL;
char *lastfail_s = NULL;
if (failcount > 0) {
failcount_s = crm_strdup_printf(" %s=%d", PCMK__FAIL_COUNT_PREFIX,
failcount);
} else {
failcount_s = strdup("");
}
if (last_failure > 0) {
buf = pcmk__epoch2str(&last_failure, 0);
lastfail_s = crm_strdup_printf(" %s='%s'",
PCMK__LAST_FAILURE_PREFIX, buf);
free(buf);
}
buf = crm_strdup_printf("%s: migration-threshold=%d%s%s",
rsc_id, rsc->migration_threshold, failcount_s,
lastfail_s? lastfail_s : "");
free(failcount_s);
free(lastfail_s);
} else {
buf = crm_strdup_printf("%s:", rsc_id);
}
return buf;
}
/*!
* \internal
* \brief Get a node's feature set for status display purposes
*
* \param[in] node Node to check
*
* \return String representation of feature set if the node is fully up (using
* "<3.15.1" for older nodes that don't set the #feature-set attribute),
* otherwise NULL
*/
static const char *
get_node_feature_set(const pcmk_node_t *node)
{
if (node->details->online && node->details->expected_up
&& !pe__is_guest_or_remote_node(node)) {
const char *feature_set = g_hash_table_lookup(node->details->attrs,
CRM_ATTR_FEATURE_SET);
/* The feature set attribute is present since 3.15.1. If it is missing,
* then the node must be running an earlier version.
*/
return pcmk__s(feature_set, "<3.15.1");
}
return NULL;
}
static bool
is_mixed_version(pcmk_scheduler_t *scheduler)
{
const char *feature_set = NULL;
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
const char *node_feature_set = get_node_feature_set(node);
if (node_feature_set != NULL) {
if (feature_set == NULL) {
feature_set = node_feature_set;
} else if (strcmp(feature_set, node_feature_set) != 0) {
return true;
}
}
}
return false;
}
static char *
formatted_xml_buf(const pcmk_resource_t *rsc, bool raw)
{
if (raw) {
return dump_xml_formatted(rsc->orig_xml ? rsc->orig_xml : rsc->xml);
} else {
return dump_xml_formatted(rsc->xml);
}
}
#define XPATH_DC_VERSION "//" XML_CIB_TAG_NVPAIR \
"[@" XML_NVPAIR_ATTR_NAME "='" \
PCMK_OPT_DC_VERSION "']"
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
if (pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, XML_NVPAIR_ATTR_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
XML_ATTR_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
XML_CIB_ATTR_WRITTEN);
const char *user = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_ORIG);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times",
scheduler->localhost, last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->ninstances, scheduler->disabled_resources,
scheduler->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
/* Always print DC if none, even if not requested */
if ((scheduler->dc_node == NULL)
|| pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, XML_NVPAIR_ATTR_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
XML_ATTR_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
XML_CIB_ATTR_WRITTEN);
const char *user = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
XML_ATTR_UPDATE_ORIG);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times",
scheduler->localhost, last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->ninstances, scheduler->disabled_resources,
scheduler->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
/* Kind of a hack - close the list we may have opened earlier in this
* function so we can put all the options into their own list. We
* only want to do this on HTML output, though.
*/
PCMK__OUTPUT_LIST_FOOTER(out, rc);
out->begin_list(out, NULL, NULL, "Config Options");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
char *
pe__node_display_name(pcmk_node_t *node, bool print_detail)
{
char *node_name;
const char *node_host = NULL;
const char *node_id = NULL;
int name_len;
CRM_ASSERT((node != NULL) && (node->details != NULL) && (node->details->uname != NULL));
/* Host is displayed only if this is a guest node and detail is requested */
if (print_detail && pe__is_guest_node(node)) {
const pcmk_resource_t *container = node->details->remote_rsc->container;
const pcmk_node_t *host_node = pe__current_node(container);
if (host_node && host_node->details) {
node_host = host_node->details->uname;
}
if (node_host == NULL) {
node_host = ""; /* so we at least get "uname@" to indicate guest */
}
}
/* Node ID is displayed if different from uname and detail is requested */
if (print_detail && !pcmk__str_eq(node->details->uname, node->details->id, pcmk__str_casei)) {
node_id = node->details->id;
}
/* Determine name length */
name_len = strlen(node->details->uname) + 1;
if (node_host) {
name_len += strlen(node_host) + 1; /* "@node_host" */
}
if (node_id) {
name_len += strlen(node_id) + 3; /* + " (node_id)" */
}
/* Allocate and populate display name */
node_name = malloc(name_len);
CRM_ASSERT(node_name != NULL);
strcpy(node_name, node->details->uname);
if (node_host) {
strcat(node_name, "@");
strcat(node_name, node_host);
}
if (node_id) {
strcat(node_name, " (");
strcat(node_name, node_id);
strcat(node_name, ")");
}
return node_name;
}
int
pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name
, size_t pairs_count, ...)
{
xmlNodePtr xml_node = NULL;
va_list args;
CRM_ASSERT(tag_name != NULL);
xml_node = pcmk__output_xml_peek_parent(out);
CRM_ASSERT(xml_node != NULL);
xml_node = create_xml_node(xml_node, tag_name);
va_start(args, pairs_count);
while(pairs_count--) {
const char *param_name = va_arg(args, const char *);
const char *param_value = va_arg(args, const char *);
if (param_name && param_value) {
crm_xml_add(xml_node, param_name, param_value);
}
};
va_end(args);
if (is_list) {
pcmk__output_xml_push_parent(out, xml_node);
}
return pcmk_rc_ok;
}
static const char *
role_desc(enum rsc_role_e role)
{
if (role == pcmk_role_promoted) {
#ifdef PCMK__COMPAT_2_0
return "as " PCMK__ROLE_PROMOTED_LEGACY " ";
#else
return "in " PCMK__ROLE_PROMOTED " role ";
#endif
}
return "";
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
char *buf = crm_strdup_printf("%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
pcmk__output_create_html_node(out, "li", NULL, NULL, buf);
free(node_name);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
out->list_item(out, NULL, "%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
free(node_name);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *promoted_only = pcmk__btoa(location->role_filter == pcmk_role_promoted);
char *weight_s = pcmk__itoa(pe_node->weight);
pcmk__output_create_xml_node(out, "ban",
"id", location->id,
"resource", location->rsc->id,
"node", pe_node->details->uname,
"weight", weight_s,
"promoted-only", promoted_only,
/* This is a deprecated alias for
* promoted_only. Removing it will break
* backward compatibility of the API schema,
* which will require an API schema major
* version bump.
*/
"master_only", promoted_only,
NULL);
free(weight_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban-list", "pcmk_scheduler_t *", "const char *", "GList *",
"uint32_t", "bool")
static int
ban_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *prefix = va_arg(args, const char *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *gIter, *gIter2;
int rc = pcmk_rc_no_output;
/* Print each ban */
for (gIter = scheduler->placement_constraints;
gIter != NULL; gIter = gIter->next) {
pcmk__location_t *location = gIter->data;
const pcmk_resource_t *rsc = location->rsc;
if (prefix != NULL && !g_str_has_prefix(location->id, prefix)) {
continue;
}
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(pe__const_top_resource(rsc, false)),
only_rsc, pcmk__str_star_matches)) {
continue;
}
for (gIter2 = location->nodes; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter2->data;
if (node->weight < 0) {
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Negative Location Constraints");
out->message(out, "ban", node, location, show_opts);
}
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_html(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "li", NULL);
char *nnodes_str = crm_strdup_printf("%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
pcmk_create_html_node(nodes_node, "span", NULL, NULL, nnodes_str);
free(nnodes_str);
if (ndisabled && nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
ndisabled);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED");
s = crm_strdup_printf(", %d ", nblocked);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED");
pcmk_create_html_node(resources_node, "span", NULL, NULL,
" from further action due to failure)");
} else if (ndisabled && !nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
ndisabled);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED");
pcmk_create_html_node(resources_node, "span", NULL, NULL, ")");
} else if (!ndisabled && nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
nblocked);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED");
pcmk_create_html_node(resources_node, "span", NULL, NULL,
" from further action due to failure)");
} else {
char *s = crm_strdup_printf("%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_text(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
out->list_item(out, NULL, "%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
if (ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED, %d BLOCKED from "
"further action due to failure)",
nresources, pcmk__plural_s(nresources), ndisabled,
nblocked);
} else if (ndisabled && !nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED)",
nresources, pcmk__plural_s(nresources), ndisabled);
} else if (!ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d BLOCKED from further action "
"due to failure)",
nresources, pcmk__plural_s(nresources), nblocked);
} else {
out->list_item(out, NULL, "%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_xml(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "nodes_configured", NULL);
xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "resources_configured", NULL);
char *s = pcmk__itoa(nnodes);
crm_xml_add(nodes_node, "number", s);
free(s);
s = pcmk__itoa(nresources);
crm_xml_add(resources_node, "number", s);
free(s);
s = pcmk__itoa(ndisabled);
crm_xml_add(resources_node, "disabled", s);
free(s);
s = pcmk__itoa(nblocked);
crm_xml_add(resources_node, "blocked", s);
free(s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, "bold", "Current DC: ");
if (dc) {
char *buf = crm_strdup_printf("%s (version %s) -", dc_name,
dc_version_s ? dc_version_s : "unknown");
pcmk_create_html_node(node, "span", NULL, NULL, buf);
free(buf);
if (mixed_version) {
pcmk_create_html_node(node, "span", NULL, "warning",
" MIXED-VERSION");
}
pcmk_create_html_node(node, "span", NULL, NULL, " partition");
if (crm_is_true(quorum)) {
pcmk_create_html_node(node, "span", NULL, NULL, " with");
} else {
pcmk_create_html_node(node, "span", NULL, "warning", " WITHOUT");
}
pcmk_create_html_node(node, "span", NULL, NULL, " quorum");
} else {
pcmk_create_html_node(node, "span", NULL, "warning", "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
out->list_item(out, "Current DC",
"%s (version %s) - %spartition %s quorum",
dc_name, dc_version_s ? dc_version_s : "unknown",
mixed_version ? "MIXED-VERSION " : "",
crm_is_true(quorum) ? "with" : "WITHOUT");
} else {
out->list_item(out, "Current DC", "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name G_GNUC_UNUSED = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
pcmk__output_create_xml_node(out, "current_dc",
"present", "true",
"version", dc_version_s ? dc_version_s : "",
"name", dc->details->uname,
"id", dc->details->id,
"with_quorum", pcmk__btoa(crm_is_true(quorum)),
"mixed_version", pcmk__btoa(mixed_version),
NULL);
} else {
pcmk__output_create_xml_node(out, "current_dc",
"present", "false",
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("maint-mode", "unsigned long long int")
static int
cluster_maint_mode_text(pcmk__output_t *out, va_list args) {
unsigned long long flags = va_arg(args, unsigned long long);
if (pcmk_is_set(flags, pcmk_sched_in_maintenance)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will not attempt to start, stop or recover services\n");
return pcmk_rc_ok;
} else if (pcmk_is_set(flags, pcmk_sched_stop_all)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will keep all resources stopped\n");
return pcmk_rc_ok;
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL, "No quorum policy: Suicide");
break;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: ");
pcmk_create_html_node(node, "span", NULL, "bold", "DISABLED");
pcmk_create_html_node(node, "span", NULL, NULL,
" (the cluster will not attempt to start, stop, or recover services)");
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: ");
pcmk_create_html_node(node, "span", NULL, "bold", "STOPPED");
pcmk_create_html_node(node, "span", NULL, NULL,
" (the cluster will keep all resources stopped)");
} else {
out->list_item(out, NULL, "Resource management: enabled");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_log(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
return out->info(out, "Resource management is DISABLED. The cluster will not attempt to start, stop or recover services.");
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) {
return out->info(out, "Resource management is DISABLED. The cluster has stopped all resources.");
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_text(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL, "No quorum policy: Suicide");
break;
}
return pcmk_rc_ok;
}
#define bv(flag) pcmk__btoa(pcmk_is_set(scheduler->flags, (flag)))
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *no_quorum_policy = NULL;
char *stonith_timeout_str = pcmk__itoa(scheduler->stonith_timeout);
char *priority_fencing_delay_str = pcmk__itoa(scheduler->priority_fencing_delay * 1000);
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
no_quorum_policy = "freeze";
break;
case pcmk_no_quorum_stop:
no_quorum_policy = "stop";
break;
case pcmk_no_quorum_demote:
no_quorum_policy = "demote";
break;
case pcmk_no_quorum_ignore:
no_quorum_policy = "ignore";
break;
case pcmk_no_quorum_fence:
no_quorum_policy = "suicide";
break;
}
pcmk__output_create_xml_node(out, "cluster_options",
PCMK_OPT_STONITH_ENABLED,
bv(pcmk_sched_fencing_enabled),
"symmetric-cluster",
bv(pcmk_sched_symmetric_cluster),
PCMK_OPT_NO_QUORUM_POLICY, no_quorum_policy,
PCMK_OPT_MAINTENANCE_MODE,
bv(pcmk_sched_in_maintenance),
- "stop-all-resources", bv(pcmk_sched_stop_all),
+ PCMK_OPT_STOP_ALL_RESOURCES,
+ bv(pcmk_sched_stop_all),
PCMK_OPT_STONITH_TIMEOUT "-ms",
stonith_timeout_str,
PCMK_OPT_PRIORITY_FENCING_DELAY "-ms",
priority_fencing_delay_str,
NULL);
free(stonith_timeout_str);
free(priority_fencing_delay_str);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_html(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, "bold", "Stack: ");
pcmk_create_html_node(node, "span", NULL, NULL, stack_s);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
pcmk_create_html_node(node, "span", NULL, NULL, " (");
pcmk_create_html_node(node, "span", NULL, NULL,
pcmk__pcmkd_state_enum2friendly(pcmkd_state));
pcmk_create_html_node(node, "span", NULL, NULL, ")");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_text(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
out->list_item(out, "Stack", "%s (%s)",
stack_s, pcmk__pcmkd_state_enum2friendly(pcmkd_state));
} else {
out->list_item(out, "Stack", "%s", stack_s);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_xml(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
const char *state_s = NULL;
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
state_s = pcmk_pacemakerd_api_daemon_state_enum2text(pcmkd_state);
}
pcmk__output_create_xml_node(out, "stack",
"type", stack_s,
"pacemakerd-state", state_s,
NULL);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_html(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
xmlNodePtr updated_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr changed_node = pcmk__output_create_xml_node(out, "li", NULL);
char *time_s = pcmk__epoch2str(NULL, 0);
pcmk_create_html_node(updated_node, "span", NULL, "bold", "Last updated: ");
pcmk_create_html_node(updated_node, "span", NULL, NULL, time_s);
if (our_nodename != NULL) {
pcmk_create_html_node(updated_node, "span", NULL, NULL, " on ");
pcmk_create_html_node(updated_node, "span", NULL, NULL, our_nodename);
}
free(time_s);
time_s = last_changed_string(last_written, user, client, origin);
pcmk_create_html_node(changed_node, "span", NULL, "bold", "Last change: ");
pcmk_create_html_node(changed_node, "span", NULL, NULL, time_s);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_xml(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
pcmk__output_create_xml_node(out, "last_update",
"time", time_s,
"origin", our_nodename,
NULL);
pcmk__output_create_xml_node(out, "last_change",
"time", last_written ? last_written : "",
"user", user ? user : "",
"client", client ? client : "",
"origin", origin ? origin : "",
NULL);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_text(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
out->list_item(out, "Last updated", "%s%s%s",
time_s, (our_nodename != NULL)? " on " : "",
pcmk__s(our_nodename, ""));
free(time_s);
time_s = last_changed_string(last_written, user, client, origin);
out->list_item(out, "Last change", " %s", time_s);
free(time_s);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Display a failed action in less-technical natural language
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_friendly(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
char *rsc_id = NULL;
char *task = NULL;
guint interval_ms = 0;
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)
|| !parse_op_key(op_key, &rsc_id, &task, &interval_ms)) {
rsc_id = strdup("unknown resource");
task = strdup("unknown action");
interval_ms = 0;
}
CRM_ASSERT((rsc_id != NULL) && (task != NULL));
str = g_string_sized_new(256); // Should be sufficient for most messages
pcmk__g_strcat(str, rsc_id, " ", NULL);
if (interval_ms != 0) {
pcmk__g_strcat(str, pcmk__readable_interval(interval_ms), "-interval ",
NULL);
}
pcmk__g_strcat(str, pcmk__readable_action(task, interval_ms), " on ",
node_name, NULL);
if (status == PCMK_EXEC_DONE) {
pcmk__g_strcat(str, " returned '", services_ocf_exitcode_str(rc), "'",
NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, " (", exit_reason, ")", NULL);
}
} else {
pcmk__g_strcat(str, " could not be executed (",
pcmk_exec_status_str(status), NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ": ", exit_reason, NULL);
}
g_string_append_c(str, ')');
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *s = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str, " at ", s, NULL);
free(s);
}
if (!pcmk__str_empty(exec_time)) {
int exec_time_ms = 0;
if ((pcmk__scan_min_int(exec_time, &exec_time_ms, 0) == pcmk_rc_ok)
&& (exec_time_ms > 0)) {
pcmk__g_strcat(str, " after ",
pcmk__readable_interval(exec_time_ms), NULL);
}
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
free(rsc_id);
free(task);
}
/*!
* \internal
* \brief Display a failed action with technical details
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_technical(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
const char *call_id = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
const char *queue_time = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
const char *exit_status = services_ocf_exitcode_str(rc);
const char *lrm_status = pcmk_exec_status_str(status);
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)) {
op_key = "unknown operation";
}
if (pcmk__str_empty(exit_status)) {
exit_status = "unknown exit status";
}
if (pcmk__str_empty(call_id)) {
call_id = "unknown";
}
str = g_string_sized_new(256);
g_string_append_printf(str, "%s on %s '%s' (%d): call=%s, status='%s'",
op_key, node_name, exit_status, rc, call_id,
lrm_status);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ", exitreason='", exit_reason, "'", NULL);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *last_change_str = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str,
", " XML_RSC_OP_LAST_CHANGE "="
"'", last_change_str, "'", NULL);
free(last_change_str);
}
if (!pcmk__str_empty(queue_time)) {
pcmk__g_strcat(str, ", queued=", queue_time, "ms", NULL);
}
if (!pcmk__str_empty(exec_time)) {
pcmk__g_strcat(str, ", exec=", exec_time, "ms", NULL);
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t")
static int
failed_action_default(pcmk__output_t *out, va_list args)
{
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts = va_arg(args, uint32_t);
const char *op_key = pe__xe_history_key(xml_op);
const char *node_name = crm_element_value(xml_op, XML_ATTR_UNAME);
const char *exit_reason = crm_element_value(xml_op,
XML_LRM_ATTR_EXIT_REASON);
const char *exec_time = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
int rc;
int status;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_RC), &rc, 0);
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
&status, 0);
if (pcmk__str_empty(node_name)) {
node_name = "unknown node";
}
if (pcmk_is_set(show_opts, pcmk_show_failed_detail)) {
failed_action_technical(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
} else {
failed_action_friendly(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t")
static int
failed_action_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *op_key = pe__xe_history_key(xml_op);
const char *op_key_name = "op_key";
int rc;
int status;
const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON);
time_t epoch = 0;
char *rc_s = NULL;
char *reason_s = crm_xml_escape(exit_reason ? exit_reason : "none");
xmlNodePtr node = NULL;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_RC), &rc, 0);
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
&status, 0);
rc_s = pcmk__itoa(rc);
if (crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY) == NULL) {
op_key_name = "id";
}
node = pcmk__output_create_xml_node(out, "failure",
op_key_name, op_key,
"node", crm_element_value(xml_op, XML_ATTR_UNAME),
"exitstatus", services_ocf_exitcode_str(rc),
"exitreason", pcmk__s(reason_s, ""),
"exitcode", rc_s,
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"status", pcmk_exec_status_str(status),
NULL);
free(rc_s);
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
guint interval_ms = 0;
char *interval_ms_s = NULL;
char *rc_change = pcmk__epoch2str(&epoch,
crm_time_log_date
|crm_time_log_timeofday
|crm_time_log_with_timezone);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
interval_ms_s = crm_strdup_printf("%u", interval_ms);
pcmk__xe_set_props(node, XML_RSC_OP_LAST_CHANGE, rc_change,
"queued", crm_element_value(xml_op, XML_RSC_OP_T_QUEUE),
"exec", crm_element_value(xml_op, XML_RSC_OP_T_EXEC),
"interval", interval_ms_s,
"task", crm_element_value(xml_op, XML_LRM_ATTR_TASK),
NULL);
free(interval_ms_s);
free(rc_change);
}
free(reason_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action-list", "pcmk_scheduler_t *", "GList *",
"GList *", "uint32_t", "bool")
static int
failed_action_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *xml_op = NULL;
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(scheduler->failed) == 0) {
return rc;
}
for (xml_op = pcmk__xml_first_child(scheduler->failed); xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
char *rsc = NULL;
if (!pcmk__str_in_list(crm_element_value(xml_op, XML_ATTR_UNAME), only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
if (pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
if (!parse_op_key(pe__xe_history_key(xml_op), &rsc, NULL, NULL)) {
continue;
}
if (!pcmk__str_in_list(rsc, only_rsc, pcmk__str_star_matches)) {
free(rsc);
continue;
}
free(rsc);
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Failed Resource Actions");
out->message(out, "failed-action", xml_op, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
static void
status_node(pcmk_node_t *node, xmlNodePtr parent, uint32_t show_opts)
{
int health = pe__node_health(node);
// Cluster membership
if (node->details->online) {
pcmk_create_html_node(parent, "span", NULL, "online", " online");
} else {
pcmk_create_html_node(parent, "span", NULL, "offline", " OFFLINE");
}
// Standby mode
if (node->details->standby_onfail && (node->details->running_rsc != NULL)) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby due to on-fail,"
" with active resources)");
} else if (node->details->standby_onfail) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby due to on-fail)");
} else if (node->details->standby && (node->details->running_rsc != NULL)) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby, with active resources)");
} else if (node->details->standby) {
pcmk_create_html_node(parent, "span", NULL, "standby", " (in standby)");
}
// Maintenance mode
if (node->details->maintenance) {
pcmk_create_html_node(parent, "span", NULL, "maint",
" (in maintenance mode)");
}
// Node health
if (health < 0) {
pcmk_create_html_node(parent, "span", NULL, "health_red",
" (health is RED)");
} else if (health == 0) {
pcmk_create_html_node(parent, "span", NULL, "health_yellow",
" (health is YELLOW)");
}
// Feature set
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
char *buf = crm_strdup_printf(", feature set %s", feature_set);
pcmk_create_html_node(parent, "span", NULL, NULL, buf);
free(buf);
}
}
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool",
"GList *", "GList *")
static int
node_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (full) {
xmlNodePtr item_node;
if (pcmk_all_flags_set(show_opts, pcmk_show_brief | pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:");
status_node(node, item_node, show_opts);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
}
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc2 = NULL;
int rc = pcmk_rc_no_output;
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:");
status_node(node, item_node, show_opts);
for (lpc2 = node->details->running_rsc; lpc2 != NULL; lpc2 = lpc2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc2->data;
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources");
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else {
char *buf = crm_strdup_printf("%s:", node_name);
item_node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, "bold", buf);
status_node(node, item_node, show_opts);
free(buf);
}
} else {
out->begin_list(out, NULL, NULL, "%s:", node_name);
}
free(node_name);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Get a human-friendly textual description of a node's status
*
* \param[in] node Node to check
*
* \return String representation of node's status
*/
static const char *
node_text_status(const pcmk_node_t *node)
{
if (node->details->unclean) {
if (node->details->online) {
return "UNCLEAN (online)";
} else if (node->details->pending) {
return "UNCLEAN (pending)";
} else {
return "UNCLEAN (offline)";
}
} else if (node->details->pending) {
return "pending";
} else if (node->details->standby_onfail && node->details->online) {
return "standby (on-fail)";
} else if (node->details->standby) {
if (node->details->online) {
if (node->details->running_rsc) {
return "standby (with active resources)";
} else {
return "standby";
}
} else {
return "OFFLINE (standby)";
}
} else if (node->details->maintenance) {
if (node->details->online) {
return "maintenance";
} else {
return "OFFLINE (maintenance)";
}
} else if (node->details->online) {
return "online";
}
return "OFFLINE";
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
GString *str = g_string_sized_new(64);
int health = pe__node_health(node);
// Create a summary line with node type, name, and status
if (pe__is_guest_node(node)) {
g_string_append(str, "GuestNode");
} else if (pe__is_remote_node(node)) {
g_string_append(str, "RemoteNode");
} else {
g_string_append(str, "Node");
}
pcmk__g_strcat(str, " ", node_name, ": ", node_text_status(node), NULL);
if (health < 0) {
g_string_append(str, " (health is RED)");
} else if (health == 0) {
g_string_append(str, " (health is YELLOW)");
}
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
pcmk__g_strcat(str, ", feature set ", feature_set, NULL);
}
}
/* If we're grouping by node, print its resources */
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (pcmk_is_set(show_opts, pcmk_show_brief)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
out->end_list(out);
g_list_free(rscs);
}
} else {
GList *gIter2 = NULL;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
for (gIter2 = node->details->running_rsc; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter2->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
out->end_list(out);
out->end_list(out);
}
} else {
out->list_item(out, NULL, "%s", str->str);
}
g_string_free(str, TRUE);
free(node_name);
} else {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
out->begin_list(out, NULL, NULL, "Node: %s", node_name);
free(node_name);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
const char *node_type = "unknown";
char *length_s = pcmk__itoa(g_list_length(node->details->running_rsc));
int health = pe__node_health(node);
const char *health_s = NULL;
const char *feature_set;
switch (node->details->type) {
case pcmk_node_variant_cluster:
node_type = "member";
break;
case pcmk_node_variant_remote:
node_type = "remote";
break;
case node_ping:
node_type = "ping";
break;
}
if (health < 0) {
health_s = "red";
} else if (health == 0) {
health_s = "yellow";
} else {
health_s = "green";
}
feature_set = get_node_feature_set(node);
pe__name_and_nvpairs_xml(out, true, "node", 15,
"name", node->details->uname,
"id", node->details->id,
"online", pcmk__btoa(node->details->online),
"standby", pcmk__btoa(node->details->standby),
"standby_onfail", pcmk__btoa(node->details->standby_onfail),
"maintenance", pcmk__btoa(node->details->maintenance),
"pending", pcmk__btoa(node->details->pending),
"unclean", pcmk__btoa(node->details->unclean),
"health", health_s,
"feature_set", feature_set,
"shutdown", pcmk__btoa(node->details->shutdown),
"expected_up", pcmk__btoa(node->details->expected_up),
"is_dc", pcmk__btoa(node->details->is_dc),
"resources_running", length_s,
"type", node_type);
if (pe__is_guest_node(node)) {
xmlNodePtr xml_node = pcmk__output_xml_peek_parent(out);
crm_xml_add(xml_node, "id_as_resource", node->details->remote_rsc->container->id);
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc = NULL;
for (lpc = node->details->running_rsc; lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
}
free(length_s);
out->end_list(out);
} else {
pcmk__output_xml_create_parent(out, "node",
"name", node->details->uname,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_text(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v;
if (value == NULL) {
v = 0;
} else {
pcmk__scan_min_int(value, &v, INT_MIN);
}
if (v <= 0) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is lost", name, value);
} else if (v < expected_score) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is degraded (Expected=%d)", name, value, expected_score);
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_html(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v;
char *s = crm_strdup_printf("%s: %s", name, value);
xmlNodePtr item_node = pcmk__output_create_xml_node(out, "li", NULL);
if (value == NULL) {
v = 0;
} else {
pcmk__scan_min_int(value, &v, INT_MIN);
}
pcmk_create_html_node(item_node, "span", NULL, NULL, s);
free(s);
if (v <= 0) {
pcmk_create_html_node(item_node, "span", NULL, "bold", "(connectivity is lost)");
} else if (v < expected_score) {
char *buf = crm_strdup_printf("(connectivity is degraded -- expected %d", expected_score);
pcmk_create_html_node(item_node, "span", NULL, "bold", buf);
free(buf);
}
} else {
out->list_item(out, NULL, "%s: %s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr")
static int
node_and_op(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
gchar *node_str = NULL;
char *last_change_str = NULL;
const char *op_rsc = crm_element_value(xml_op, "resource");
int status;
time_t last_change = 0;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
&status, PCMK_EXEC_UNKNOWN);
rsc = pe_find_resource(scheduler->resources, op_rsc);
if (rsc) {
const pcmk_node_t *node = pe__current_node(rsc);
const char *target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
uint32_t show_opts = pcmk_show_rsc_only | pcmk_show_pending;
if (node == NULL) {
node = rsc->pending_node;
}
node_str = pcmk__native_output_string(rsc, rsc_printable_id(rsc), node,
show_opts, target_role, false);
} else {
node_str = crm_strdup_printf("Unknown resource %s", op_rsc);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change) == pcmk_ok) {
last_change_str = crm_strdup_printf(", %s='%s', exec=%sms",
XML_RSC_OP_LAST_CHANGE,
pcmk__trim(ctime(&last_change)),
crm_element_value(xml_op, XML_RSC_OP_T_EXEC));
}
out->list_item(out, NULL, "%s: %s (node=%s, call=%s, rc=%s%s): %s",
node_str, pe__xe_history_key(xml_op),
crm_element_value(xml_op, XML_ATTR_UNAME),
crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
crm_element_value(xml_op, XML_LRM_ATTR_RC),
last_change_str ? last_change_str : "",
pcmk_exec_status_str(status));
g_free(node_str);
free(last_change_str);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr")
static int
node_and_op_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
const char *op_rsc = crm_element_value(xml_op, "resource");
int status;
time_t last_change = 0;
xmlNode *node = NULL;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
&status, PCMK_EXEC_UNKNOWN);
node = pcmk__output_create_xml_node(out, "operation",
"op", pe__xe_history_key(xml_op),
"node", crm_element_value(xml_op, XML_ATTR_UNAME),
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"rc", crm_element_value(xml_op, XML_LRM_ATTR_RC),
"status", pcmk_exec_status_str(status),
NULL);
rsc = pe_find_resource(scheduler->resources, op_rsc);
if (rsc) {
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE);
char *agent_tuple = NULL;
agent_tuple = crm_strdup_printf("%s:%s:%s", class,
pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) ? crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER) : "",
kind);
pcmk__xe_set_props(node, "rsc", rsc_printable_id(rsc),
"agent", agent_tuple,
NULL);
free(agent_tuple);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change) == pcmk_ok) {
pcmk__xe_set_props(node, XML_RSC_OP_LAST_CHANGE,
pcmk__trim(ctime(&last_change)),
XML_RSC_OP_T_EXEC, crm_element_value(xml_op, XML_RSC_OP_T_EXEC),
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_xml(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "attribute",
"name", name,
"value", value,
NULL);
if (add_extra) {
char *buf = pcmk__itoa(expected_score);
crm_xml_add(node, "expected", buf);
free(buf);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute-list", "pcmk_scheduler_t *", "uint32_t",
"bool", "GList *", "GList *")
static int
node_attribute_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
int rc = pcmk_rc_no_output;
/* Display each node's attributes */
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
GList *attr_list = NULL;
GHashTableIter iter;
gpointer key;
if (!node || !node->details || !node->details->online) {
continue;
}
g_hash_table_iter_init(&iter, node->details->attrs);
while (g_hash_table_iter_next (&iter, &key, NULL)) {
attr_list = filter_attr_list(attr_list, key);
}
if (attr_list == NULL) {
continue;
}
if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) {
g_list_free(attr_list);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node Attributes");
out->message(out, "node", node, show_opts, false, only_node, only_rsc);
for (GList *aIter = attr_list; aIter != NULL; aIter = aIter->next) {
const char *name = aIter->data;
const char *value = NULL;
int expected_score = 0;
bool add_extra = false;
value = pe_node_attribute_raw(node, name);
add_extra = add_extra_info(node, node->details->running_rsc,
scheduler, name, &expected_score);
/* Print attribute name and value */
out->message(out, "node-attribute", name, value, add_extra,
expected_score);
}
g_list_free(attr_list);
out->end_list(out);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *comment = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s capacity:",
comment, pe__node_name(node));
g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity_xml(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *comment = va_arg(args, const char *);
xmlNodePtr xml_node = pcmk__output_create_xml_node(out, "capacity",
"node", node->details->uname,
"comment", comment,
NULL);
g_hash_table_foreach(node->details->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-history-list", "pcmk_scheduler_t *", "pcmk_node_t *",
"xmlNodePtr", "GList *", "GList *", "uint32_t", "uint32_t")
static int
node_history_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
xmlNode *node_state = va_arg(args, xmlNode *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
xmlNode *lrm_rsc = NULL;
xmlNode *rsc_entry = NULL;
int rc = pcmk_rc_no_output;
lrm_rsc = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
lrm_rsc = find_xml_node(lrm_rsc, XML_LRM_TAG_RESOURCES, FALSE);
/* Print history of each of the node's resources */
for (rsc_entry = first_named_child(lrm_rsc, XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID);
pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, rsc_id);
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
/* We can't use is_filtered here to filter group resources. For is_filtered,
* we have to decide whether to check the parent or not. If we check the
* parent, all elements of a group will always be printed because that's how
* is_filtered works for groups. If we do not check the parent, sometimes
* this will filter everything out.
*
* For other resource types, is_filtered is okay.
*/
if (parent->variant == pcmk_rsc_variant_group) {
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(parent), only_rsc,
pcmk__str_star_matches)) {
continue;
}
} else {
if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) {
continue;
}
}
if (!pcmk_is_set(section_opts, pcmk_section_operations)) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
if (failcount <= 0) {
continue;
}
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-history", rsc, rsc_id, false,
failcount, last_failure, false);
} else {
GList *op_list = get_operation_list(rsc_entry);
pcmk_resource_t *rsc = pe_find_resource(scheduler->resources,
crm_element_value(rsc_entry, XML_ATTR_ID));
if (op_list == NULL) {
continue;
}
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-operation-list", scheduler, rsc, node,
op_list, show_opts);
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_html(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Node List");
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_text(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
/* space-separated lists of node names */
GString *online_nodes = NULL;
GString *online_remote_nodes = NULL;
GString *online_guest_nodes = NULL;
GString *offline_nodes = NULL;
GString *offline_remote_nodes = NULL;
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
free(node_name);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node List");
// Determine whether to display node individually or in a list
if (node->details->unclean || node->details->pending
|| (node->details->standby_onfail && node->details->online)
|| node->details->standby || node->details->maintenance
|| pcmk_is_set(show_opts, pcmk_show_rscs_by_node)
|| pcmk_is_set(show_opts, pcmk_show_feature_set)
|| (pe__node_health(node) <= 0)) {
// Display node individually
} else if (node->details->online) {
// Display online node in a list
if (pe__is_guest_node(node)) {
pcmk__add_word(&online_guest_nodes, 1024, node_name);
} else if (pe__is_remote_node(node)) {
pcmk__add_word(&online_remote_nodes, 1024, node_name);
} else {
pcmk__add_word(&online_nodes, 1024, node_name);
}
free(node_name);
continue;
} else {
// Display offline node in a list
if (pe__is_remote_node(node)) {
pcmk__add_word(&offline_remote_nodes, 1024, node_name);
} else if (pe__is_guest_node(node)) {
/* ignore offline guest nodes */
} else {
pcmk__add_word(&offline_nodes, 1024, node_name);
}
free(node_name);
continue;
}
/* If we get here, node is in bad state, or we're grouping by node */
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
free(node_name);
}
/* If we're not grouping by node, summarize nodes by status */
if (online_nodes != NULL) {
out->list_item(out, "Online", "[ %s ]",
(const char *) online_nodes->str);
g_string_free(online_nodes, TRUE);
}
if (offline_nodes != NULL) {
out->list_item(out, "OFFLINE", "[ %s ]",
(const char *) offline_nodes->str);
g_string_free(offline_nodes, TRUE);
}
if (online_remote_nodes) {
out->list_item(out, "RemoteOnline", "[ %s ]",
(const char *) online_remote_nodes->str);
g_string_free(online_remote_nodes, TRUE);
}
if (offline_remote_nodes) {
out->list_item(out, "RemoteOFFLINE", "[ %s ]",
(const char *) offline_remote_nodes->str);
g_string_free(offline_remote_nodes, TRUE);
}
if (online_guest_nodes != NULL) {
out->list_item(out, "GuestOnline", "[ %s ]",
(const char *) online_guest_nodes->str);
g_string_free(online_guest_nodes, TRUE);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_xml(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
out->begin_list(out, NULL, NULL, "nodes");
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
out->end_list(out);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-summary", "pcmk_scheduler_t *", "GList *", "GList *",
"uint32_t", "uint32_t", "bool")
static int
node_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *node_state = NULL;
xmlNode *cib_status = pcmk_find_cib_element(scheduler->input,
XML_CIB_TAG_STATUS);
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(cib_status) == 0) {
return rc;
}
for (node_state = first_named_child(cib_status, XML_CIB_TAG_STATE);
node_state != NULL; node_state = crm_next_same_xml(node_state)) {
pcmk_node_t *node = pe_find_node_id(scheduler->nodes, ID(node_state));
if (!node || !node->details || !node->details->online) {
continue;
}
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc,
pcmk_is_set(section_opts, pcmk_section_operations) ? "Operations" : "Migration Summary");
out->message(out, "node-history-list", scheduler, node, node_state,
only_node, only_rsc, section_opts, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
if (rsc) {
out->list_item(out, NULL, "%s: %s allocation score on %s: %s",
prefix, rsc->id, uname, score);
} else {
out->list_item(out, NULL, "%s: %s = %s", prefix, uname, score);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight_xml(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, "node_weight",
"function", prefix,
"node", uname,
"score", score,
NULL);
if (rsc) {
crm_xml_add(node, "id", rsc->id);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t")
static int
op_history_text(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
char *buf = op_history_string(xml_op, task, interval_ms_s, rc,
pcmk_is_set(show_opts, pcmk_show_timing));
out->list_item(out, NULL, "%s", buf);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t")
static int
op_history_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
char *rc_s = pcmk__itoa(rc);
xmlNodePtr node = pcmk__output_create_xml_node(out, "operation_history",
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"task", task,
"rc", rc_s,
"rc_text", services_ocf_exitcode_str(rc),
NULL);
free(rc_s);
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *s = crm_strdup_printf("%sms", interval_ms_s);
crm_xml_add(node, "interval", s);
free(s);
}
if (pcmk_is_set(show_opts, pcmk_show_timing)) {
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
char *s = pcmk__epoch2str(&epoch, 0);
crm_xml_add(node, XML_RSC_OP_LAST_CHANGE, s);
free(s);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, XML_RSC_OP_T_EXEC, s);
free(s);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, XML_RSC_OP_T_QUEUE, s);
free(s);
}
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
out->list_item(out, NULL, "%s promotion score on %s: %s",
child_rsc->id,
chosen? chosen->details->uname : "none",
score);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, "promotion_score",
"id", child_rsc->id,
"score", score,
NULL);
if (chosen) {
crm_xml_add(node, "node", chosen->details->uname);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config(pcmk__output_t *out, va_list args) {
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
bool raw = va_arg(args, int);
char *rsc_xml = formatted_xml_buf(rsc, raw);
out->output_xml(out, "xml", rsc_xml);
free(rsc_xml);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config_text(pcmk__output_t *out, va_list args) {
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
bool raw = va_arg(args, int);
char *rsc_xml = formatted_xml_buf(rsc, raw);
pcmk__formatted_printf(out, "Resource XML:\n");
out->output_xml(out, "xml", rsc_xml);
free(rsc_xml);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_text(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
char *buf = resource_history_string(rsc, rsc_id, all, failcount, last_failure);
if (as_header) {
out->begin_list(out, NULL, NULL, "%s", buf);
} else {
out->list_item(out, NULL, "%s", buf);
}
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_xml(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
xmlNodePtr node = pcmk__output_xml_create_parent(out, "resource_history",
"id", rsc_id,
NULL);
if (rsc == NULL) {
pcmk__xe_set_bool_attr(node, "orphan", true);
} else if (all || failcount || last_failure > 0) {
char *migration_s = pcmk__itoa(rsc->migration_threshold);
pcmk__xe_set_props(node, "orphan", "false",
"migration-threshold", migration_s,
NULL);
free(migration_s);
if (failcount > 0) {
char *s = pcmk__itoa(failcount);
crm_xml_add(node, PCMK__FAIL_COUNT_PREFIX, s);
free(s);
}
if (last_failure > 0) {
char *s = pcmk__epoch2str(&last_failure, 0);
crm_xml_add(node, PCMK__LAST_FAILURE_PREFIX, s);
free(s);
}
}
if (!as_header) {
pcmk__output_xml_pop_parent(out);
}
return pcmk_rc_ok;
}
static void
print_resource_header(pcmk__output_t *out, uint32_t show_opts)
{
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
/* Active resources have already been printed by node */
out->begin_list(out, NULL, NULL, "Inactive Resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->begin_list(out, NULL, NULL, "Full List of Resources");
} else {
out->begin_list(out, NULL, NULL, "Active Resources");
}
}
PCMK__OUTPUT_ARGS("resource-list", "pcmk_scheduler_t *", "uint32_t", "bool",
"GList *", "GList *", "bool")
static int
resource_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_summary = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
bool print_spacer = va_arg(args, int);
GList *rsc_iter;
int rc = pcmk_rc_no_output;
bool printed_header = false;
/* If we already showed active resources by node, and
* we're not showing inactive resources, we have nothing to do
*/
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node) &&
!pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
return rc;
}
/* If we haven't already printed resources grouped by node,
* and brief output was requested, print resource summary */
if (pcmk_is_set(show_opts, pcmk_show_brief)
&& !pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(scheduler->resources, only_rsc);
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
rc = pe__rscs_brief_output(out, rscs, show_opts);
g_list_free(rscs);
}
/* For each resource, display it if appropriate */
for (rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) rsc_iter->data;
int x;
/* Complex resources may have some sub-resources active and some inactive */
gboolean is_active = rsc->fns->active(rsc, TRUE);
gboolean partially_active = rsc->fns->active(rsc, FALSE);
/* Skip inactive orphans (deleted but still in CIB) */
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed) && !is_active) {
continue;
/* Skip active resources if we already displayed them by node */
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (is_active) {
continue;
}
/* Skip primitives already counted in a brief summary */
} else if (pcmk_is_set(show_opts, pcmk_show_brief)
&& (rsc->variant == pcmk_rsc_variant_primitive)) {
continue;
/* Skip resources that aren't at least partially active,
* unless we're displaying inactive resources
*/
} else if (!partially_active && !pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
continue;
} else if (partially_active && !pe__rsc_running_on_any(rsc, only_node)) {
continue;
}
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
/* Print this resource */
x = out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc,
only_node, only_rsc);
if (x == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
if (print_summary && rc != pcmk_rc_ok) {
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
out->list_item(out, NULL, "No inactive resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->list_item(out, NULL, "No resources");
} else {
out->list_item(out, NULL, "No active resources");
}
}
if (printed_header) {
out->end_list(out);
}
return rc;
}
PCMK__OUTPUT_ARGS("resource-operation-list", "pcmk_scheduler_t *",
"pcmk_resource_t *", "pcmk_node_t *", "GList *", "uint32_t")
static int
resource_operation_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler G_GNUC_UNUSED = va_arg(args,
pcmk_scheduler_t *);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
GList *op_list = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
GList *gIter = NULL;
int rc = pcmk_rc_no_output;
/* Print each operation */
for (gIter = op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *xml_op = (xmlNode *) gIter->data;
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 *op_rc = crm_element_value(xml_op, XML_LRM_ATTR_RC);
int op_rc_i;
pcmk__scan_min_int(op_rc, &op_rc_i, 0);
/* Display 0-interval monitors as "probe" */
if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) {
task = "probe";
}
/* If this is the first printed operation, print heading for resource */
if (rc == pcmk_rc_no_output) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
out->message(out, "resource-history", rsc, rsc_printable_id(rsc), true,
failcount, last_failure, true);
rc = pcmk_rc_ok;
}
/* Print the operation */
out->message(out, "op-history", xml_op, task, interval_ms_s,
op_rc_i, show_opts);
}
/* Free the list we created (no need to free the individual items) */
g_list_free(op_list);
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *fn = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s utilization on %s:",
fn, rsc->id, pe__node_name(node));
g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *fn = va_arg(args, const char *);
xmlNodePtr xml_node = pcmk__output_create_xml_node(out, "utilization",
"resource", rsc->id,
"node", node->details->uname,
"function", fn,
NULL);
g_hash_table_foreach(rsc->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_html(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
if (ticket->last_granted > -1) {
char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0);
out->list_item(out, NULL, "%s:\t%s%s %s=\"%s\"", ticket->id,
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "",
"last-granted", pcmk__s(epoch_str, ""));
free(epoch_str);
} else {
out->list_item(out, NULL, "%s:\t%s%s", ticket->id,
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_text(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
if (ticket->last_granted > -1) {
char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0);
out->list_item(out, ticket->id, "%s%s %s=\"%s\"",
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "",
"last-granted", pcmk__s(epoch_str, ""));
free(epoch_str);
} else {
out->list_item(out, ticket->id, "%s%s",
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_xml(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
xmlNodePtr node = NULL;
node = pcmk__output_create_xml_node(out, "ticket",
"id", ticket->id,
"status", ticket->granted ? "granted" : "revoked",
"standby", pcmk__btoa(ticket->standby),
NULL);
if (ticket->last_granted > -1) {
char *buf = pcmk__epoch2str(&ticket->last_granted, 0);
crm_xml_add(node, "last-granted", buf);
free(buf);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket-list", "pcmk_scheduler_t *", "bool")
static int
ticket_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
bool print_spacer = va_arg(args, int);
GHashTableIter iter;
gpointer key, value;
if (g_hash_table_size(scheduler->tickets) == 0) {
return pcmk_rc_no_output;
}
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
/* Print section heading */
out->begin_list(out, NULL, NULL, "Tickets");
/* Print each ticket */
g_hash_table_iter_init(&iter, scheduler->tickets);
while (g_hash_table_iter_next(&iter, &key, &value)) {
pcmk_ticket_t *ticket = (pcmk_ticket_t *) value;
out->message(out, "ticket", ticket);
}
/* Close section */
out->end_list(out);
return pcmk_rc_ok;
}
static pcmk__message_entry_t fmt_functions[] = {
{ "ban", "default", ban_text },
{ "ban", "html", ban_html },
{ "ban", "xml", ban_xml },
{ "ban-list", "default", ban_list },
{ "bundle", "default", pe__bundle_text },
{ "bundle", "xml", pe__bundle_xml },
{ "bundle", "html", pe__bundle_html },
{ "clone", "default", pe__clone_default },
{ "clone", "xml", pe__clone_xml },
{ "cluster-counts", "default", cluster_counts_text },
{ "cluster-counts", "html", cluster_counts_html },
{ "cluster-counts", "xml", cluster_counts_xml },
{ "cluster-dc", "default", cluster_dc_text },
{ "cluster-dc", "html", cluster_dc_html },
{ "cluster-dc", "xml", cluster_dc_xml },
{ "cluster-options", "default", cluster_options_text },
{ "cluster-options", "html", cluster_options_html },
{ "cluster-options", "log", cluster_options_log },
{ "cluster-options", "xml", cluster_options_xml },
{ "cluster-summary", "default", cluster_summary },
{ "cluster-summary", "html", cluster_summary_html },
{ "cluster-stack", "default", cluster_stack_text },
{ "cluster-stack", "html", cluster_stack_html },
{ "cluster-stack", "xml", cluster_stack_xml },
{ "cluster-times", "default", cluster_times_text },
{ "cluster-times", "html", cluster_times_html },
{ "cluster-times", "xml", cluster_times_xml },
{ "failed-action", "default", failed_action_default },
{ "failed-action", "xml", failed_action_xml },
{ "failed-action-list", "default", failed_action_list },
{ "group", "default", pe__group_default},
{ "group", "xml", pe__group_xml },
{ "maint-mode", "text", cluster_maint_mode_text },
{ "node", "default", node_text },
{ "node", "html", node_html },
{ "node", "xml", node_xml },
{ "node-and-op", "default", node_and_op },
{ "node-and-op", "xml", node_and_op_xml },
{ "node-capacity", "default", node_capacity },
{ "node-capacity", "xml", node_capacity_xml },
{ "node-history-list", "default", node_history_list },
{ "node-list", "default", node_list_text },
{ "node-list", "html", node_list_html },
{ "node-list", "xml", node_list_xml },
{ "node-weight", "default", node_weight },
{ "node-weight", "xml", node_weight_xml },
{ "node-attribute", "default", node_attribute_text },
{ "node-attribute", "html", node_attribute_html },
{ "node-attribute", "xml", node_attribute_xml },
{ "node-attribute-list", "default", node_attribute_list },
{ "node-summary", "default", node_summary },
{ "op-history", "default", op_history_text },
{ "op-history", "xml", op_history_xml },
{ "primitive", "default", pe__resource_text },
{ "primitive", "xml", pe__resource_xml },
{ "primitive", "html", pe__resource_html },
{ "promotion-score", "default", promotion_score },
{ "promotion-score", "xml", promotion_score_xml },
{ "resource-config", "default", resource_config },
{ "resource-config", "text", resource_config_text },
{ "resource-history", "default", resource_history_text },
{ "resource-history", "xml", resource_history_xml },
{ "resource-list", "default", resource_list },
{ "resource-operation-list", "default", resource_operation_list },
{ "resource-util", "default", resource_util },
{ "resource-util", "xml", resource_util_xml },
{ "ticket", "default", ticket_text },
{ "ticket", "html", ticket_html },
{ "ticket", "xml", ticket_xml },
{ "ticket-list", "default", ticket_list },
{ NULL, NULL, NULL }
};
void
pe__register_messages(pcmk__output_t *out) {
pcmk__register_messages(out, fmt_functions);
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index 93175a3dc2..a21c4c160c 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5109 +1,5110 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
// A (parsed) resource action history entry
struct action_history {
pcmk_resource_t *rsc; // Resource that history is for
pcmk_node_t *node; // Node that history is for
xmlNode *xml; // History entry XML
// Parsed from entry XML
const char *id; // XML ID of history entry
const char *key; // Operation key of action
const char *task; // Action name
const char *exit_reason; // Exit reason given for result
guint interval_ms; // Action interval
int call_id; // Call ID of action
int expected_exit_status; // Expected exit status of action
int exit_status; // Actual exit status of action
int execution_status; // Execution status of action
};
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pe__set_working_set_flags()/pe__clear_working_set_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(scheduler, option, flag) do { \
const char *scf_value = pe_pref((scheduler)->config_hash, (option)); \
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} else { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node,
xmlNode *xml_op, xmlNode **last_failure,
enum action_fail_response *failed);
static void determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node);
static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node,
bool overwrite, pcmk_scheduler_t *scheduler);
static void determine_online_status(const xmlNode *node_state,
pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler);
static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler);
static gboolean
is_dangling_guest_node(pcmk_node_t *node)
{
/* we are looking for a remote-node that was supposed to be mapped to a
* container resource, but all traces of that container have disappeared
* from both the config and the status section. */
if (pe__is_guest_or_remote_node(node) &&
node->details->remote_rsc &&
node->details->remote_rsc->container == NULL &&
pcmk_is_set(node->details->remote_rsc->flags,
pcmk_rsc_removed_filler)) {
return TRUE;
}
return FALSE;
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider
* \c PCMK_OPT_PRIORITY_FENCING_DELAY
*/
void
pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node,
const char *reason, bool priority_delay)
{
CRM_CHECK(node, return);
/* A guest node is fenced by marking its container as failed */
if (pe__is_guest_node(node)) {
pcmk_resource_t *rsc = node->details->remote_rsc->container;
if (!pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
crm_notice("Not fencing guest node %s "
"(otherwise would because %s): "
"its guest resource %s is unmanaged",
pe__node_name(node), reason, rsc->id);
} else {
pcmk__sched_warn("Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pe__node_name(node), rsc->id, reason);
/* We don't mark the node as unclean because that would prevent the
* node from running resources. We want to allow it to run resources
* in this transition if the recovery succeeds.
*/
node->details->remote_requires_reset = TRUE;
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
}
}
} else if (is_dangling_guest_node(node)) {
crm_info("Cleaning up dangling connection for guest node %s: "
"fencing was already done because %s, "
"and guest resource no longer exists",
pe__node_name(node), reason);
pe__set_resource_flags(node->details->remote_rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
} else if (pe__is_remote_node(node)) {
pcmk_resource_t *rsc = node->details->remote_rsc;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
crm_notice("Not fencing remote node %s "
"(otherwise would because %s): connection is unmanaged",
pe__node_name(node), reason);
} else if(node->details->remote_requires_reset == FALSE) {
node->details->remote_requires_reset = TRUE;
pcmk__sched_warn("Remote node %s %s: %s",
pe__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pe__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
pcmk__sched_warn("Cluster node %s %s: %s",
pe__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \
"[(@" XML_NVPAIR_ATTR_NAME "='" PCMK_STONITH_PROVIDES "'" \
"or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \
"and @" XML_NVPAIR_ATTR_VALUE "='" PCMK__VALUE_UNFENCING "']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \
"//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \
"|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \
"/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(scheduler->flags, flag)) {
result = xpath_search(scheduler->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pe__set_working_set_flags(scheduler, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
GHashTable *config_hash = pcmk__strkey_table(free, free);
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
scheduler->config_hash = config_hash;
pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash,
CIB_OPTIONS_FIRST, FALSE, scheduler);
verify_pe_options(scheduler->config_hash);
set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES,
pcmk_sched_probe_resources);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pe_pref(scheduler->config_hash, PCMK_OPT_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" is nonzero");
pe__set_working_set_flags(scheduler, pcmk_sched_have_fencing);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING,
scheduler);
value = pe_pref(scheduler->config_hash, PCMK_OPT_STONITH_TIMEOUT);
scheduler->stonith_timeout = (int) crm_parse_interval_spec(value);
crm_debug("STONITH timeout: %d", scheduler->stonith_timeout);
set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED,
pcmk_sched_fencing_enabled);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
crm_debug("STONITH of failed nodes is enabled");
} else {
crm_debug("STONITH of failed nodes is disabled");
}
scheduler->stonith_action = pe_pref(scheduler->config_hash,
PCMK_OPT_STONITH_ACTION);
if (!strcmp(scheduler->stonith_action, "poweroff")) {
pcmk__warn_once(pcmk__wo_poweroff,
"Support for " PCMK_OPT_STONITH_ACTION " of "
"'poweroff' is deprecated and will be removed in a "
"future release (use 'off' instead)");
scheduler->stonith_action = PCMK_ACTION_OFF;
}
crm_trace("STONITH will %s nodes", scheduler->stonith_action);
set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING,
pcmk_sched_concurrent_fencing);
if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) {
crm_debug("Concurrent fencing is enabled");
} else {
crm_debug("Concurrent fencing is disabled");
}
value = pe_pref(scheduler->config_hash,
PCMK_OPT_PRIORITY_FENCING_DELAY);
if (value) {
scheduler->priority_fencing_delay = crm_parse_interval_spec(value)
/ 1000;
crm_trace("Priority fencing delay is %ds",
scheduler->priority_fencing_delay);
}
- set_config_flag(scheduler, "stop-all-resources", pcmk_sched_stop_all);
+ set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES,
+ pcmk_sched_stop_all);
crm_debug("Stop all active resources: %s",
pcmk__btoa(pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)));
set_config_flag(scheduler, "symmetric-cluster",
pcmk_sched_symmetric_cluster);
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pe_pref(scheduler->config_hash, PCMK_OPT_NO_QUORUM_POLICY);
if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) {
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
int do_panic = 0;
crm_element_value_int(scheduler->input, XML_ATTR_QUORUM_PANIC,
&do_panic);
if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
scheduler->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop': cluster has never had quorum");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop' because fencing is disabled");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
crm_debug("On loss of quorum: Freeze resources");
break;
case pcmk_no_quorum_stop:
crm_debug("On loss of quorum: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case pcmk_no_quorum_fence:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case pcmk_no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(scheduler, "stop-orphan-resources",
pcmk_sched_stop_removed_resources);
if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
crm_trace("Orphan resources are stopped");
} else {
crm_trace("Orphan resources are ignored");
}
set_config_flag(scheduler, "stop-orphan-actions",
pcmk_sched_cancel_removed_actions);
if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) {
crm_trace("Orphan resource actions are stopped");
} else {
crm_trace("Orphan resource actions are ignored");
}
value = pe_pref(scheduler->config_hash, PCMK__OPT_REMOVE_AFTER_STOP);
if (value != NULL) {
if (crm_is_true(value)) {
pe__set_working_set_flags(scheduler, pcmk_sched_remove_after_stop);
#ifndef PCMK__COMPAT_2_0
pcmk__warn_once(pcmk__wo_remove_after,
"Support for the " PCMK__OPT_REMOVE_AFTER_STOP
" cluster property is deprecated and will be "
"removed in a future release");
#endif
} else {
pe__clear_working_set_flags(scheduler,
pcmk_sched_remove_after_stop);
}
}
set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE,
pcmk_sched_in_maintenance);
crm_trace("Maintenance mode: %s",
pcmk__btoa(pcmk_is_set(scheduler->flags,
pcmk_sched_in_maintenance)));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk_sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk_sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->placement_strategy = pe_pref(scheduler->config_hash,
PCMK_OPT_PLACEMENT_STRATEGY);
crm_trace("Placement strategy: %s", scheduler->placement_strategy);
set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK,
pcmk_sched_shutdown_lock);
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
value = pe_pref(scheduler->config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT);
scheduler->shutdown_lock = crm_parse_interval_spec(value) / 1000;
crm_trace("Resources will be locked to nodes that were cleanly "
"shut down (locks expire after %s)",
pcmk__readable_interval(scheduler->shutdown_lock));
} else {
crm_trace("Resources will not be locked to nodes that were cleanly "
"shut down");
}
value = pe_pref(scheduler->config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT);
scheduler->node_pending_timeout = crm_parse_interval_spec(value) / 1000;
if (scheduler->node_pending_timeout == 0) {
crm_trace("Do not fence pending nodes");
} else {
crm_trace("Fence pending nodes after %s",
pcmk__readable_interval(scheduler->node_pending_timeout
* 1000));
}
return TRUE;
}
pcmk_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pcmk_scheduler_t *scheduler)
{
pcmk_node_t *new_node = NULL;
if (pe_find_node(scheduler->nodes, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
new_node = calloc(1, sizeof(pcmk_node_t));
if (new_node == NULL) {
pcmk__sched_err("Could not allocate memory for node %s", uname);
return NULL;
}
new_node->weight = char2score(score);
new_node->details = calloc(1, sizeof(struct pe_node_shared_s));
if (new_node->details == NULL) {
free(new_node);
pcmk__sched_err("Could not allocate memory for node %s", uname);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->details->id = id;
new_node->details->uname = uname;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->rsc_discovery_enabled = TRUE;
new_node->details->running_rsc = NULL;
new_node->details->data_set = scheduler;
if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) {
new_node->details->type = pcmk_node_variant_cluster;
} else if (pcmk__str_eq(type, "remote", pcmk__str_casei)) {
new_node->details->type = pcmk_node_variant_remote;
pe__set_working_set_flags(scheduler, pcmk_sched_have_remote_nodes);
} else {
/* @COMPAT 'ping' is the default for backward compatibility, but it
* should be changed to 'member' at a compatibility break
*/
if (!pcmk__str_eq(type, "ping", pcmk__str_casei)) {
pcmk__config_warn("Node %s has unrecognized type '%s', "
"assuming 'ping'", pcmk__s(uname, "without name"),
type);
}
pcmk__warn_once(pcmk__wo_ping_node,
"Support for nodes of type 'ping' (such as %s) is "
"deprecated and will be removed in a future release",
pcmk__s(uname, "unnamed node"));
new_node->details->type = node_ping;
}
new_node->details->attrs = pcmk__strkey_table(free, free);
if (pe__is_guest_or_remote_node(new_node)) {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("remote"));
} else {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("cluster"));
}
new_node->details->utilization = pcmk__strkey_table(free, free);
new_node->details->digest_cache = pcmk__strkey_table(free,
pe__free_digests);
scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = ID(xml_obj);
const char *remote_name = NULL;
const char *remote_server = NULL;
const char *remote_port = NULL;
const char *connect_timeout = "60s";
const char *remote_allow_migrate=NULL;
const char *is_managed = NULL;
for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL;
attr_set = pcmk__xe_next(attr_set)) {
if (!pcmk__str_eq((const char *)attr_set->name, XML_TAG_META_SETS,
pcmk__str_casei)) {
continue;
}
for (attr = pcmk__xe_first_child(attr_set); attr != NULL;
attr = pcmk__xe_next(attr)) {
const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
if (pcmk__str_eq(name, XML_RSC_ATTR_REMOTE_NODE, pcmk__str_casei)) {
remote_name = value;
} else if (pcmk__str_eq(name, "remote-addr", pcmk__str_casei)) {
remote_server = value;
} else if (pcmk__str_eq(name, "remote-port", pcmk__str_casei)) {
remote_port = value;
} else if (pcmk__str_eq(name, "remote-connect-timeout", pcmk__str_casei)) {
connect_timeout = value;
} else if (pcmk__str_eq(name, "remote-allow-migrate", pcmk__str_casei)) {
remote_allow_migrate=value;
} else if (pcmk__str_eq(name, XML_RSC_ATTR_MANAGED, pcmk__str_casei)) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node)
{
if ((new_node->details->type == pcmk_node_variant_remote)
&& (new_node->details->remote_rsc == NULL)) {
/* Ignore fencing for remote nodes that don't have a connection resource
* associated with them. This happens when remote node entries get left
* in the nodes section after the connection resource is removed.
*/
return;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
// All nodes are unclean until we've seen their status entry
new_node->details->unclean = TRUE;
} else {
// Blind faith ...
new_node->details->unclean = FALSE;
}
/* We need to be able to determine if a node's status section
* exists or not separate from whether the node is unclean. */
new_node->details->unseen = TRUE;
}
gboolean
unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
pcmk_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
const char *score = NULL;
for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) {
new_node = NULL;
id = crm_element_value(xml_obj, XML_ATTR_ID);
uname = crm_element_value(xml_obj, XML_ATTR_UNAME);
type = crm_element_value(xml_obj, XML_ATTR_TYPE);
score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE
"> entry in configuration without id");
continue;
}
new_node = pe_create_node(id, uname, type, score, scheduler);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(scheduler, new_node);
add_node_attrs(xml_obj, new_node, FALSE, scheduler);
crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME));
}
}
if (scheduler->localhost
&& (pe_find_node(scheduler->nodes, scheduler->localhost) == NULL)) {
crm_info("Creating a fake local node");
pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0,
scheduler);
}
return TRUE;
}
static void
setup_container(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *container_id = NULL;
if (rsc->children) {
g_list_foreach(rsc->children, (GFunc) setup_container, scheduler);
return;
}
container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER);
if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) {
pcmk_resource_t *container = pe_find_resource(scheduler->resources,
container_id);
if (container) {
rsc->container = container;
pe__set_resource_flags(container, pcmk_rsc_has_filler);
container->fillers = g_list_append(container->fillers, rsc);
pcmk__rsc_trace(rsc, "Resource %s's container is %s",
rsc->id, container_id);
} else {
pcmk__config_err("Resource %s: Unknown resource container (%s)",
rsc->id, container_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
/* Create remote nodes and guest nodes from the resource configuration
* before unpacking resources.
*/
for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
const char *new_node_id = NULL;
/* Check for remote nodes, which are defined by ocf:pacemaker:remote
* primitives.
*/
if (xml_contains_remote_node(xml_obj)) {
new_node_id = ID(xml_obj);
/* The "pe_find_node" check is here to make sure we don't iterate over
* an expanded node that has already been added to the node list. */
if (new_node_id
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) {
/* This will add an ocf:pacemaker:remote primitive to the
* configuration for the guest node's connection, to be unpacked
* later.
*/
new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources,
scheduler);
if (new_node_id
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL;
xml_obj2 = pcmk__xe_next(xml_obj2)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources,
scheduler);
if (new_node_id
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, ID(xml_obj2), ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc)
{
pcmk_node_t *remote_node = NULL;
if (new_rsc->is_remote_node == FALSE) {
return;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pe_find_node(scheduler->nodes, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s",
new_rsc->id, pe__node_name(remote_node));
remote_node->details->remote_rsc = new_rsc;
if (new_rsc->container == NULL) {
/* Handle start-up fencing for remote nodes (as opposed to guest nodes)
* the same as is done for cluster nodes.
*/
handle_startup_fencing(scheduler, remote_node);
} else {
/* pe_create_node() marks the new node as "remote" or "cluster"; now
* that we know the node is a guest node, update it correctly.
*/
g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("container"));
}
}
static void
destroy_tag(gpointer data)
{
pcmk_tag_t *tag = data;
if (tag) {
free(tag->id);
g_list_free_full(tag->refs, free);
free(tag);
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] scheduler Scheduler data
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when pe__unpack_resource() calls resource_location()
*/
gboolean
unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag);
for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
pcmk_resource_t *new_rsc = NULL;
const char *id = ID(xml_obj);
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring <%s> resource without ID",
xml_obj->name);
continue;
}
if (pcmk__str_eq((const char *) xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE,
pcmk__str_none)) {
if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
g_hash_table_insert(scheduler->template_rsc_sets, strdup(id),
NULL);
}
continue;
}
crm_trace("Unpacking <%s " XML_ATTR_ID "='%s'>",
xml_obj->name, id);
if (pe__unpack_resource(xml_obj, &new_rsc, NULL,
scheduler) == pcmk_rc_ok) {
scheduler->resources = g_list_append(scheduler->resources, new_rsc);
pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
xml_obj->name, id);
}
}
for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
setup_container(rsc, scheduler);
link_rsc2remotenode(scheduler, rsc);
}
scheduler->resources = g_list_sort(scheduler->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
/* Ignore */
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) {
pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined");
pcmk__config_err("Either configure some or disable STONITH with the "
PCMK_OPT_STONITH_ENABLED " option");
pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity");
}
return TRUE;
}
gboolean
unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_tag = NULL;
scheduler->tags = pcmk__strkey_table(free, destroy_tag);
for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL;
xml_tag = pcmk__xe_next(xml_tag)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = ID(xml_tag);
if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) {
continue;
}
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " XML_ATTR_ID,
(const char *) xml_tag->name);
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL;
xml_obj_ref = pcmk__xe_next(xml_obj_ref)) {
const char *obj_ref = ID(xml_obj_ref);
if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) {
continue;
}
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " XML_ATTR_ID,
xml_obj_ref->name, tag_id);
continue;
}
if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) {
return FALSE;
}
}
}
return TRUE;
}
/* The ticket state section:
* "/cib/status/tickets/ticket_state" */
static gboolean
unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler)
{
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pcmk_ticket_t *ticket = NULL;
ticket_id = ID(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return FALSE;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(scheduler->tickets, ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, scheduler);
if (ticket == NULL) {
return FALSE;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
if (pcmk__str_eq(prop_name, XML_ATTR_ID, pcmk__str_none)) {
continue;
}
g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value));
}
granted = g_hash_table_lookup(ticket->state, "granted");
if (granted && crm_is_true(granted)) {
ticket->granted = TRUE;
crm_info("We have ticket '%s'", ticket->id);
} else {
ticket->granted = FALSE;
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, "last-granted");
if (last_granted) {
long long last_granted_ll;
pcmk__scan_ll(last_granted, &last_granted_ll, 0LL);
ticket->last_granted = (time_t) last_granted_ll;
}
standby = g_hash_table_lookup(ticket->state, "standby");
if (standby && crm_is_true(standby)) {
ticket->standby = TRUE;
if (ticket->granted) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
ticket->standby = FALSE;
}
crm_trace("Done with ticket state for %s", ticket_id);
return TRUE;
}
static gboolean
unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (!pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) {
continue;
}
unpack_ticket_state(xml_obj, scheduler);
}
return TRUE;
}
static void
unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state,
pcmk_scheduler_t *scheduler)
{
const char *resource_discovery_enabled = NULL;
const xmlNode *attrs = NULL;
pcmk_resource_t *rsc = NULL;
if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
return;
}
if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) {
return;
}
crm_trace("Processing Pacemaker Remote node %s", pe__node_name(this_node));
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_MAINTENANCE),
&(this_node->details->remote_maintenance), 0);
rsc = this_node->details->remote_rsc;
if (this_node->details->remote_requires_reset == FALSE) {
this_node->details->unclean = FALSE;
this_node->details->unseen = FALSE;
}
attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pe__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) {
crm_info("%s is in standby mode", pe__node_name(this_node));
this_node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) ||
((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed))) {
crm_info("%s is in maintenance mode", pe__node_name(this_node));
this_node->details->maintenance = TRUE;
}
resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY);
if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) {
if (pe__is_remote_node(this_node)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
pcmk__config_warn("Ignoring " XML_NODE_ATTR_RSC_DISCOVERY
" attribute on Pacemaker Remote node %s"
" because fencing is disabled",
pe__node_name(this_node));
} else {
/* This is either a remote node with fencing enabled, or a guest
* node. We don't care whether fencing is enabled when fencing guest
* nodes, because they are "fenced" by recovering their containing
* resource.
*/
crm_info("%s has resource discovery disabled",
pe__node_name(this_node));
this_node->details->rsc_discovery_enabled = FALSE;
}
}
}
/*!
* \internal
* \brief Unpack a cluster node's transient attributes
*
* \param[in] state CIB node state XML
* \param[in,out] node Cluster node whose attributes are being unpacked
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS,
FALSE);
add_node_attrs(attrs, node, TRUE, scheduler);
if (crm_is_true(pe_node_attribute_raw(node, "standby"))) {
crm_info("%s is in standby mode", pe__node_name(node));
node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) {
crm_info("%s is in maintenance mode", pe__node_name(node));
node->details->maintenance = TRUE;
}
discovery = pe_node_attribute_raw(node, XML_NODE_ATTR_RSC_DISCOVERY);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__config_warn("Ignoring " XML_NODE_ATTR_RSC_DISCOVERY " attribute "
"for %s because disabling resource discovery is "
"not allowed for cluster nodes", pe__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* node_state element itself and node attributes inside it, but not the
* resource history inside it. Multiple passes through the status are needed to
* fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = crm_element_value(state, XML_ATTR_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " XML_CIB_TAG_STATE " entry without "
XML_ATTR_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = crm_element_value(state, XML_ATTR_UNAME);
if (uname == NULL) {
/* If a joining peer makes the cluster acquire the quorum from corosync
* meanwhile it has not joined CPG membership of pacemaker-controld yet,
* it's possible that the created node_state entry doesn't have an uname
* yet. We should recognize the node as `pending` and wait for it to
* join CPG.
*/
crm_trace("Handling " XML_CIB_TAG_STATE " entry with id=\"%s\" without "
XML_ATTR_UNAME, id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
pcmk__config_warn("Ignoring recorded node state for id=\"%s\" (%s) "
"because it is no longer in the configuration",
id, pcmk__s(uname, "uname unknown"));
return;
}
if (pe__is_guest_or_remote_node(this_node)) {
/* We can't determine the online status of Pacemaker Remote nodes until
* after all resource history has been unpacked. In this first pass, we
* do need to mark whether the node has been fenced, as this plays a
* role during unpacking cluster node resource state.
*/
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_FENCED),
&(this_node->details->remote_was_fenced), 0);
return;
}
unpack_transient_attributes(state, this_node, scheduler);
/* Provisionally mark this cluster node as clean. We have at least seen it
* in the current cluster's lifetime.
*/
this_node->details->unclean = FALSE;
this_node->details->unseen = FALSE;
crm_trace("Determining online status of cluster node %s (id %s)",
pe__node_name(this_node), id);
determine_online_status(state, this_node, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)
&& this_node->details->online
&& (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(scheduler, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(const xmlNode *status, bool fence,
pcmk_scheduler_t *scheduler)
{
int rc = pcmk_rc_ok;
// Loop through all node_state entries in CIB status
for (const xmlNode *state = first_named_child(status, XML_CIB_TAG_STATE);
state != NULL; state = crm_next_same_xml(state)) {
const char *id = ID(state);
const char *uname = crm_element_value(state, XML_ATTR_UNAME);
pcmk_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
XML_CIB_TAG_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (this_node->details->unpacked) {
crm_trace("Not unpacking resource history for node %s because "
"already unpacked", id);
continue;
}
if (fence) {
// We're processing all remaining nodes
} else if (pe__is_guest_node(this_node)) {
/* We can unpack a guest node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection and containing resource are both up.
*/
pcmk_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL) || (rsc->role != pcmk_role_started)
|| (rsc->container->role != pcmk_role_started)) {
crm_trace("Not unpacking resource history for guest node %s "
"because container and connection are not known to "
"be up", id);
continue;
}
} else if (pe__is_remote_node(this_node)) {
/* We can unpack a remote node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection is up, with the exception of when shutdown locks are
* in use.
*/
pcmk_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL)
|| (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)
&& (rsc->role != pcmk_role_started))) {
crm_trace("Not unpacking resource history for remote node %s "
"because connection is not known to be up", id);
continue;
}
/* If fencing and shutdown locks are disabled and we're not processing
* unseen nodes, then we don't want to unpack offline nodes until online
* nodes have been unpacked. This allows us to number active clone
* instances first.
*/
} else if (!pcmk_any_flags_set(scheduler->flags,
pcmk_sched_fencing_enabled
|pcmk_sched_shutdown_lock)
&& !this_node->details->online) {
crm_trace("Not unpacking resource history for offline "
"cluster node %s", id);
continue;
}
if (pe__is_guest_or_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
unpack_handle_remote_attrs(this_node, state, scheduler);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
this_node->details->unpacked = TRUE;
unpack_node_lrm(this_node, state, scheduler);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (scheduler->tickets == NULL) {
scheduler->tickets = pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status); state != NULL;
state = pcmk__xe_next(state)) {
if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) {
unpack_tickets_state((xmlNode *) state, scheduler);
} else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
unpack_node_state(state, scheduler);
}
}
while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(scheduler->flags,
pcmk_sched_fencing_enabled),
scheduler);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (scheduler->stop_needed != NULL) {
for (GList *item = scheduler->stop_needed; item; item = item->next) {
pcmk_resource_t *container = item->data;
pcmk_node_t *node = pe__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(scheduler->stop_needed);
scheduler->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *this_node = gIter->data;
if (!pe__is_guest_or_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->details->remote_rsc != NULL)) {
pe__set_next_role(this_node->details->remote_rsc, pcmk_role_stopped,
"remote shutdown");
}
if (!this_node->details->unpacked) {
determine_remote_online_status(scheduler, this_node);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack node's time when it became a member at the cluster layer
*
* \param[in] node_state Node's node_state entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM);
int member = 0;
if (member_time == NULL) {
return -1LL;
} else if (crm_str_to_boolean(member_time, &member) == 1) {
/* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was
* recorded as a boolean for a DC < 2.1.7, or the node is pending
* shutdown and has left the CPG, in which case it was set to 1 to avoid
* fencing for PCMK_OPT_NODE_PENDING_TIMEOUT.
*
* We return the effective time for in_ccm=1 because what's important to
* avoid fencing is that effective time minus this value is less than
* the pending node timeout.
*/
return member? (long long) get_effective_time(scheduler) : 0LL;
} else {
long long when_member = 0LL;
if ((pcmk__scan_ll(member_time, &when_member,
0LL) != pcmk_rc_ok) || (when_member < 0LL)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM
" in " XML_CIB_TAG_STATE " entry", member_time);
return -1LL;
}
return when_member;
}
}
/*!
* \internal
* \brief Unpack node's time when it became online in process group
*
* \param[in] node_state Node's node_state entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = crm_element_value(node_state, PCMK__XA_CRMD);
// @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline"
if (pcmk__str_eq(peer_time, OFFLINESTATUS,
pcmk__str_casei|pcmk__str_null_matches)) {
return 0LL;
} else if (pcmk__str_eq(peer_time, ONLINESTATUS, pcmk__str_casei)) {
return 1LL;
} else {
long long when_online = 0LL;
if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok)
|| (when_online < 0)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_CRMD " in "
XML_CIB_TAG_STATE " entry, assuming offline", peer_time);
return 0LL;
}
return when_online;
}
}
/*!
* \internal
* \brief Unpack node attribute for user-requested fencing
*
* \param[in] node Node to check
* \param[in] node_state Node's node_state entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
long long value = 0LL;
int value_i = 0;
const char *value_s = pe_node_attribute_raw(node, PCMK_NODE_ATTR_TERMINATE);
// Value may be boolean or an epoch time
if (crm_str_to_boolean(value_s, &value_i) == 1) {
return (value_i != 0);
}
if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) {
return (value > 0);
}
crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE
"node attribute for %s", value_s, pe__node_name(node));
return false;
}
static gboolean
determine_online_status_no_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
if (when_member <= 0) {
crm_trace("Node %s is %sdown", pe__node_name(this_node),
((when_member < 0)? "presumed " : ""));
} else if (when_online > 0) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node %s is not ready to run resources: %s",
pe__node_name(this_node), join);
}
} else if (this_node->details->expected_up == FALSE) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pe__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE);
crm_info("Node %s member@%lld online@%lld join=%s expected=%s",
pe__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->node_pending_timeout > 0)
&& (when_member > 0) && (when_online <= 0)) {
// There is a timeout on pending nodes, and node is pending
time_t timeout = when_member + scheduler->node_pending_timeout;
if (get_effective_time(node->details->data_set) >= timeout) {
return true; // Node has timed out
}
// Node is pending, but still has time
pe__update_recheck_time(timeout, scheduler, "pending node timeout");
}
return false;
}
static bool
determine_online_status_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
bool termination_requested = unpack_node_terminate(this_node, node_state);
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
/*
- PCMK__XA_JOIN ::= member|down|pending|banned
- PCMK__XA_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- PCMK__XA_IN_CCM ::= true|false
- PCMK__XA_CRMD ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- PCMK__XA_IN_CCM ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- PCMK__XA_CRMD ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s",
pe__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"),
(termination_requested? " (termination requested)" : ""));
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", pe__node_name(this_node));
/* Slightly different criteria since we can't shut down a dead peer */
return (when_online > 0);
}
if (when_member < 0) {
pe_fence_node(scheduler, this_node,
"peer has not been seen by the cluster", FALSE);
return false;
}
if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) {
pe_fence_node(scheduler, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (termination_requested) {
if ((when_member <= 0) && (when_online <= 0)
&& pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) {
crm_info("%s was fenced as requested", pe__node_name(this_node));
return false;
}
pe_fence_node(scheduler, this_node, "fencing was requested", false);
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN,
pcmk__str_null_matches)) {
if (pending_too_long(scheduler, this_node, when_member, when_online)) {
pe_fence_node(scheduler, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if ((when_member > 0) || (when_online > 0)) {
crm_info("- %s is not ready to run resources",
pe__node_name(this_node));
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pe__node_name(this_node));
}
} else if (when_member <= 0) {
// Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes
pe_fence_node(scheduler, this_node,
"peer is no longer part of the cluster", TRUE);
} else if (when_online <= 0) {
pe_fence_node(scheduler, this_node,
"peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) {
crm_info("%s is active", pe__node_name(this_node));
} else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING,
CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("%s is not ready to run resources", pe__node_name(this_node));
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
pe_fence_node(scheduler, this_node, "peer was in an unknown state",
FALSE);
}
return (when_member > 0);
}
static void
determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node)
{
pcmk_resource_t *rsc = this_node->details->remote_rsc;
pcmk_resource_t *container = NULL;
pcmk_node_t *host = NULL;
/* If there is a node state entry for a (former) Pacemaker Remote node
* but no resource creating that node, the node's connection resource will
* be NULL. Consider it an offline remote node in that case.
*/
if (rsc == NULL) {
this_node->details->online = FALSE;
goto remote_online_done;
}
container = rsc->container;
if (container && pcmk__list_of_1(rsc->running_on)) {
host = rsc->running_on->data;
}
/* If the resource is currently started, mark it online. */
if (rsc->role == pcmk_role_started) {
crm_trace("%s node %s presumed ONLINE because connection resource is started",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if ((rsc->role == pcmk_role_started)
&& (rsc->next_role == pcmk_role_stopped)) {
crm_trace("%s node %s shutting down because connection resource is stopping",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if(container && pcmk_is_set(container->flags, pcmk_rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
} else if ((rsc->role == pcmk_role_stopped)
|| ((container != NULL)
&& (container->role == pcmk_role_stopped))) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = FALSE;
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
}
remote_online_done:
crm_trace("Remote node %s online=%s",
this_node->details->id, this_node->details->online ? "TRUE" : "FALSE");
}
static void
determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
this_node->details->expected_up = FALSE;
if (pe__shutdown_requested(this_node)) {
this_node->details->shutdown = TRUE;
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
this_node->details->expected_up = TRUE;
}
if (this_node->details->type == node_ping) {
this_node->details->unclean = FALSE;
online = FALSE; /* As far as resource management is concerned,
* the node is safely offline.
* Anyone caught abusing this logic will be shot
*/
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
online = determine_online_status_no_fencing(scheduler, node_state,
this_node);
} else {
online = determine_online_status_fencing(scheduler, node_state,
this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->fixed = TRUE; // @COMPAT deprecated and unused
this_node->weight = -INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->fixed = TRUE; // @COMPAT deprecated and unused
this_node->weight = -INFINITY;
}
if (this_node->details->type == node_ping) {
crm_info("%s is not a Pacemaker node", pe__node_name(this_node));
} else if (this_node->details->unclean) {
pcmk__sched_warn("%s is unclean", pe__node_name(this_node));
} else if (this_node->details->online) {
crm_info("%s is %s", pe__node_name(this_node),
this_node->details->shutdown ? "shutting down" :
this_node->details->pending ? "pending" :
this_node->details->standby ? "standby" :
this_node->details->maintenance ? "maintenance" : "online");
} else {
crm_trace("%s is offline", pe__node_name(this_node));
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
CRM_ASSERT(end);
basename = strndup(last_rsc_id, end - last_rsc_id + 1);
CRM_ASSERT(basename);
return basename;
}
/*!
* \internal
* \brief Get the name of the first instance of a cloned resource
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name plus :0
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_zero(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
size_t base_name_len = end - last_rsc_id + 1;
char *zero = NULL;
CRM_ASSERT(end);
zero = calloc(base_name_len + 3, sizeof(char));
CRM_ASSERT(zero);
memcpy(zero, last_rsc_id, base_name_len);
zero[base_name_len] = ':';
zero[base_name_len + 1] = '0';
return zero;
}
static pcmk_resource_t *
create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE);
copy_in_properties(xml_rsc, rsc_entry);
crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Orphan resource");
if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pcmk_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pe_find_node(scheduler->nodes, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, "remote", NULL, scheduler);
}
link_rsc2remotenode(scheduler, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER)) {
/* This orphaned rsc needs to be mapped to a container. */
crm_trace("Detected orphaned container filler %s", rsc_id);
pe__set_resource_flags(rsc, pcmk_rsc_removed_filler);
}
pe__set_resource_flags(rsc, pcmk_rsc_removed);
scheduler->resources = g_list_append(scheduler->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create orphan instance for anonymous clone resource history
*
* \param[in,out] parent Clone resource that orphan will be added to
* \param[in] rsc_id Orphan's resource ID
* \param[in] node Where orphan is active (for logging only)
* \param[in,out] scheduler Scheduler data
*
* \return Newly added orphaned instance of \p parent
*/
static pcmk_resource_t *
create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *top = pe__create_clone_child(parent, scheduler);
// find_rsc() because we might be a cloned group
pcmk_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s",
top->id, parent->id, rsc_id, pe__node_name(node));
return orphan;
}
/*!
* \internal
* \brief Check a node for an instance of an anonymous clone
*
* Return a child instance of the specified anonymous clone, in order of
* preference: (1) the instance running on the specified node, if any;
* (2) an inactive instance (i.e. within the total of clone-max instances);
* (3) a newly created orphan (i.e. clone-max instances are already active).
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node on which to check for instance
* \param[in,out] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (no instance)
*/
static pcmk_resource_t *
find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
pcmk_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
CRM_ASSERT(parent != NULL);
CRM_ASSERT(pe_rsc_is_clone(parent));
CRM_ASSERT(!pcmk_is_set(parent->flags, pcmk_rsc_unique));
// Check for active (or partially active, for cloned groups) instance
pcmk__rsc_trace(parent, "Looking for %s on %s in %s",
rsc_id, pe__node_name(node), parent->id);
for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) {
GList *locations = NULL;
pcmk_resource_t *child = rIter->data;
/* Check whether this instance is already known to be active or pending
* anywhere, at this stage of unpacking. Because this function is called
* for a resource before the resource's individual operation history
* entries are unpacked, locations will generally not contain the
* desired node.
*
* However, there are three exceptions:
* (1) when child is a cloned group and we have already unpacked the
* history of another member of the group on the same node;
* (2) when we've already unpacked the history of another numbered
* instance on the same node (which can happen if globally-unique
* was flipped from true to false); and
* (3) when we re-run calculations on the same scheduler data as part of
* a simulation.
*/
child->fns->location(child, &locations, 2);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (((pcmk_node_t *) locations->data)->details == node->details) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry is orphaned, rsc will be NULL.
*/
rsc = parent->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
if (rsc) {
/* If there are multiple instance history entries for an
* anonymous clone in a single node's history (which can
* happen if globally-unique is switched from true to
* false), we want to consider the instances beyond the
* first as orphans, even if there are inactive instance
* numbers available.
*/
if (rsc->running_on) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple (orphan) instance histories on %s",
parent->id, pe__node_name(node));
skip_inactive = TRUE;
rsc = NULL;
} else {
pcmk__rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pcmk_rsc_blocked)) {
// Remember one inactive instance in case we don't find active
inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
/* ... but don't use it if it was already associated with a
* pending action on another node
*/
if (inactive_instance && inactive_instance->pending_node
&& (inactive_instance->pending_node->details != node->details)) {
inactive_instance = NULL;
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pcmk__rsc_trace(parent, "Resource %s, empty slot",
inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has "requires" set to "quorum" or "nothing", and we don't
* have a clone instance for every node, we don't want to consume a valid
* instance number for unclean nodes. Such instances may appear to be active
* according to the history, but should be considered inactive, so we can
* start an instance elsewhere. Treat such instances as orphans.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pe__is_guest_node(node)
&& !pe__is_universal_clone(parent, scheduler)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler);
pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pcmk_resource_t *
unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
const char *rsc_id)
{
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(scheduler->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when clone-max=0, we create
* a single :0 orphan to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources,
clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pcmk_rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (orphan)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->variant > pcmk_rsc_variant_primitive) {
crm_trace("Resource history for %s is orphaned because it is no longer primitive",
rsc_id);
return NULL;
} else {
parent = uber_parent(rsc);
}
if (pe_rsc_is_anon_clone(parent)) {
if (pe_rsc_is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(scheduler, node, parent, base);
free(base);
CRM_ASSERT(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei)
&& !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) {
pcmk__str_update(&rsc->clone_name, rsc_id);
pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pe__node_name(node), rsc->id,
pcmk_is_set(rsc->flags, pcmk_rsc_removed)? " (ORPHAN)" : "");
}
return rsc;
}
static pcmk_resource_t *
process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID);
crm_debug("Detected orphan resource %s on %s", rsc_id, pe__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
pe__clear_resource_flags(rsc, pcmk_rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__",
scheduler);
}
return rsc;
}
static void
process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node,
enum action_fail_response on_fail)
{
pcmk_node_t *tmpnode = NULL;
char *reason = NULL;
enum action_fail_response save_on_fail = pcmk_on_fail_ignore;
CRM_ASSERT(rsc);
pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, role2text(rsc->role), pe__node_name(node),
fail2text(on_fail));
/* process current state */
if (rsc->role != pcmk_role_unknown) {
pcmk_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) {
pcmk_node_t *n = pe__copy_node(node);
pcmk__rsc_trace(rsc, "%s%s%s known on %s",
rsc->id,
((rsc->clone_name == NULL)? "" : " also known as "),
((rsc->clone_name == NULL)? "" : rsc->clone_name),
pe__node_name(n));
g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n);
}
if (pcmk_is_set(iter->flags, pcmk_rsc_unique)) {
break;
}
iter = iter->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if ((rsc->role > pcmk_role_stopped)
&& node->details->online == FALSE
&& node->details->maintenance == FALSE
&& pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
gboolean should_fence = FALSE;
/* If this is a guest node, fence it (regardless of whether fencing is
* enabled, because guest node fencing is done by recovery of the
* container resource rather than by the fencer). Mark the resource
* we're processing as failed. When the guest comes back up, its
* operation history in the CIB will be cleared, freeing the affected
* resource to run again once we are sure we know its state.
*/
if (pe__is_guest_node(node)) {
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
should_fence = TRUE;
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& !pcmk_is_set(node->details->remote_rsc->flags,
pcmk_rsc_failed)) {
/* Setting unseen means that fencing of the remote node will
* occur only if the connection resource is not going to start
* somewhere. This allows connection resources on a failed
* cluster node to move to another node without requiring the
* remote nodes to be fenced as well.
*/
node->details->unseen = TRUE;
reason = crm_strdup_printf("%s is active there (fencing will be"
" revoked if remote connection can "
"be re-established elsewhere)",
rsc->id);
}
should_fence = TRUE;
}
if (should_fence) {
if (reason == NULL) {
reason = crm_strdup_printf("%s is thought to be active there", rsc->id);
}
pe_fence_node(rsc->cluster, node, reason, FALSE);
}
free(reason);
}
/* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */
save_on_fail = on_fail;
if (node->details->unclean) {
/* No extra processing needed
* Also allows resources to be started again after a node is shot
*/
on_fail = pcmk_on_fail_ignore;
}
switch (on_fail) {
case pcmk_on_fail_ignore:
/* nothing to do */
break;
case pcmk_on_fail_demote:
pe__set_resource_flags(rsc, pcmk_rsc_failed);
demote_action(rsc, node, FALSE);
break;
case pcmk_on_fail_fence_node:
/* treat it as if it is still running
* but also mark the node as unclean
*/
reason = crm_strdup_printf("%s failed there", rsc->id);
pe_fence_node(rsc->cluster, node, reason, FALSE);
free(reason);
break;
case pcmk_on_fail_standby_node:
node->details->standby = TRUE;
node->details->standby_onfail = TRUE;
break;
case pcmk_on_fail_block:
/* is_managed == FALSE will prevent any
* actions being sent for the resource
*/
pe__clear_resource_flags(rsc, pcmk_rsc_managed);
pe__set_resource_flags(rsc, pcmk_rsc_blocked);
break;
case pcmk_on_fail_ban:
/* make sure it comes up somewhere else
* or not at all
*/
resource_location(rsc, node, -INFINITY, "__action_migration_auto__",
rsc->cluster);
break;
case pcmk_on_fail_stop:
pe__set_next_role(rsc, pcmk_role_stopped, "on-fail=stop");
break;
case pcmk_on_fail_restart:
if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
stop_action(rsc, node, FALSE);
}
break;
case pcmk_on_fail_restart_container:
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
if (rsc->container && pe_rsc_is_bundled(rsc)) {
/* A bundle's remote connection can run on a different node than
* the bundle's container. We don't necessarily know where the
* container is running yet, so remember it and add a stop
* action for it later.
*/
rsc->cluster->stop_needed =
g_list_prepend(rsc->cluster->stop_needed, rsc->container);
} else if (rsc->container) {
stop_action(rsc->container, node, FALSE);
} else if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
stop_action(rsc, node, FALSE);
}
break;
case pcmk_on_fail_reset_remote:
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) {
tmpnode = NULL;
if (rsc->is_remote_node) {
tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id);
}
if (tmpnode &&
pe__is_remote_node(tmpnode) &&
tmpnode->details->remote_was_fenced == 0) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(rsc->cluster, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (rsc->role > pcmk_role_stopped) {
stop_action(rsc, node, FALSE);
}
/* if reconnect delay is in use, prevent the connection from exiting the
* "STOPPED" role until the failure is cleared by the delay timeout. */
if (rsc->remote_reconnect_ms) {
pe__set_next_role(rsc, pcmk_role_stopped, "remote reset");
}
break;
}
/* ensure a remote-node connection failure forces an unclean remote-node
* to be fenced. By setting unseen = FALSE, the remote-node failure will
* result in a fencing operation regardless if we're going to attempt to
* reconnect to the remote-node in this transition or not. */
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed) && rsc->is_remote_node) {
tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
tmpnode->details->unseen = FALSE;
}
}
if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__config_warn("Detected active orphan %s running on %s",
rsc->id, pe__node_name(node));
} else {
pcmk__config_warn("Resource '%s' must be stopped manually on "
"%s because cluster is configured not to "
"stop active orphans",
rsc->id, pe__node_name(node));
}
}
native_add_running(rsc, node, rsc->cluster,
(save_on_fail != pcmk_on_fail_ignore));
switch (on_fail) {
case pcmk_on_fail_ignore:
break;
case pcmk_on_fail_demote:
case pcmk_on_fail_block:
pe__set_resource_flags(rsc, pcmk_rsc_failed);
break;
default:
pe__set_resource_flags(rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
break;
}
} else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) {
/* Only do this for older status sections that included instance numbers
* Otherwise stopped instances will appear as orphans
*/
pcmk__rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)",
rsc->clone_name, rsc->id);
free(rsc->clone_name);
rsc->clone_name = NULL;
} else {
GList *possible_matches = pe__resource_actions(rsc, node,
PCMK_ACTION_STOP, FALSE);
GList *gIter = possible_matches;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *stop = (pcmk_action_t *) gIter->data;
pe__set_action_flags(stop, pcmk_action_optional);
}
g_list_free(possible_matches);
}
/* A successful stop after migrate_to on the migration source doesn't make
* the partially migrated resource stopped on the migration target.
*/
if ((rsc->role == pcmk_role_stopped)
&& rsc->partial_migration_source
&& rsc->partial_migration_source->details == node->details
&& rsc->partial_migration_target
&& rsc->running_on) {
rsc->role = pcmk_role_started;
}
}
/* create active recurring operations as optional */
static void
process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc,
int start_index, int stop_index,
GList *sorted_op_list, pcmk_scheduler_t *scheduler)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
CRM_ASSERT(rsc);
pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d",
rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = ID(rsc_op);
counter++;
if (node->details->online == FALSE) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline",
rsc->id, pe__node_name(node));
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pe__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pe__node_name(node), counter);
continue;
}
crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if (interval_ms == 0) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring",
id, pe__node_name(node));
continue;
}
status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: status",
id, pe__node_name(node));
continue;
}
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk__rsc_trace(rsc, "Creating %s on %s", key, pe__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei)
&& pcmk__str_eq(status, "0", pcmk__str_casei)) {
*stop_index = counter;
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
*start_index = counter;
} else if ((implied_monitor_start <= *stop_index)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((scheduler->shutdown_lock > 0)
&& (get_effective_time(scheduler)
> (lock_time + scheduler->shutdown_lock))) {
pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pe__node_name(node));
pe__clear_resource_history(rsc, node);
} else {
/* @COMPAT I don't like breaking const signatures, but
* rsc->lock_node should really be const -- we just can't change it
* until the next API compatibility break.
*/
rsc->lock_node = (pcmk_node_t *) node;
rsc->lock_time = lock_time;
}
}
}
/*!
* \internal
* \brief Unpack one lrm_resource entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry lrm_resource XML being unpacked
* \param[in,out] scheduler Scheduler data
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pcmk_resource_t *
unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource,
pcmk_scheduler_t *scheduler)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = pcmk_role_unknown;
const char *rsc_id = ID(lrm_resource);
pcmk_resource_t *rsc = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
xmlNode *rsc_op = NULL;
xmlNode *last_failure = NULL;
enum action_fail_response on_fail = pcmk_on_fail_ignore;
enum rsc_role_e saved_role = pcmk_role_unknown;
if (rsc_id == NULL) {
pcmk__config_err("Ignoring invalid " XML_LRM_TAG_RESOURCE
" entry: No " XML_ATTR_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " XML_LRM_TAG_RESOURCE " for %s on %s",
rsc_id, pe__node_name(node));
// Build a list of individual lrm_rsc_op entries, so we can sort them
for (rsc_op = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(scheduler, node, rsc_id);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_orphan_resource(lrm_resource, node, scheduler);
}
}
CRM_ASSERT(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, scheduler);
}
/* process operations */
saved_role = rsc->role;
rsc->role = pcmk_role_unknown;
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail);
}
/* create active recurring operations as optional */
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
process_recurring(node, rsc, start_index, stop_index, sorted_op_list,
scheduler);
/* no need to free the contents */
g_list_free(sorted_op_list);
process_rsc_state(rsc, node, on_fail);
if (get_target_role(rsc, &req_role)) {
if ((rsc->next_role == pcmk_role_unknown)
|| (req_role < rsc->next_role)) {
pe__set_next_role(rsc, req_role, XML_RSC_ATTR_TARGET_ROLE);
} else if (req_role > rsc->next_role) {
pcmk__rsc_info(rsc,
"%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, role2text(rsc->next_role),
role2text(req_role));
}
}
if (saved_role > rsc->role) {
rsc->role = saved_role;
}
return rsc;
}
static void
handle_orphaned_container_fillers(const xmlNode *lrm_rsc_list,
pcmk_scheduler_t *scheduler)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list);
rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) {
pcmk_resource_t *rsc;
pcmk_resource_t *container;
const char *rsc_id;
const char *container_id;
if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) {
continue;
}
container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER);
rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID);
if (container_id == NULL || rsc_id == NULL) {
continue;
}
container = pe_find_resource(scheduler->resources, container_id);
if (container == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->resources, rsc_id);
if ((rsc == NULL) || (rsc->container != NULL)
|| !pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) {
continue;
}
pcmk__rsc_trace(rsc, "Mapped container of orphaned resource %s to %s",
rsc->id, container_id);
rsc->container = container;
container->fillers = g_list_append(container->fillers, rsc);
}
}
/*!
* \internal
* \brief Unpack one node's lrm status section
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] xml CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler)
{
bool found_orphaned_container_filler = false;
// Drill down to lrm_resources section
xml = find_xml_node(xml, XML_CIB_TAG_LRM, FALSE);
if (xml == NULL) {
return;
}
xml = find_xml_node(xml, XML_LRM_TAG_RESOURCES, FALSE);
if (xml == NULL) {
return;
}
// Unpack each lrm_resource entry
for (const xmlNode *rsc_entry = first_named_child(xml, XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) {
found_orphaned_container_filler = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* orphaned container fillers to their container resource.
*/
if (found_orphaned_container_filler) {
handle_orphaned_container_fillers(xml, scheduler);
}
}
static void
set_active(pcmk_resource_t *rsc)
{
const pcmk_resource_t *top = pe__const_top_resource(rsc, false);
if (top && pcmk_is_set(top->flags, pcmk_rsc_promotable)) {
rsc->role = pcmk_role_unpromoted;
} else {
rsc->role = pcmk_role_started;
}
}
static void
set_node_score(gpointer key, gpointer value, gpointer user_data)
{
pcmk_node_t *node = value;
int *score = user_data;
node->weight = *score;
}
#define XPATH_NODE_STATE "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE
#define SUB_XPATH_LRM_RESOURCE "/" XML_CIB_TAG_LRM \
"/" XML_LRM_TAG_RESOURCES \
"/" XML_LRM_TAG_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" XML_LRM_TAG_RSC_OP
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
int target_rc, pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL),
return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" XML_ATTR_UNAME "='", node, "']"
SUB_XPATH_LRM_RESOURCE "[@" XML_ATTR_ID "='", resource, "']"
SUB_XPATH_LRM_RSC_OP "[@" XML_LRM_ATTR_TASK "='", op, "'",
NULL);
/* Need to check against transition_magic too? */
if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) {
pcmk__g_strcat(xpath,
" and @" XML_LRM_ATTR_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" XML_LRM_ATTR_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
if (xml && target_rc >= 0) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_EXEC_ERROR;
crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status);
if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) {
return NULL;
}
}
return xml;
}
static xmlNode *
find_lrm_resource(const char *rsc_id, const char *node_name,
pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" XML_ATTR_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" XML_ATTR_ID "='", rsc_id, "']",
NULL);
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
return xml;
}
/*!
* \internal
* \brief Check whether a resource has no completed action history on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node_name Node to check
*
* \return true if \p rsc_id is unknown on \p node_name, otherwise false
*/
static bool
unknown_on_node(pcmk_resource_t *rsc, const char *node_name)
{
bool result = false;
xmlXPathObjectPtr search;
GString *xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" XML_ATTR_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" XML_ATTR_ID "='", rsc->id, "']"
SUB_XPATH_LRM_RSC_OP "[@" XML_LRM_ATTR_RC "!='193']",
NULL);
search = xpath_search(rsc->cluster->input, (const char *) xpath->str);
result = (numXpathResults(search) == 0);
freeXpathObject(search);
g_string_free(xpath, TRUE);
return result;
}
/*!
* \brief Check whether a probe/monitor indicating the resource was not running
* on a node happened after some event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that monitor is being compared to
* \param[in] same_node Whether the operations are on the same node
* \param[in,out] scheduler Scheduler data
*
* \return true if such a monitor happened after event, false otherwise
*/
static bool
monitor_not_running_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, bool same_node,
pcmk_scheduler_t *scheduler)
{
/* Any probe/monitor operation on the node indicating it was not running
* there
*/
xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name,
NULL, PCMK_OCF_NOT_RUNNING, scheduler);
return (monitor && pe__is_newer_op(monitor, xml_op, same_node) > 0);
}
/*!
* \brief Check whether any non-monitor operation on a node happened after some
* event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that non-monitor is being compared to
* \param[in] same_node Whether the operations are on the same node
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
non_monitor_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, bool same_node,
pcmk_scheduler_t *scheduler)
{
xmlNode *lrm_resource = NULL;
lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler);
if (lrm_resource == NULL) {
return false;
}
for (xmlNode *op = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP);
op != NULL; op = crm_next_same_xml(op)) {
const char * task = NULL;
if (op == xml_op) {
continue;
}
task = crm_element_value(op, XML_LRM_ATTR_TASK);
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
NULL)
&& pe__is_newer_op(op, xml_op, same_node) > 0) {
return true;
}
}
return false;
}
/*!
* \brief Check whether the resource has newer state on a node after a migration
* attempt
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] migrate_to Any migrate_to event that is being compared to
* \param[in] migrate_from Any migrate_from event that is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
newer_state_after_migrate(const char *rsc_id, const char *node_name,
const xmlNode *migrate_to,
const xmlNode *migrate_from,
pcmk_scheduler_t *scheduler)
{
const xmlNode *xml_op = migrate_to;
const char *source = NULL;
const char *target = NULL;
bool same_node = false;
if (migrate_from) {
xml_op = migrate_from;
}
source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE);
target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
/* It's preferred to compare to the migrate event on the same node if
* existing, since call ids are more reliable.
*/
if (pcmk__str_eq(node_name, target, pcmk__str_casei)) {
if (migrate_from) {
xml_op = migrate_from;
same_node = true;
} else {
xml_op = migrate_to;
}
} else if (pcmk__str_eq(node_name, source, pcmk__str_casei)) {
if (migrate_to) {
xml_op = migrate_to;
same_node = true;
} else {
xml_op = migrate_from;
}
}
/* If there's any newer non-monitor operation on the node, or any newer
* probe/monitor operation on the node indicating it was not running there,
* the migration events potentially no longer matter for the node.
*/
return non_monitor_after(rsc_id, node_name, xml_op, same_node, scheduler)
|| monitor_not_running_after(rsc_id, node_name, xml_op, same_node,
scheduler);
}
/*!
* \internal
* \brief Parse migration source and target node names from history entry
*
* \param[in] entry Resource history entry for a migration action
* \param[in] source_node If not NULL, source must match this node
* \param[in] target_node If not NULL, target must match this node
* \param[out] source_name Where to store migration source node name
* \param[out] target_name Where to store migration target node name
*
* \return Standard Pacemaker return code
*/
static int
get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node,
const pcmk_node_t *target_node,
const char **source_name, const char **target_name)
{
*source_name = crm_element_value(entry, XML_LRM_ATTR_MIGRATE_SOURCE);
*target_name = crm_element_value(entry, XML_LRM_ATTR_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
XML_LRM_ATTR_MIGRATE_SOURCE " and "
XML_LRM_ATTR_MIGRATE_TARGET, ID(entry));
return pcmk_rc_unpack_error;
}
if ((source_node != NULL)
&& !pcmk__str_eq(*source_name, source_node->details->uname,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
XML_LRM_ATTR_MIGRATE_SOURCE "='%s' does not match %s",
ID(entry), *source_name, pe__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->details->uname,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
XML_LRM_ATTR_MIGRATE_TARGET "='%s' does not match %s",
ID(entry), *target_name, pe__node_name(target_node));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*
* \internal
* \brief Add a migration source to a resource's list of dangling migrations
*
* If the migrate_to and migrate_from actions in a live migration both
* succeeded, but there is no stop on the source, the migration is considered
* "dangling." Add the source to the resource's dangling migration list, which
* will be used to schedule a stop on the source without affecting the target.
*
* \param[in,out] rsc Resource involved in migration
* \param[in] node Migration source
*/
static void
add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s",
rsc->id, pe__node_name(node));
rsc->role = pcmk_role_stopped;
rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations,
(gpointer) node);
}
/*!
* \internal
* \brief Update resource role etc. after a successful migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_success(struct action_history *history)
{
/* A complete migration sequence is:
* 1. migrate_to on source node (which succeeded if we get to this function)
* 2. migrate_from on target node
* 3. stop on source node
*
* If no migrate_from has happened, the migration is considered to be
* "partial". If the migrate_from succeeded but no stop has happened, the
* migration is considered to be "dangling".
*
* If a successful migrate_to and stop have happened on the source node, we
* still need to check for a partial migration, due to scenarios (easier to
* produce with batch-limit=1) like:
*
* - A resource is migrating from node1 to node2, and a migrate_to is
* initiated for it on node1.
*
* - node2 goes into standby mode while the migrate_to is pending, which
* aborts the transition.
*
* - Upon completion of the migrate_to, a new transition schedules a stop
* on both nodes and a start on node1.
*
* - If the new transition is aborted for any reason while the resource is
* stopping on node1, the transition after that stop completes will see
* the migrate_to and stop on the source, but it's still a partial
* migration, and the resource must be stopped on node2 because it is
* potentially active there due to the migrate_to.
*
* We also need to take into account that either node's history may be
* cleared at any point in the migration process.
*/
int from_rc = PCMK_OCF_OK;
int from_status = PCMK_EXEC_PENDING;
pcmk_node_t *target_node = NULL;
xmlNode *migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
bool source_newer_op = false;
bool target_newer_state = false;
bool active_on_target = false;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
// Check for newer state on the source
source_newer_op = non_monitor_after(history->rsc->id, source, history->xml,
true, history->rsc->cluster);
// Check for a migrate_from action from this source on the target
migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM,
target, source, -1, history->rsc->cluster);
if (migrate_from != NULL) {
if (source_newer_op) {
/* There's a newer non-monitor operation on the source and a
* migrate_from on the target, so this migrate_to is irrelevant to
* the resource's state.
*/
return;
}
crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc);
crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS,
&from_status);
}
/* If the resource has newer state on both the source and target after the
* migration events, this migrate_to is irrelevant to the resource's state.
*/
target_newer_state = newer_state_after_migrate(history->rsc->id, target,
history->xml, migrate_from,
history->rsc->cluster);
if (source_newer_op && target_newer_state) {
return;
}
/* Check for dangling migration (migrate_from succeeded but stop not done).
* We know there's no stop because we already returned if the target has a
* migrate_from and the source has any newer non-monitor operation.
*/
if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) {
add_dangling_migration(history->rsc, history->node);
return;
}
/* Without newer state, this migrate_to implies the resource is active.
* (Clones are not allowed to migrate, so role can't be promoted.)
*/
history->rsc->role = pcmk_role_started;
target_node = pe_find_node(history->rsc->cluster->nodes, target);
active_on_target = !target_newer_state && (target_node != NULL)
&& target_node->details->online;
if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target
if (active_on_target) {
native_add_running(history->rsc, target_node, history->rsc->cluster,
TRUE);
} else {
// Mark resource as failed, require recovery, and prevent migration
pe__set_resource_flags(history->rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
pe__clear_resource_flags(history->rsc, pcmk_rsc_migratable);
}
return;
}
// The migrate_from is pending, complete but erased, or to be scheduled
/* If there is no history at all for the resource on an online target, then
* it was likely cleaned. Just return, and we'll schedule a probe. Once we
* have the probe result, it will be reflected in target_newer_state.
*/
if ((target_node != NULL) && target_node->details->online
&& unknown_on_node(history->rsc, target)) {
return;
}
if (active_on_target) {
pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes,
source);
native_add_running(history->rsc, target_node, history->rsc->cluster,
FALSE);
if ((source_node != NULL) && source_node->details->online) {
/* This is a partial migration: the migrate_to completed
* successfully on the source, but the migrate_from has not
* completed. Remember the source and target; if the newly
* chosen target remains the same when we schedule actions
* later, we may continue with the migration.
*/
history->rsc->partial_migration_target = target_node;
history->rsc->partial_migration_source = source_node;
}
} else if (!source_newer_op) {
// Mark resource as failed, require recovery, and prevent migration
pe__set_resource_flags(history->rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
pe__clear_resource_flags(history->rsc, pcmk_rsc_migratable);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_failure(struct action_history *history)
{
xmlNode *target_migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->role = pcmk_role_started;
// Check for migrate_from on the target
target_migrate_from = find_lrm_op(history->rsc->id,
PCMK_ACTION_MIGRATE_FROM, target, source,
PCMK_OCF_OK, history->rsc->cluster);
if (/* If the resource state is unknown on the target, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, target)
/* If the resource has newer state on the target after the migration
* events, this migrate_to no longer matters for the target.
*/
&& !newer_state_after_migrate(history->rsc->id, target, history->xml,
target_migrate_from,
history->rsc->cluster)) {
/* The resource has no newer state on the target, so assume it's still
* active there.
* (if it is up).
*/
pcmk_node_t *target_node = pe_find_node(history->rsc->cluster->nodes,
target);
if (target_node && target_node->details->online) {
native_add_running(history->rsc, target_node, history->rsc->cluster,
FALSE);
}
} else if (!non_monitor_after(history->rsc->id, source, history->xml, true,
history->rsc->cluster)) {
/* We know the resource has newer state on the target, but this
* migrate_to still matters for the source as long as there's no newer
* non-monitor operation there.
*/
// Mark node as having dangling migration so we can force a stop later
history->rsc->dangling_migrations =
g_list_prepend(history->rsc->dangling_migrations,
(gpointer) history->node);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_from action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_from_failure(struct action_history *history)
{
xmlNode *source_migrate_to = NULL;
const char *source = NULL;
const char *target = NULL;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, NULL, history->node, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->role = pcmk_role_started;
// Check for a migrate_to on the source
source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO,
source, target, PCMK_OCF_OK,
history->rsc->cluster);
if (/* If the resource state is unknown on the source, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, source)
/* If the resource has newer state on the source after the migration
* events, this migrate_from no longer matters for the source.
*/
&& !newer_state_after_migrate(history->rsc->id, source,
source_migrate_to, history->xml,
history->rsc->cluster)) {
/* The resource has no newer state on the source, so assume it's still
* active there (if it is up).
*/
pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes,
source);
if (source_node && source_node->details->online) {
native_add_running(history->rsc, source_node, history->rsc->cluster,
TRUE);
}
}
}
/*!
* \internal
* \brief Add an action to cluster's list of failed actions
*
* \param[in,out] history Parsed action result history
*/
static void
record_failed_op(struct action_history *history)
{
if (!(history->node->details->online)) {
return;
}
for (const xmlNode *xIter = history->rsc->cluster->failed->children;
xIter != NULL; xIter = xIter->next) {
const char *key = pe__xe_history_key(xIter);
const char *uname = crm_element_value(xIter, XML_ATTR_UNAME);
if (pcmk__str_eq(history->key, key, pcmk__str_none)
&& pcmk__str_eq(uname, history->node->details->uname,
pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s",
history->key, pe__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pe__node_name(history->node));
crm_xml_add(history->xml, XML_ATTR_UNAME, history->node->details->uname);
crm_xml_add(history->xml, XML_LRM_ATTR_RSCID, history->rsc->id);
add_node_copy(history->rsc->cluster->failed, history->xml);
}
static char *
last_change_str(const xmlNode *xml_op)
{
time_t when;
char *result = NULL;
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&when) == pcmk_ok) {
char *when_s = pcmk__epoch2str(&when, 0);
const char *p = strchr(when_s, ' ');
// Skip day of week to make message shorter
if ((p != NULL) && (*(++p) != '\0')) {
result = strdup(p);
CRM_ASSERT(result != NULL);
}
free(when_s);
}
if (result == NULL) {
result = strdup("unknown time");
CRM_ASSERT(result != NULL);
}
return result;
}
/*!
* \internal
* \brief Compare two on-fail values
*
* \param[in] first One on-fail value to compare
* \param[in] second The other on-fail value to compare
*
* \return A negative number if second is more severe than first, zero if they
* are equal, or a positive number if first is more severe than second.
* \note This is only needed until the action_fail_response values can be
* renumbered at the next API compatibility break.
*/
static int
cmp_on_fail(enum action_fail_response first, enum action_fail_response second)
{
switch (first) {
case pcmk_on_fail_demote:
switch (second) {
case pcmk_on_fail_ignore:
return 1;
case pcmk_on_fail_demote:
return 0;
default:
return -1;
}
break;
case pcmk_on_fail_reset_remote:
switch (second) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
return 1;
case pcmk_on_fail_reset_remote:
return 0;
default:
return -1;
}
break;
case pcmk_on_fail_restart_container:
switch (second) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
case pcmk_on_fail_reset_remote:
return 1;
case pcmk_on_fail_restart_container:
return 0;
default:
return -1;
}
break;
default:
break;
}
switch (second) {
case pcmk_on_fail_demote:
return (first == pcmk_on_fail_ignore)? -1 : 1;
case pcmk_on_fail_reset_remote:
switch (first) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
return -1;
default:
return 1;
}
break;
case pcmk_on_fail_restart_container:
switch (first) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
case pcmk_on_fail_reset_remote:
return -1;
default:
return 1;
}
break;
default:
break;
}
return first - second;
}
/*!
* \internal
* \brief Ban a resource (or its clone if an anonymous instance) from all nodes
*
* \param[in,out] rsc Resource to ban
*/
static void
ban_from_all_nodes(pcmk_resource_t *rsc)
{
int score = -INFINITY;
pcmk_resource_t *fail_rsc = rsc;
if (fail_rsc->parent != NULL) {
pcmk_resource_t *parent = uber_parent(fail_rsc);
if (pe_rsc_is_anon_clone(parent)) {
/* For anonymous clones, if an operation with on-fail=stop fails for
* any instance, the entire clone must stop.
*/
fail_rsc = parent;
}
}
// Ban the resource from all nodes
crm_notice("%s will not be started under current conditions", fail_rsc->id);
if (fail_rsc->allowed_nodes != NULL) {
g_hash_table_destroy(fail_rsc->allowed_nodes);
}
fail_rsc->allowed_nodes = pe__node_list2table(rsc->cluster->nodes);
g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score);
}
/*!
* \internal
* \brief Get configured failure handling and role after failure for an action
*
* \param[in,out] history Unpacked action history entry
* \param[out] on_fail Where to set configured failure handling
* \param[out] fail_role Where to set to role after failure
*/
static void
unpack_failure_handling(struct action_history *history,
enum action_fail_response *on_fail,
enum rsc_role_e *fail_role)
{
xmlNode *config = pcmk__find_action_config(history->rsc, history->task,
history->interval_ms, true);
GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node,
history->task,
history->interval_ms, config);
const char *on_fail_str = g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
*on_fail = pcmk__parse_on_fail(history->rsc, history->task,
history->interval_ms, on_fail_str);
*fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail,
meta);
g_hash_table_destroy(meta);
}
/*!
* \internal
* \brief Update resource role, failure handling, etc., after a failed action
*
* \param[in,out] history Parsed action result history
* \param[in] config_on_fail Action failure handling from configuration
* \param[in] fail_role Resource's role after failure of this action
* \param[out] last_failure This will be set to the history XML
* \param[in,out] on_fail Actual handling of action result
*/
static void
unpack_rsc_op_failure(struct action_history *history,
enum action_fail_response config_on_fail,
enum rsc_role_e fail_role, xmlNode **last_failure,
enum action_fail_response *on_fail)
{
bool is_probe = false;
char *last_change_s = NULL;
*last_failure = history->xml;
is_probe = pcmk_xe_is_probe(history->xml);
last_change_s = last_change_str(history->xml);
if (!pcmk_is_set(history->rsc->cluster->flags, pcmk_sched_symmetric_cluster)
&& (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
crm_trace("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " CRM_XS " exit-status=%d id=%s",
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pe__node_name(history->node), last_change_s,
history->exit_status, history->id);
} else {
pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of "
"%s on %s at %s " CRM_XS " exit-status=%d id=%s",
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pe__node_name(history->node), last_change_s,
history->exit_status, history->id);
if (is_probe && (history->exit_status != PCMK_OCF_OK)
&& (history->exit_status != PCMK_OCF_NOT_RUNNING)
&& (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) {
/* A failed (not just unexpected) probe result could mean the user
* didn't know resources will be probed even where they can't run.
*/
crm_notice("If it is not possible for %s to run on %s, see "
"the resource-discovery option for location constraints",
history->rsc->id, pe__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (cmp_on_fail(*on_fail, config_on_fail) < 0) {
pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s",
fail2text(*on_fail), fail2text(config_on_fail),
history->key);
*on_fail = config_on_fail;
}
if (strcmp(history->task, PCMK_ACTION_STOP) == 0) {
resource_location(history->rsc, history->node, -INFINITY,
"__stop_fail__", history->rsc->cluster);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) {
unpack_migrate_to_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) {
unpack_migrate_from_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->role = pcmk_role_promoted;
} else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) {
if (config_on_fail == pcmk_on_fail_block) {
history->rsc->role = pcmk_role_promoted;
pe__set_next_role(history->rsc, pcmk_role_stopped,
"demote with on-fail=block");
} else if (history->exit_status == PCMK_OCF_NOT_RUNNING) {
history->rsc->role = pcmk_role_stopped;
} else {
/* Staying in the promoted role would put the scheduler and
* controller into a loop. Setting the role to unpromoted is not
* dangerous because the resource will be stopped as part of
* recovery, and any promotion will be ordered after that stop.
*/
history->rsc->role = pcmk_role_unpromoted;
}
}
if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
/* leave stopped */
pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id);
history->rsc->role = pcmk_role_stopped;
} else if (history->rsc->role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id);
set_active(history->rsc);
}
pcmk__rsc_trace(history->rsc,
"Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s",
history->rsc->id, role2text(history->rsc->role),
pcmk__btoa(history->node->details->unclean),
fail2text(config_on_fail), role2text(fail_role));
if ((fail_role != pcmk_role_started)
&& (history->rsc->next_role < fail_role)) {
pe__set_next_role(history->rsc, fail_role, "failure");
}
if (fail_role == pcmk_role_stopped) {
ban_from_all_nodes(history->rsc);
}
}
/*!
* \internal
* \brief Block a resource with a failed action if it cannot be recovered
*
* If resource action is a failed stop and fencing is not possible, mark the
* resource as unmanaged and blocked, since recovery cannot be done.
*
* \param[in,out] history Parsed action history entry
*/
static void
block_if_unrecoverable(struct action_history *history)
{
char *last_change_s = NULL;
if (strcmp(history->task, PCMK_ACTION_STOP) != 0) {
return; // All actions besides stop are always recoverable
}
if (pe_can_fence(history->node->details->data_set, history->node)) {
return; // Failed stops are recoverable via fencing
}
last_change_s = last_change_str(history->xml);
pcmk__sched_err("No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
CRM_XS " rc=%d id=%s",
history->rsc->id, history->task,
pe__node_name(history->node),
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
last_change_s, history->exit_status, history->id);
free(last_change_s);
pe__clear_resource_flags(history->rsc, pcmk_rsc_managed);
pe__set_resource_flags(history->rsc, pcmk_rsc_blocked);
}
/*!
* \internal
* \brief Update action history's execution status and why
*
* \param[in,out] history Parsed action history entry
* \param[out] why Where to store reason for update
* \param[in] value New value
* \param[in] reason Description of why value was changed
*/
static inline void
remap_because(struct action_history *history, const char **why, int value,
const char *reason)
{
if (history->execution_status != value) {
history->execution_status = value;
*why = reason;
}
}
/*!
* \internal
* \brief Remap informational monitor results and operation status
*
* For the monitor results, certain OCF codes are for providing extended information
* to the user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by Pacemaker.
*
* For operation status, the action result can be used to determine an appropriate
* status for the purposes of responding to the action. The status provided by the
* executor is not directly usable since the executor does not know what was expected.
*
* \param[in,out] history Parsed action history entry
* \param[in,out] on_fail What should be done about the result
* \param[in] expired Whether result is expired
*
* \note If the result is remapped and the node is not shutting down or failed,
* the operation will be recorded in the scheduler data's list of failed
* operations to highlight it for the user.
*
* \note This may update the resource's current and next role.
*/
static void
remap_operation(struct action_history *history,
enum action_fail_response *on_fail, bool expired)
{
bool is_probe = false;
int orig_exit_status = history->exit_status;
int orig_exec_status = history->execution_status;
const char *why = NULL;
const char *task = history->task;
// Remap degraded results to their successful counterparts
history->exit_status = pcmk__effective_rc(history->exit_status);
if (history->exit_status != orig_exit_status) {
why = "degraded result";
if (!expired && (!history->node->details->shutdown
|| history->node->details->online)) {
record_failed_op(history);
}
}
if (!pe_rsc_is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& ((history->execution_status != PCMK_EXEC_DONE)
|| (history->exit_status != PCMK_OCF_NOT_RUNNING))) {
history->execution_status = PCMK_EXEC_DONE;
history->exit_status = PCMK_OCF_NOT_RUNNING;
why = "equivalent probe result";
}
/* If the executor reported an execution status of anything but done or
* error, consider that final. But for done or error, we know better whether
* it should be treated as a failure or not, because we know the expected
* result.
*/
switch (history->execution_status) {
case PCMK_EXEC_DONE:
case PCMK_EXEC_ERROR:
break;
// These should be treated as node-fatal
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"node-fatal error");
goto remap_done;
default:
goto remap_done;
}
is_probe = pcmk_xe_is_probe(history->xml);
if (is_probe) {
task = "probe";
}
if (history->expected_exit_status < 0) {
/* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with
* Heartbeat 2.0.7 or earlier as the cluster layer, did not include the
* expected exit status in the transition key, which (along with the
* similar case of a corrupted transition key in the CIB) will be
* reported to this function as -1. Pacemaker 2.0+ does not support
* rolling upgrades from those versions or processing of saved CIB files
* from those versions, so we do not need to care much about this case.
*/
remap_because(history, &why, PCMK_EXEC_ERROR,
"obsolete history format");
pcmk__config_warn("Expected result not found for %s on %s "
"(corrupt or obsolete CIB?)",
history->key, pe__node_name(history->node));
} else if (history->exit_status == history->expected_exit_status) {
remap_because(history, &why, PCMK_EXEC_DONE, "expected result");
} else {
remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result");
pcmk__rsc_debug(history->rsc,
"%s on %s: expected %d (%s), got %d (%s%s%s)",
history->key, pe__node_name(history->node),
history->expected_exit_status,
services_ocf_exitcode_str(history->expected_exit_status),
history->exit_status,
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""));
}
switch (history->exit_status) {
case PCMK_OCF_OK:
if (is_probe
&& (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active on %s at %s",
history->rsc->id, pe__node_name(history->node),
last_change_s);
free(last_change_s);
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe
|| (history->expected_exit_status == history->exit_status)
|| !pcmk_is_set(history->rsc->flags, pcmk_rsc_managed)) {
/* For probes, recurring monitors for the Stopped role, and
* unmanaged resources, "not running" is not considered a
* failure.
*/
remap_because(history, &why, PCMK_EXEC_DONE, "exit status");
history->rsc->role = pcmk_role_stopped;
*on_fail = pcmk_on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe
&& (history->exit_status != history->expected_exit_status)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active and promoted on %s at %s",
history->rsc->id, pe__node_name(history->node),
last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->role = pcmk_role_promoted;
}
break;
case PCMK_OCF_FAILED_PROMOTED:
if (!expired) {
history->rsc->role = pcmk_role_promoted;
}
remap_because(history, &why, PCMK_EXEC_ERROR, "exit status");
break;
case PCMK_OCF_NOT_CONFIGURED:
remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status");
break;
case PCMK_OCF_UNIMPLEMENT_FEATURE:
{
guint interval_ms = 0;
crm_element_value_ms(history->xml, XML_LRM_ATTR_INTERVAL_MS,
&interval_ms);
if (interval_ms == 0) {
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"exit status");
} else {
remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED,
"exit status");
}
}
break;
case PCMK_OCF_NOT_INSTALLED:
case PCMK_OCF_INVALID_PARAM:
case PCMK_OCF_INSUFFICIENT_PRIV:
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status");
break;
default:
if (history->execution_status == PCMK_EXEC_DONE) {
char *last_change_s = last_change_str(history->xml);
crm_info("Treating unknown exit status %d from %s of %s "
"on %s at %s as failure",
history->exit_status, task, history->rsc->id,
pe__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure,
rsc->cluster);
} else {
pcmk__op_digest_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->cluster);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, pe__xe_history_key(xml_op),
node->details->id);
break;
case pcmk__digest_match:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn,
pcmk_scheduler_t *scheduler)
{
pcmk_node_t *remote_node = pe_find_node(scheduler->nodes, remote_conn->id);
if (remote_node) {
pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, scheduler);
order_actions(fence, action, pcmk__ar_first_implies_then);
}
}
static bool
should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task,
guint interval_ms, bool is_last_failure)
{
/* Clearing failures of recurring monitors has special concerns. The
* executor reports only changes in the monitor result, so if the
* monitor is still active and still getting the same failure result,
* that will go undetected after the failure is cleared.
*
* Also, the operation history will have the time when the recurring
* monitor result changed to the given code, not the time when the
* result last happened.
*
* @TODO We probably should clear such failures only when the failure
* timeout has passed since the last occurrence of the failed result.
* However we don't record that information. We could maybe approximate
* that by clearing only if there is a more recent successful monitor or
* stop result, but we don't even have that information at this point
* since we are still unpacking the resource's operation history.
*
* This is especially important for remote connection resources with a
* reconnect interval, so in that case, we skip clearing failures
* if the remote node hasn't been fenced.
*/
if (rsc->remote_reconnect_ms
&& pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pcmk_node_t *remote_node = pe_find_node(rsc->cluster->nodes, rsc->id);
if (remote_node && !remote_node->details->remote_was_fenced) {
if (is_last_failure) {
crm_info("Waiting to clear monitor failure for remote node %s"
" until fencing has occurred", rsc->id);
}
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Check operation age and schedule failure clearing when appropriate
*
* This function has two distinct purposes. The first is to check whether an
* operation history entry is expired (i.e. the resource has a failure timeout,
* the entry is older than the timeout, and the resource either has no fail
* count or its fail count is entirely older than the timeout). The second is to
* schedule fail count clearing when appropriate (i.e. the operation is expired
* and either the resource has an expired fail count or the operation is a
* last_failure for a remote connection resource with a reconnect interval,
* or the operation is a last_failure for a start or monitor operation and the
* resource's parameters have changed since the operation).
*
* \param[in,out] history Parsed action result history
*
* \return true if operation history entry is expired, otherwise false
*/
static bool
check_operation_expiry(struct action_history *history)
{
bool expired = false;
bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0");
time_t last_run = 0;
int unexpired_fail_count = 0;
const char *clear_reason = NULL;
if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) {
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not expired: "
"Not Installed does not expire",
history->id, pe__node_name(history->node));
return false; // "Not installed" must always be cleared manually
}
if ((history->rsc->failure_timeout > 0)
&& (crm_element_value_epoch(history->xml, XML_RSC_OP_LAST_CHANGE,
&last_run) == 0)) {
// Resource has a failure-timeout, and history entry has a timestamp
time_t now = get_effective_time(history->rsc->cluster);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + history->rsc->failure_timeout))
&& !should_ignore_failure_timeout(history->rsc, history->task,
history->interval_ms,
is_last_failure)) {
expired = true;
}
// Does the resource as a whole have an unexpired fail count?
unexpired_fail_count = pe_get_failcount(history->node, history->rsc,
&last_failure,
pcmk__fc_effective,
history->xml);
// Update scheduler recheck time according to *last* failure
crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds"
" last-failure@%lld",
history->id, (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count,
history->rsc->failure_timeout, (long long) last_failure);
last_failure += history->rsc->failure_timeout + 1;
if (unexpired_fail_count && (now < last_failure)) {
pe__update_recheck_time(last_failure, history->rsc->cluster,
"fail count expiration");
}
}
if (expired) {
if (pe_get_failcount(history->node, history->rsc, NULL,
pcmk__fc_default, history->xml)) {
// There is a fail count ignoring timeout
if (unexpired_fail_count == 0) {
// There is no fail count considering timeout
clear_reason = "it expired";
} else {
/* This operation is old, but there is an unexpired fail count.
* In a properly functioning cluster, this should only be
* possible if this operation is not a failure (otherwise the
* fail count should be expired too), so this is really just a
* failsafe.
*/
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Unexpired fail count",
history->id, pe__node_name(history->node));
expired = false;
}
} else if (is_last_failure
&& (history->rsc->remote_reconnect_ms != 0)) {
/* Clear any expired last failure when reconnect interval is set,
* even if there is no fail count.
*/
clear_reason = "reconnect interval is set";
}
}
if (!expired && is_last_failure
&& should_clear_for_param_change(history->xml, history->task,
history->rsc, history->node)) {
clear_reason = "resource parameters have changed";
}
if (clear_reason != NULL) {
pcmk_action_t *clear_op = NULL;
// Schedule clearing of the fail count
clear_op = pe__clear_failcount(history->rsc, history->node,
clear_reason, history->rsc->cluster);
if (pcmk_is_set(history->rsc->cluster->flags,
pcmk_sched_fencing_enabled)
&& (history->rsc->remote_reconnect_ms != 0)) {
/* If we're clearing a remote connection due to a reconnect
* interval, we want to wait until any scheduled fencing
* completes.
*
* We could limit this to remote_node->details->unclean, but at
* this point, that's always true (it won't be reliable until
* after unpack_node_history() is done).
*/
crm_info("Clearing %s failure will wait until any scheduled "
"fencing of %s completes",
history->task, history->rsc->id);
order_after_remote_fencing(clear_op, history->rsc,
history->rsc->cluster);
}
}
if (expired && (history->interval_ms == 0)
&& pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
switch (history->exit_status) {
case PCMK_OCF_OK:
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Don't expire probes that return these values
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Probe result",
history->id, pe__node_name(history->node));
expired = false;
break;
}
}
return expired;
}
int
pe__target_rc_from_xml(const xmlNode *xml_op)
{
int target_rc = 0;
const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
/*!
* \internal
* \brief Update a resource's state for an action result
*
* \param[in,out] history Parsed action history entry
* \param[in] exit_status Exit status to base new state on
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
update_resource_state(struct action_history *history, int exit_status,
const xmlNode *last_failure,
enum action_fail_response *on_fail)
{
bool clear_past_failure = false;
if ((exit_status == PCMK_OCF_NOT_INSTALLED)
|| (!pe_rsc_is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml))) {
history->rsc->role = pcmk_role_stopped;
} else if (exit_status == PCMK_OCF_NOT_RUNNING) {
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR,
pcmk__str_none)) {
if ((last_failure != NULL)
&& pcmk__str_eq(history->key, pe__xe_history_key(last_failure),
pcmk__str_none)) {
clear_past_failure = true;
}
if (history->rsc->role < pcmk_role_started) {
set_active(history->rsc);
}
} else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) {
history->rsc->role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) {
history->rsc->role = pcmk_role_stopped;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
history->rsc->role = pcmk_role_promoted;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE,
pcmk__str_none)) {
if (*on_fail == pcmk_on_fail_demote) {
// Demote clears an error only if on-fail=demote
clear_past_failure = true;
}
history->rsc->role = pcmk_role_unpromoted;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
history->rsc->role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
unpack_migrate_to_success(history);
} else if (history->rsc->role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "%s active on %s",
history->rsc->id, pe__node_name(history->node));
set_active(history->rsc);
}
if (!clear_past_failure) {
return;
}
switch (*on_fail) {
case pcmk_on_fail_stop:
case pcmk_on_fail_ban:
case pcmk_on_fail_standby_node:
case pcmk_on_fail_fence_node:
pcmk__rsc_trace(history->rsc,
"%s (%s) is not cleared by a completed %s",
history->rsc->id, fail2text(*on_fail),
history->task);
break;
case pcmk_on_fail_block:
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
case pcmk_on_fail_restart_container:
*on_fail = pcmk_on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures");
break;
case pcmk_on_fail_reset_remote:
if (history->rsc->remote_reconnect_ms == 0) {
/* With no reconnect interval, the connection is allowed to
* start again after the remote node is fenced and
* completely stopped. (With a reconnect interval, we wait
* for the failure to be cleared entirely before attempting
* to reconnect.)
*/
*on_fail = pcmk_on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures and reset remote");
}
break;
}
}
/*!
* \internal
* \brief Check whether a given history entry matters for resource state
*
* \param[in] history Parsed action history entry
*
* \return true if action can affect resource state, otherwise false
*/
static inline bool
can_affect_state(struct action_history *history)
{
#if 0
/* @COMPAT It might be better to parse only actions we know we're interested
* in, rather than exclude a couple we don't. However that would be a
* behavioral change that should be done at a major or minor series release.
* Currently, unknown operations can affect whether a resource is considered
* active and/or failed.
*/
return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
"asyncmon", NULL);
#else
return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY,
PCMK_ACTION_META_DATA, NULL);
#endif
}
/*!
* \internal
* \brief Unpack execution/exit status and exit reason from a history entry
*
* \param[in,out] history Action history entry to unpack
*
* \return Standard Pacemaker return code
*/
static int
unpack_action_result(struct action_history *history)
{
if ((crm_element_value_int(history->xml, XML_LRM_ATTR_OPSTATUS,
&(history->execution_status)) < 0)
|| (history->execution_status < PCMK_EXEC_PENDING)
|| (history->execution_status > PCMK_EXEC_MAX)
|| (history->execution_status == PCMK_EXEC_CANCELLED)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " XML_LRM_ATTR_OPSTATUS " '%s'",
history->id, history->rsc->id,
pe__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
XML_LRM_ATTR_OPSTATUS),
""));
return pcmk_rc_unpack_error;
}
if ((crm_element_value_int(history->xml, XML_LRM_ATTR_RC,
&(history->exit_status)) < 0)
|| (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) {
#if 0
/* @COMPAT We should ignore malformed entries, but since that would
* change behavior, it should be done at a major or minor series
* release.
*/
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " XML_LRM_ATTR_RC " '%s'",
history->id, history->rsc->id,
pe__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
XML_LRM_ATTR_RC),
""));
return pcmk_rc_unpack_error;
#else
history->exit_status = CRM_EX_ERROR;
#endif
}
history->exit_reason = crm_element_value(history->xml,
XML_LRM_ATTR_EXIT_REASON);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Process an action history entry whose result expired
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
*
* \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the
* entry needs no further processing)
*/
static int
process_expired_result(struct action_history *history, int orig_exit_status)
{
if (!pe_rsc_is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& (orig_exit_status != history->expected_exit_status)) {
if (history->rsc->role <= pcmk_role_stopped) {
history->rsc->role = pcmk_role_unknown;
}
crm_trace("Ignoring resource history entry %s for probe of %s on %s: "
"Masked failure expired",
history->id, history->rsc->id,
pe__node_name(history->node));
return pcmk_rc_ok;
}
if (history->exit_status == history->expected_exit_status) {
return pcmk_rc_undetermined; // Only failures expire
}
if (history->interval_ms == 0) {
crm_notice("Ignoring resource history entry %s for %s of %s on %s: "
"Expired failure",
history->id, history->task, history->rsc->id,
pe__node_name(history->node));
return pcmk_rc_ok;
}
if (history->node->details->online && !history->node->details->unclean) {
/* Reschedule the recurring action. schedule_cancel() won't work at
* this stage, so as a hacky workaround, forcibly change the restart
* digest so pcmk__check_action_config() does what we want later.
*
* @TODO We should skip this if there is a newer successful monitor.
* Also, this causes rescheduling only if the history entry
* has an op-digest (which the expire-non-blocked-failure
* scheduler regression test doesn't, but that may not be a
* realistic scenario in production).
*/
crm_notice("Rescheduling %s-interval %s of %s on %s "
"after failure expired",
pcmk__readable_interval(history->interval_ms), history->task,
history->rsc->id, pe__node_name(history->node));
crm_xml_add(history->xml, XML_LRM_ATTR_RESTART_DIGEST,
"calculated-failure-timeout");
return pcmk_rc_ok;
}
return pcmk_rc_undetermined;
}
/*!
* \internal
* \brief Process a masked probe failure
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
mask_probe_failure(struct action_history *history, int orig_exit_status,
const xmlNode *last_failure,
enum action_fail_response *on_fail)
{
pcmk_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pcmk_rsc_unique)) {
ban_rsc = uber_parent(history->rsc);
}
crm_notice("Treating probe result '%s' for %s on %s as 'not running'",
services_ocf_exitcode_str(orig_exit_status), history->rsc->id,
pe__node_name(history->node));
update_resource_state(history, history->expected_exit_status, last_failure,
on_fail);
crm_xml_add(history->xml, XML_ATTR_UNAME, history->node->details->uname);
record_failed_op(history);
resource_location(ban_rsc, history->node, -INFINITY, "masked-probe-failure",
history->rsc->cluster);
}
/*!
* \internal Check whether a given failure is for a given pending action
*
* \param[in] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*
* \return true if \p last_failure is failure of pending action in \p history,
* otherwise false
* \note Both \p history and \p last_failure must come from the same
* lrm_resource block, as node and resource are assumed to be the same.
*/
static bool
failure_is_newer(const struct action_history *history,
const xmlNode *last_failure)
{
guint failure_interval_ms = 0U;
long long failure_change = 0LL;
long long this_change = 0LL;
if (last_failure == NULL) {
return false; // Resource has no last_failure entry
}
if (!pcmk__str_eq(history->task,
crm_element_value(last_failure, XML_LRM_ATTR_TASK),
pcmk__str_none)) {
return false; // last_failure is for different action
}
if ((crm_element_value_ms(last_failure, XML_LRM_ATTR_INTERVAL_MS,
&failure_interval_ms) != pcmk_ok)
|| (history->interval_ms != failure_interval_ms)) {
return false; // last_failure is for action with different interval
}
if ((pcmk__scan_ll(crm_element_value(history->xml, XML_RSC_OP_LAST_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(crm_element_value(last_failure,
XML_RSC_OP_LAST_CHANGE),
&failure_change, 0LL) != pcmk_rc_ok)
|| (failure_change < this_change)) {
return false; // Failure is not known to be newer
}
return true;
}
/*!
* \internal
* \brief Update a resource's role etc. for a pending action
*
* \param[in,out] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*/
static void
process_pending_action(struct action_history *history,
const xmlNode *last_failure)
{
/* For recurring monitors, a failure is recorded only in RSC_last_failure_0,
* and there might be a RSC_monitor_INTERVAL entry with the last successful
* or pending result.
*
* If last_failure contains the failure of the pending recurring monitor
* we're processing here, and is newer, the action is no longer pending.
* (Pending results have call ID -1, which sorts last, so the last failure
* if any should be known.)
*/
if (failure_is_newer(history, last_failure)) {
return;
}
if (strcmp(history->task, PCMK_ACTION_START) == 0) {
pe__set_resource_flags(history->rsc, pcmk_rsc_start_pending);
set_active(history->rsc);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->role = pcmk_role_promoted;
} else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0)
&& history->node->details->unclean) {
/* A migrate_to action is pending on a unclean source, so force a stop
* on the target.
*/
const char *migrate_target = NULL;
pcmk_node_t *target = NULL;
migrate_target = crm_element_value(history->xml,
XML_LRM_ATTR_MIGRATE_TARGET);
target = pe_find_node(history->rsc->cluster->nodes, migrate_target);
if (target != NULL) {
stop_action(history->rsc, target, FALSE);
}
}
if (history->rsc->pending_task != NULL) {
/* There should never be multiple pending actions, but as a failsafe,
* just remember the first one processed for display purposes.
*/
return;
}
if (pcmk_is_probe(history->task, history->interval_ms)) {
/* Pending probes are currently never displayed, even if pending
* operations are requested. If we ever want to change that,
* enable the below and the corresponding part of
* native.c:native_pending_task().
*/
#if 0
history->rsc->pending_task = strdup("probe");
history->rsc->pending_node = history->node;
#endif
} else {
history->rsc->pending_task = strdup(history->task);
history->rsc->pending_node = history->node;
}
}
static void
unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum action_fail_response *on_fail)
{
int old_rc = 0;
bool expired = false;
pcmk_resource_t *parent = rsc;
enum rsc_role_e fail_role = pcmk_role_unknown;
enum action_fail_response failure_strategy = pcmk_on_fail_restart;
struct action_history history = {
.rsc = rsc,
.node = node,
.xml = xml_op,
.execution_status = PCMK_EXEC_UNKNOWN,
};
CRM_CHECK(rsc && node && xml_op, return);
history.id = ID(xml_op);
if (history.id == NULL) {
pcmk__config_err("Ignoring resource history entry for %s on %s "
"without ID", rsc->id, pe__node_name(node));
return;
}
// Task and interval
history.task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
if (history.task == NULL) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"without " XML_LRM_ATTR_TASK,
history.id, rsc->id, pe__node_name(node));
return;
}
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS,
&(history.interval_ms));
if (!can_affect_state(&history)) {
pcmk__rsc_trace(rsc,
"Ignoring resource history entry %s for %s on %s "
"with irrelevant action '%s'",
history.id, rsc->id, pe__node_name(node), history.task);
return;
}
if (unpack_action_result(&history) != pcmk_rc_ok) {
return; // Error already logged
}
history.expected_exit_status = pe__target_rc_from_xml(xml_op);
history.key = pe__xe_history_key(xml_op);
crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pe__node_name(node),
pcmk_exec_status_str(history.execution_status),
crm_exit_str(history.exit_status));
if (node->details->unclean) {
pcmk__rsc_trace(rsc,
"%s is running on %s, which is unclean (further action "
"depends on value of stop's on-fail attribute)",
rsc->id, pe__node_name(node));
}
expired = check_operation_expiry(&history);
old_rc = history.exit_status;
remap_operation(&history, on_fail, expired);
if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) {
goto done;
}
if (!pe_rsc_is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) {
mask_probe_failure(&history, old_rc, *last_failure, on_fail);
goto done;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) {
parent = uber_parent(rsc);
}
switch (history.execution_status) {
case PCMK_EXEC_PENDING:
process_pending_action(&history, *last_failure);
goto done;
case PCMK_EXEC_DONE:
update_resource_state(&history, history.exit_status, *last_failure,
on_fail);
goto done;
case PCMK_EXEC_NOT_INSTALLED:
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if (failure_strategy == pcmk_on_fail_ignore) {
crm_warn("Cannot ignore failed %s of %s on %s: "
"Resource agent doesn't exist "
CRM_XS " status=%d rc=%d id=%s",
history.task, rsc->id, pe__node_name(node),
history.execution_status, history.exit_status,
history.id);
/* Also for printing it as "FAILED" by marking it as
* pcmk_rsc_failed later
*/
*on_fail = pcmk_on_fail_ban;
}
resource_location(parent, node, -INFINITY, "hard-error",
rsc->cluster);
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
goto done;
case PCMK_EXEC_NOT_CONNECTED:
if (pe__is_guest_or_remote_node(node)
&& pcmk_is_set(node->details->remote_rsc->flags,
pcmk_rsc_managed)) {
/* We should never get into a situation where a managed remote
* connection resource is considered OK but a resource action
* behind the connection gets a "not connected" status. But as a
* fail-safe in case a bug or unusual circumstances do lead to
* that, ensure the remote connection is considered failed.
*/
pe__set_resource_flags(node->details->remote_rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
}
break; // Not done, do error handling
case PCMK_EXEC_ERROR:
case PCMK_EXEC_ERROR_HARD:
case PCMK_EXEC_ERROR_FATAL:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_INVALID:
break; // Not done, do error handling
default: // No other value should be possible at this point
break;
}
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if ((failure_strategy == pcmk_on_fail_ignore)
|| ((failure_strategy == pcmk_on_fail_restart_container)
&& (strcmp(history.task, PCMK_ACTION_STOP) == 0))) {
char *last_change_s = last_change_str(xml_op);
crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded "
CRM_XS " %s",
history.task, services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), rsc->id, pe__node_name(node),
last_change_s, history.id);
free(last_change_s);
update_resource_state(&history, history.expected_exit_status,
*last_failure, on_fail);
crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname);
pe__set_resource_flags(rsc, pcmk_rsc_ignore_failure);
record_failed_op(&history);
if ((failure_strategy == pcmk_on_fail_restart_container)
&& cmp_on_fail(*on_fail, pcmk_on_fail_restart) <= 0) {
*on_fail = failure_strategy;
}
} else {
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
if (history.execution_status == PCMK_EXEC_ERROR_HARD) {
uint8_t log_level = LOG_ERR;
if (history.exit_status == PCMK_OCF_NOT_INSTALLED) {
log_level = LOG_NOTICE;
}
do_crm_log(log_level,
"Preventing %s from restarting on %s because "
"of hard failure (%s%s%s) " CRM_XS " %s",
parent->id, pe__node_name(node),
services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, node, -INFINITY, "hard-error",
rsc->cluster);
} else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) {
pcmk__sched_err("Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " CRM_XS " %s",
parent->id,
services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, NULL, -INFINITY, "fatal-error",
rsc->cluster);
}
}
done:
pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)",
rsc->id, pe__node_name(node), history.id,
role2text(rsc->role), role2text(rsc->next_role));
}
static void
add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite,
pcmk_scheduler_t *scheduler)
{
const char *cluster_name = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_UNAME), strdup(node->details->uname));
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID),
strdup(node->details->id));
if (pcmk__str_eq(node->details->id, scheduler->dc_uuid, pcmk__str_casei)) {
scheduler->dc_node = node;
node->details->is_dc = TRUE;
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE));
} else {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE));
}
cluster_name = g_hash_table_lookup(scheduler->config_hash,
PCMK_OPT_CLUSTER_NAME);
if (cluster_name) {
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME),
strdup(cluster_name));
}
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data,
node->details->attrs, NULL, overwrite,
scheduler);
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data,
node->details->utilization, NULL,
FALSE, scheduler);
if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) {
const char *site_name = pe_node_attribute_raw(node, "site-name");
if (site_name) {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(site_name));
} else if (cluster_name) {
/* Default to cluster-name if unset */
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(cluster_name));
}
}
}
static GList *
extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter)
{
int counter = -1;
int stop_index = -1;
int start_index = -1;
xmlNode *rsc_op = NULL;
GList *gIter = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
/* extract operations */
op_list = NULL;
sorted_op_list = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) {
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP,
pcmk__str_none)) {
crm_xml_add(rsc_op, "resource", rsc);
crm_xml_add(rsc_op, XML_ATTR_UNAME, node);
op_list = g_list_prepend(op_list, rsc_op);
}
}
if (op_list == NULL) {
/* if there are no operations, there is nothing to do */
return NULL;
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
/* create active recurring operations as optional */
if (active_filter == FALSE) {
return sorted_op_list;
}
op_list = NULL;
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;
counter++;
if (start_index < stop_index) {
crm_trace("Skipping %s: not active", ID(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", ID(rsc_op));
continue;
}
op_list = g_list_append(op_list, rsc_op);
}
g_list_free(sorted_op_list);
return op_list;
}
GList *
find_operations(const char *rsc, const char *node, gboolean active_filter,
pcmk_scheduler_t *scheduler)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = find_xml_node(scheduler->input, XML_CIB_TAG_STATUS, TRUE);
pcmk_node_t *this_node = NULL;
xmlNode *node_state = NULL;
for (node_state = pcmk__xe_first_child(status); node_state != NULL;
node_state = pcmk__xe_next(node_state)) {
if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
const char *uname = crm_element_value(node_state, XML_ATTR_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pe_find_node(scheduler->nodes, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pe__is_guest_or_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
} else {
determine_online_status(node_state, this_node, scheduler);
}
if (this_node->details->online
|| pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
/* offline nodes run no resources...
* unless stonith is enabled in which case we need to
* make sure rsc start events happen after the stonith
*/
xmlNode *lrm_rsc = NULL;
tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE);
for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL;
lrm_rsc = pcmk__xe_next(lrm_rsc)) {
if (pcmk__str_eq((const char *)lrm_rsc->name,
XML_LRM_TAG_RESOURCE, pcmk__str_none)) {
const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_ID);
if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) {
continue;
}
intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
output = g_list_concat(output, intermediate);
}
}
}
}
}
return output;
}

File Metadata

Mime Type
text/x-diff
Expires
Mon, Apr 21, 6:04 PM (1 d, 6 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1665041
Default Alt Text
(418 KB)

Event Timeline