diff --git a/include/crm/pengine/pe_types.h b/include/crm/pengine/pe_types.h
index 260bfbbac7..843cf869d3 100644
--- a/include/crm/pengine/pe_types.h
+++ b/include/crm/pengine/pe_types.h
@@ -1,528 +1,525 @@
/*
* Copyright 2004-2023 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_PENGINE_PE_TYPES__H
# define PCMK__CRM_PENGINE_PE_TYPES__H
# include <stdbool.h> // bool
# include <sys/types.h> // time_t
# include <libxml/tree.h> // xmlNode
# include <glib.h> // gboolean, guint, GList, GHashTable
# include <crm/common/iso8601.h>
# include <crm/common/scheduler.h>
# include <crm/pengine/common.h>
#ifdef __cplusplus
extern "C" {
#endif
/*!
* \file
* \brief Data types for cluster status
* \ingroup pengine
*/
typedef struct pe_node_s pe_node_t;
typedef struct pe_action_s pe_action_t;
typedef struct pe_resource_s pe_resource_t;
typedef struct pe_working_set_s pe_working_set_t;
typedef struct resource_object_functions_s {
gboolean (*unpack) (pe_resource_t*, pe_working_set_t*);
pe_resource_t *(*find_rsc) (pe_resource_t *parent, const char *search,
const pe_node_t *node, int flags);
/* parameter result must be free'd */
char *(*parameter) (pe_resource_t*, pe_node_t*, gboolean, const char*,
pe_working_set_t*);
//! \deprecated will be removed in a future release
void (*print) (pe_resource_t*, const char*, long, void*);
gboolean (*active) (pe_resource_t*, gboolean);
enum rsc_role_e (*state) (const pe_resource_t*, gboolean);
pe_node_t *(*location) (const pe_resource_t*, GList**, int);
void (*free) (pe_resource_t*);
void (*count) (pe_resource_t*);
gboolean (*is_filtered) (const pe_resource_t*, GList *, gboolean);
/*!
* \brief Find a node (and optionally count all) where resource is active
*
* \param[in] rsc Resource to check
* \param[out] count_all If not NULL, set this to count of active nodes
* \param[out] count_clean If not NULL, set this to count of clean nodes
*
* \return A node where the resource is active, preferring the source node
* if the resource is involved in a partial migration or a clean,
* online node if the resource's "requires" is "quorum" or
* "nothing", or NULL if the resource is inactive.
*/
pe_node_t *(*active_node)(const pe_resource_t *rsc, unsigned int *count_all,
unsigned int *count_clean);
/*!
* \brief Get maximum resource instances per node
*
* \param[in] rsc Resource to check
*
* \return Maximum number of \p rsc instances that can be active on one node
*/
unsigned int (*max_per_node)(const pe_resource_t *rsc);
} resource_object_functions_t;
typedef struct resource_alloc_functions_s resource_alloc_functions_t;
-// @TODO Make these an enum
-
-# define pe_flag_have_quorum 0x00000001ULL
# define pe_flag_symmetric_cluster 0x00000002ULL
# define pe_flag_maintenance_mode 0x00000008ULL
# define pe_flag_stonith_enabled 0x00000010ULL
# define pe_flag_have_stonith_resource 0x00000020ULL
# define pe_flag_enable_unfencing 0x00000040ULL
# define pe_flag_concurrent_fencing 0x00000080ULL
# define pe_flag_stop_rsc_orphans 0x00000100ULL
# define pe_flag_stop_action_orphans 0x00000200ULL
# define pe_flag_stop_everything 0x00000400ULL
# define pe_flag_start_failure_fatal 0x00001000ULL
//! \deprecated
# define pe_flag_remove_after_stop 0x00002000ULL
# define pe_flag_startup_fencing 0x00004000ULL
# define pe_flag_shutdown_lock 0x00008000ULL
# define pe_flag_startup_probes 0x00010000ULL
# define pe_flag_have_status 0x00020000ULL
# define pe_flag_have_remote_nodes 0x00040000ULL
# define pe_flag_quick_location 0x00100000ULL
# define pe_flag_sanitized 0x00200000ULL
//! \deprecated
# define pe_flag_stdout 0x00400000ULL
//! Don't count total, disabled and blocked resource instances
# define pe_flag_no_counts 0x00800000ULL
/*! Skip deprecated code that is kept solely for backward API compatibility.
* (Internal code should always set this.)
*/
# define pe_flag_no_compat 0x01000000ULL
# define pe_flag_show_scores 0x02000000ULL
# define pe_flag_show_utilization 0x04000000ULL
/*!
* When scheduling, only unpack the CIB (including constraints), calculate
* as much cluster status as possible, and apply node health.
*/
# define pe_flag_check_config 0x08000000ULL
struct pe_working_set_s {
xmlNode *input;
crm_time_t *now;
/* options extracted from the input */
char *dc_uuid;
pe_node_t *dc_node;
const char *stonith_action;
const char *placement_strategy;
unsigned long long flags;
int stonith_timeout;
enum pe_quorum_policy no_quorum_policy;
GHashTable *config_hash;
GHashTable *tickets;
// Actions for which there can be only one (e.g. fence nodeX)
GHashTable *singletons;
GList *nodes;
GList *resources;
GList *placement_constraints;
GList *ordering_constraints;
GList *colocation_constraints;
GList *ticket_constraints;
GList *actions;
xmlNode *failed;
xmlNode *op_defaults;
xmlNode *rsc_defaults;
/* stats */
int num_synapse;
int max_valid_nodes; //! Deprecated (will be removed in a future release)
int order_id;
int action_id;
/* final output */
xmlNode *graph;
GHashTable *template_rsc_sets;
const char *localhost;
GHashTable *tags;
int blocked_resources;
int disabled_resources;
GList *param_check; // History entries that need to be checked
GList *stop_needed; // Containers that need stop actions
time_t recheck_by; // Hint to controller to re-run scheduler by this time
int ninstances; // Total number of resource instances
guint shutdown_lock;// How long (seconds) to lock resources to shutdown node
int priority_fencing_delay; // Priority fencing delay
void *priv;
guint node_pending_timeout; // Node pending timeout
};
struct pe_node_shared_s {
const char *id;
const char *uname;
enum node_type type;
/* @TODO convert these flags into a bitfield */
gboolean online;
gboolean standby;
gboolean standby_onfail;
gboolean pending;
gboolean unclean;
gboolean unseen;
gboolean shutdown;
gboolean expected_up;
gboolean is_dc;
gboolean maintenance;
gboolean rsc_discovery_enabled;
gboolean remote_requires_reset;
gboolean remote_was_fenced;
gboolean remote_maintenance; /* what the remote-rsc is thinking */
gboolean unpacked;
int num_resources;
pe_resource_t *remote_rsc;
GList *running_rsc; /* pe_resource_t* */
GList *allocated_rsc; /* pe_resource_t* */
GHashTable *attrs; /* char* => char* */
GHashTable *utilization;
GHashTable *digest_cache; //!< cache of calculated resource digests
int priority; // calculated based on the priority of resources running on the node
pe_working_set_t *data_set; //!< Cluster that this node is part of
};
struct pe_node_s {
int weight;
gboolean fixed; //!< \deprecated Will be removed in a future release
int count;
struct pe_node_shared_s *details;
int rsc_discover_mode;
};
# define pe_rsc_orphan 0x00000001ULL
# define pe_rsc_managed 0x00000002ULL
# define pe_rsc_block 0x00000004ULL
# define pe_rsc_orphan_container_filler 0x00000008ULL
# define pe_rsc_notify 0x00000010ULL
# define pe_rsc_unique 0x00000020ULL
# define pe_rsc_fence_device 0x00000040ULL
# define pe_rsc_promotable 0x00000080ULL
# define pe_rsc_provisional 0x00000100ULL
# define pe_rsc_allocating 0x00000200ULL
# define pe_rsc_merging 0x00000400ULL
# define pe_rsc_restarting 0x00000800ULL
# define pe_rsc_stop 0x00001000ULL
# define pe_rsc_reload 0x00002000ULL
# define pe_rsc_allow_remote_remotes 0x00004000ULL
# define pe_rsc_critical 0x00008000ULL
# define pe_rsc_failed 0x00010000ULL
# define pe_rsc_detect_loop 0x00020000ULL
# define pe_rsc_runnable 0x00040000ULL
# define pe_rsc_start_pending 0x00080000ULL
//!< \deprecated Do not use
# define pe_rsc_starting 0x00100000ULL
//!< \deprecated Do not use
# define pe_rsc_stopping 0x00200000ULL
# define pe_rsc_stop_unexpected 0x00400000ULL
# define pe_rsc_allow_migrate 0x00800000ULL
# define pe_rsc_failure_ignored 0x01000000ULL
# define pe_rsc_replica_container 0x02000000ULL
# define pe_rsc_maintenance 0x04000000ULL
# define pe_rsc_is_container 0x08000000ULL
# define pe_rsc_needs_quorum 0x10000000ULL
# define pe_rsc_needs_fencing 0x20000000ULL
# define pe_rsc_needs_unfencing 0x40000000ULL
/* *INDENT-OFF* */
enum pe_action_flags {
pe_action_pseudo = 0x00001,
pe_action_runnable = 0x00002,
pe_action_optional = 0x00004,
pe_action_print_always = 0x00008,
pe_action_have_node_attrs = 0x00010,
pe_action_implied_by_stonith = 0x00040,
pe_action_migrate_runnable = 0x00080,
pe_action_dumped = 0x00100,
pe_action_processed = 0x00200,
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
pe_action_clear = 0x00400, //! \deprecated Unused
#endif
pe_action_dangle = 0x00800,
/* This action requires one or more of its dependencies to be runnable.
* We use this to clear the runnable flag before checking dependencies.
*/
pe_action_requires_any = 0x01000,
pe_action_reschedule = 0x02000,
pe_action_tracking = 0x04000,
pe_action_dedup = 0x08000, //! Internal state tracking when creating graph
pe_action_dc = 0x10000, //! Action may run on DC instead of target
};
/* *INDENT-ON* */
struct pe_resource_s {
char *id;
char *clone_name;
xmlNode *xml;
xmlNode *orig_xml;
xmlNode *ops_xml;
pe_working_set_t *cluster;
pe_resource_t *parent;
enum pe_obj_types variant;
void *variant_opaque;
resource_object_functions_t *fns;
resource_alloc_functions_t *cmds;
enum rsc_recovery_type recovery_type;
enum pe_restart restart_type; //!< \deprecated will be removed in future release
int priority;
int stickiness;
int sort_index;
int failure_timeout;
int migration_threshold;
guint remote_reconnect_ms;
char *pending_task;
unsigned long long flags;
// @TODO merge these into flags
gboolean is_remote_node;
gboolean exclusive_discover;
/* Pay special attention to whether you want to use rsc_cons_lhs and
* rsc_cons directly, which include only colocations explicitly involving
* this resource, or call libpacemaker's pcmk__with_this_colocations() and
* pcmk__this_with_colocations() functions, which may return relevant
* colocations involving the resource's ancestors as well.
*/
//!@{
//! This field should be treated as internal to Pacemaker
GList *rsc_cons_lhs; // List of pcmk__colocation_t*
GList *rsc_cons; // List of pcmk__colocation_t*
GList *rsc_location; // List of pe__location_t*
GList *actions; // List of pe_action_t*
GList *rsc_tickets; // List of rsc_ticket*
//!@}
pe_node_t *allocated_to;
pe_node_t *partial_migration_target;
pe_node_t *partial_migration_source;
GList *running_on; /* pe_node_t* */
GHashTable *known_on; /* pe_node_t* */
GHashTable *allowed_nodes; /* pe_node_t* */
enum rsc_role_e role;
enum rsc_role_e next_role;
GHashTable *meta;
GHashTable *parameters; //! \deprecated Use pe_rsc_params() instead
GHashTable *utilization;
GList *children; /* pe_resource_t* */
GList *dangling_migrations; /* pe_node_t* */
pe_resource_t *container;
GList *fillers;
// @COMPAT These should be made const at next API compatibility break
pe_node_t *pending_node; // Node on which pending_task is happening
pe_node_t *lock_node; // Resource is shutdown-locked to this node
time_t lock_time; // When shutdown lock started
/* Resource parameters may have node-attribute-based rules, which means the
* values can vary by node. This table is a cache of parameter name/value
* tables for each node (as needed). Use pe_rsc_params() to get the table
* for a given node.
*/
GHashTable *parameter_cache; // Key = node name, value = parameters table
};
struct pe_action_s {
int id;
int priority;
pe_resource_t *rsc;
pe_node_t *node;
xmlNode *op_entry;
char *task;
char *uuid;
char *cancel_task;
char *reason;
enum pe_action_flags flags;
enum rsc_start_requirement needs;
enum action_fail_response on_fail;
enum rsc_role_e fail_role;
GHashTable *meta;
GHashTable *extra;
/*
* These two varables are associated with the constraint logic
* that involves first having one or more actions runnable before
* then allowing this action to execute.
*
* These varables are used with features such as 'clone-min' which
* requires at minimum X number of cloned instances to be running
* before an order dependency can run. Another option that uses
* this is 'require-all=false' in ordering constrants. This option
* says "only require one instance of a resource to start before
* allowing dependencies to start" -- basically, require-all=false is
* the same as clone-min=1.
*/
/* current number of known runnable actions in the before list. */
int runnable_before;
/* the number of "before" runnable actions required for this action
* to be considered runnable */
int required_runnable_before;
GList *actions_before; /* pe_action_wrapper_t* */
GList *actions_after; /* pe_action_wrapper_t* */
/* Some of the above fields could be moved to the details,
* except for API backward compatibility.
*/
void *action_details; // varies by type of action
};
typedef struct pe_ticket_s {
char *id;
gboolean granted;
time_t last_granted;
gboolean standby;
GHashTable *state;
} pe_ticket_t;
typedef struct pe_tag_s {
char *id;
GList *refs;
} pe_tag_t;
//! Internal tracking for transition graph creation
enum pe_link_state {
pe_link_not_dumped, //! Internal tracking for transition graph creation
pe_link_dumped, //! Internal tracking for transition graph creation
pe_link_dup, //! \deprecated No longer used by Pacemaker
};
enum pe_discover_e {
pe_discover_always = 0,
pe_discover_never,
pe_discover_exclusive,
};
/* *INDENT-OFF* */
enum pe_ordering {
pe_order_none = 0x0, /* deleted */
pe_order_optional = 0x1, /* pure ordering, nothing implied */
pe_order_apply_first_non_migratable = 0x2, /* Only apply this constraint's ordering if first is not migratable. */
pe_order_implies_first = 0x10, /* If 'then' is required, ensure 'first' is too */
pe_order_implies_then = 0x20, /* If 'first' is required, ensure 'then' is too */
pe_order_promoted_implies_first = 0x40, /* If 'then' is required and then's rsc is promoted, ensure 'first' becomes required too */
/* first requires then to be both runnable and migrate runnable. */
pe_order_implies_first_migratable = 0x80,
pe_order_runnable_left = 0x100, /* 'then' requires 'first' to be runnable */
pe_order_pseudo_left = 0x200, /* 'then' can only be pseudo if 'first' is runnable */
pe_order_implies_then_on_node = 0x400, /* If 'first' is required on 'nodeX',
* ensure instances of 'then' on 'nodeX' are too.
* Only really useful if 'then' is a clone and 'first' is not
*/
pe_order_probe = 0x800, /* If 'first->rsc' is
* - running but about to stop, ignore the constraint
* - otherwise, behave as runnable_left
*/
pe_order_restart = 0x1000, /* 'then' is runnable if 'first' is optional or runnable */
pe_order_stonith_stop = 0x2000, //<! \deprecated Will be removed in future release
pe_order_serialize_only = 0x4000, /* serialize */
pe_order_same_node = 0x8000, /* applies only if 'first' and 'then' are on same node */
pe_order_implies_first_printed = 0x10000, /* Like ..implies_first but only ensures 'first' is printed, not mandatory */
pe_order_implies_then_printed = 0x20000, /* Like ..implies_then but only ensures 'then' is printed, not mandatory */
pe_order_asymmetrical = 0x100000, /* Indicates asymmetrical one way ordering constraint. */
pe_order_load = 0x200000, /* Only relevant if... */
pe_order_one_or_more = 0x400000, /* 'then' is runnable only if one or more of its dependencies are too */
pe_order_anti_colocation = 0x800000,
pe_order_preserve = 0x1000000, /* Hack for breaking user ordering constraints with container resources */
pe_order_then_cancels_first = 0x2000000, // if 'then' becomes required, 'first' becomes optional
pe_order_trace = 0x4000000, /* test marker */
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
// \deprecated Use pe_order_promoted_implies_first instead
pe_order_implies_first_master = pe_order_promoted_implies_first,
#endif
};
/* *INDENT-ON* */
typedef struct pe_action_wrapper_s {
enum pe_ordering type;
enum pe_link_state state;
pe_action_t *action;
} pe_action_wrapper_t;
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
#include <crm/pengine/pe_types_compat.h>
#endif
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_PENGINE_PE_TYPES__H
diff --git a/include/crm/pengine/pe_types_compat.h b/include/crm/pengine/pe_types_compat.h
index 8164fc7023..5f3f035f93 100644
--- a/include/crm/pengine/pe_types_compat.h
+++ b/include/crm/pengine/pe_types_compat.h
@@ -1,71 +1,74 @@
/*
- * Copyright 2004-2022 the Pacemaker project contributors
+ * Copyright 2004-2023 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_PENGINE_PE_TYPES_COMPAT__H
# define PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H
#include <crm/pengine/pe_types.h>
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief Deprecated Pacemaker scheduler API
* \ingroup pengine
* \deprecated Do not include this header directly. The scheduler APIs in this
* header, and the header itself, will be removed in a future
* release.
*/
+//! \deprecated Use pcmk_sched_quorate instead
+#define pe_flag_have_quorum pcmk_sched_quorate
+
//!@{
//! \deprecated Do not use (unused by Pacemaker)
enum pe_graph_flags {
pe_graph_none = 0x00000,
pe_graph_updated_first = 0x00001,
pe_graph_updated_then = 0x00002,
pe_graph_disable = 0x00004,
};
//!@}
//!@{
//! \deprecated Do not use
enum pe_check_parameters {
pe_check_last_failure,
pe_check_active,
};
//!@}
//!< \deprecated Use pe_action_t instead
typedef struct pe_action_s action_t;
//!< \deprecated Use pe_action_wrapper_t instead
typedef struct pe_action_wrapper_s action_wrapper_t;
//!< \deprecated Use pe_node_t instead
typedef struct pe_node_s node_t;
//!< \deprecated Use enum pe_quorum_policy instead
typedef enum pe_quorum_policy no_quorum_policy_t;
//!< \deprecated use pe_resource_t instead
typedef struct pe_resource_s resource_t;
//!< \deprecated Use pe_tag_t instead
typedef struct pe_tag_s tag_t;
//!< \deprecated Use pe_ticket_t instead
typedef struct pe_ticket_s ticket_t;
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H
diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c
index 0a7616e9f5..0d13f3066f 100644
--- a/lib/pacemaker/pcmk_graph_producer.c
+++ b/lib/pacemaker/pcmk_graph_producer.c
@@ -1,1094 +1,1094 @@
/*
* Copyright 2004-2023 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 <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <glib.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), pe_action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pe_action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pe_action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->details->uname)
/*!
* \internal
* \brief Add an XML node tag for a specified ID
*
* \param[in] id Node UUID to add
* \param[in,out] xml Parent XML tag to add to
*/
static xmlNode*
add_node_to_xml_by_id(const char *id, xmlNode *xml)
{
xmlNode *node_xml;
node_xml = create_xml_node(xml, XML_CIB_TAG_NODE);
crm_xml_add(node_xml, XML_ATTR_ID, id);
return node_xml;
}
/*!
* \internal
* \brief Add an XML node tag for a specified node
*
* \param[in] node Node to add
* \param[in,out] xml XML to add node to
*/
static void
add_node_to_xml(const pe_node_t *node, void *xml)
{
add_node_to_xml_by_id(node->details->id, (xmlNode *) xml);
}
/*!
* \internal
* \brief Count (optionally add to XML) nodes needing maintenance state update
*
* \param[in,out] xml Parent XML tag to add to, if any
* \param[in] data_set Working set for cluster
*
* \return Count of nodes added
* \note Only Pacemaker Remote nodes are considered currently
*/
static int
add_maintenance_nodes(xmlNode *xml, const pe_working_set_t *data_set)
{
xmlNode *maintenance = NULL;
int count = 0;
if (xml != NULL) {
maintenance = create_xml_node(xml, XML_GRAPH_TAG_MAINTENANCE);
}
for (const GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
const pe_node_t *node = iter->data;
if (pe__is_guest_or_remote_node(node) &&
(node->details->maintenance != node->details->remote_maintenance)) {
if (maintenance != NULL) {
crm_xml_add(add_node_to_xml_by_id(node->details->id,
maintenance),
XML_NODE_IS_MAINTENANCE,
(node->details->maintenance? "1" : "0"));
}
count++;
}
}
crm_trace("%s %d nodes in need of maintenance mode update in state",
((maintenance == NULL)? "Counted" : "Added"), count);
return count;
}
/*!
* \internal
* \brief Add pseudo action with nodes needing maintenance state update
*
* \param[in,out] data_set Working set for cluster
*/
static void
add_maintenance_update(pe_working_set_t *data_set)
{
pe_action_t *action = NULL;
if (add_maintenance_nodes(NULL, data_set) != 0) {
action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, data_set);
pe__set_action_flags(action, pe_action_print_always);
}
}
/*!
* \internal
* \brief Add XML with nodes that an action is expected to bring down
*
* If a specified action is expected to bring any nodes down, add an XML block
* with their UUIDs. When a node is lost, this allows the controller to
* determine whether it was expected.
*
* \param[in,out] xml Parent XML tag to add to
* \param[in] action Action to check for downed nodes
*/
static void
add_downed_nodes(xmlNode *xml, const pe_action_t *action)
{
CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL),
return);
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) {
/* Shutdown makes the action's node down */
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
/* Fencing makes the action's node and any hosted guest nodes down */
const char *fence = g_hash_table_lookup(action->meta, "stonith_action");
if (pcmk__is_fencing_action(fence)) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
pe_foreach_guest_node(action->node->details->data_set, action->node,
add_node_to_xml, downed);
}
} else if (action->rsc && action->rsc->is_remote_node
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
/* Stopping a remote connection resource makes connected node down,
* unless it's part of a migration
*/
GList *iter;
pe_action_t *input;
bool migrating = false;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((pe_action_wrapper_t *) iter->data)->action;
if ((input->rsc != NULL)
&& pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none)
&& pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
migrating = true;
break;
}
}
if (!migrating) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->rsc->id, downed);
}
}
}
/*!
* \internal
* \brief Create a transition graph operation key for a clone action
*
* \param[in] action Clone action
* \param[in] interval_ms Action interval in milliseconds
*
* \return Newly allocated string with transition graph operation key
*/
static char *
clone_op_key(const pe_action_t *action, guint interval_ms)
{
if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = g_hash_table_lookup(action->meta, "notify_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_operation");
CRM_LOG_ASSERT((n_type != NULL) && (n_task != NULL));
return pcmk__notify_key(action->rsc->clone_name, n_type, n_task);
} else if (action->cancel_task != NULL) {
return pcmk__op_key(action->rsc->clone_name, action->cancel_task,
interval_ms);
} else {
return pcmk__op_key(action->rsc->clone_name, action->task, interval_ms);
}
}
/*!
* \internal
* \brief Add node details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] xml Transition graph action XML for \p action
*/
static void
add_node_details(const pe_action_t *action, xmlNode *xml)
{
pe_node_t *router_node = pcmk__connection_host_for_action(action);
crm_xml_add(xml, XML_LRM_ATTR_TARGET, action->node->details->uname);
crm_xml_add(xml, XML_LRM_ATTR_TARGET_UUID, action->node->details->id);
if (router_node != NULL) {
crm_xml_add(xml, XML_LRM_ATTR_ROUTER_NODE, router_node->details->uname);
}
}
/*!
* \internal
* \brief Add resource details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_resource_details(const pe_action_t *action, xmlNode *action_xml)
{
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER,
XML_ATTR_TYPE
};
/* If a resource is locked to a node via shutdown-lock, mark its actions
* so the controller can preserve the lock when the action completes.
*/
if (pcmk__action_locks_rsc_to_node(action)) {
crm_xml_add_ll(action_xml, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
(long long) action->rsc->lock_time);
}
// List affected resource
rsc_xml = create_xml_node(action_xml,
(const char *) action->rsc->xml->name);
if (pcmk_is_set(action->rsc->flags, pe_rsc_orphan)
&& (action->rsc->clone_name != NULL)) {
/* Use the numbered instance name here, because if there is more
* than one instance on a node, we need to make sure the command
* goes to the right one.
*
* This is important even for anonymous clones, because the clone's
* unique meta-attribute might have just been toggled from on to
* off.
*/
crm_debug("Using orphan clone name %s instead of %s",
action->rsc->id, action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
} else if (!pcmk_is_set(action->rsc->flags, pe_rsc_unique)) {
const char *xml_id = ID(action->rsc->xml);
crm_debug("Using anonymous clone name %s for %s (aka %s)",
xml_id, action->rsc->id, action->rsc->clone_name);
/* ID is what we'd like client to use
* ID_LONG is what they might know it as instead
*
* ID_LONG is only strictly needed /here/ during the
* transition period until all nodes in the cluster
* are running the new software /and/ have rebooted
* once (meaning that they've only ever spoken to a DC
* supporting this feature).
*
* If anyone toggles the unique flag to 'on', the
* 'instance free' name will correspond to an orphan
* and fall into the clause above instead
*/
crm_xml_add(rsc_xml, XML_ATTR_ID, xml_id);
if ((action->rsc->clone_name != NULL)
&& !pcmk__str_eq(xml_id, action->rsc->clone_name,
pcmk__str_none)) {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->clone_name);
} else {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
}
} else {
CRM_ASSERT(action->rsc->clone_name == NULL);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->id);
}
for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) {
crm_xml_add(rsc_xml, attr_list[lpc],
g_hash_table_lookup(action->rsc->meta, attr_list[lpc]));
}
}
/*!
* \internal
* \brief Add action attributes to transition graph action XML
*
* \param[in,out] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_action_attributes(pe_action_t *action, xmlNode *action_xml)
{
xmlNode *args_xml = NULL;
/* We create free-standing XML to start, so we can sort the attributes
* before adding it to action_xml, which keeps the scheduler regression
* test graphs comparable.
*/
args_xml = create_xml_node(NULL, XML_TAG_ATTRS);
crm_xml_add(args_xml, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
g_hash_table_foreach(action->extra, hash2field, args_xml);
if ((action->rsc != NULL) && (action->node != NULL)) {
// Get the resource instance attributes, evaluated properly for node
GHashTable *params = pe_rsc_params(action->rsc, action->node,
action->rsc->cluster);
pcmk__substitute_remote_addr(action->rsc, params);
g_hash_table_foreach(params, hash2smartfield, args_xml);
} else if ((action->rsc != NULL)
&& (action->rsc->variant <= pcmk_rsc_variant_primitive)) {
GHashTable *params = pe_rsc_params(action->rsc, NULL,
action->rsc->cluster);
g_hash_table_foreach(params, hash2smartfield, args_xml);
}
g_hash_table_foreach(action->meta, hash2metafield, args_xml);
if (action->rsc != NULL) {
pe_resource_t *parent = action->rsc;
while (parent != NULL) {
parent->cmds->add_graph_meta(parent, args_xml);
parent = parent->parent;
}
pcmk__add_bundle_meta_to_xml(args_xml, action);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)
&& (action->node != NULL)) {
/* Pass the node's attributes as meta-attributes.
*
* @TODO: Determine whether it is still necessary to do this. It was
* added in 33d99707, probably for the libfence-based implementation in
* c9a90bd, which is no longer used.
*/
g_hash_table_foreach(action->node->details->attrs, hash2metafield,
args_xml);
}
sorted_xml(args_xml, action_xml, FALSE);
free_xml(args_xml);
}
/*!
* \internal
* \brief Create the transition graph XML for a scheduled action
*
* \param[in,out] parent Parent XML element to add action to
* \param[in,out] action Scheduled action
* \param[in] skip_details If false, add action details as sub-elements
* \param[in] data_set Cluster working set
*/
static void
create_graph_action(xmlNode *parent, pe_action_t *action, bool skip_details,
const pe_working_set_t *data_set)
{
bool needs_node_info = true;
bool needs_maintenance_info = false;
xmlNode *action_xml = NULL;
if ((action == NULL) || (data_set == NULL)) {
return;
}
// Create the top-level element based on task
if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) {
/* All fences need node info; guest node fences are pseudo-events */
if (pcmk_is_set(action->flags, pe_action_pseudo)) {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT);
} else {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
}
} else if (pcmk__str_any_of(action->task,
PCMK_ACTION_DO_SHUTDOWN,
PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE,
pcmk__str_none)) {
// CIB-only clean-up for shutdown locks
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
crm_xml_add(action_xml, PCMK__XA_MODE, XML_TAG_CIB);
} else if (pcmk_is_set(action->flags, pe_action_pseudo)) {
if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES,
pcmk__str_none)) {
needs_maintenance_info = true;
}
action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT);
needs_node_info = false;
} else {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_RSC_OP);
}
crm_xml_add_int(action_xml, XML_ATTR_ID, action->id);
crm_xml_add(action_xml, XML_LRM_ATTR_TASK, action->task);
if ((action->rsc != NULL) && (action->rsc->clone_name != NULL)) {
char *clone_key = NULL;
guint interval_ms;
if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
clone_key = clone_op_key(action, interval_ms);
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, clone_key);
crm_xml_add(action_xml, "internal_" XML_LRM_ATTR_TASK_KEY,
action->uuid);
free(clone_key);
} else {
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, action->uuid);
}
if (needs_node_info && (action->node != NULL)) {
add_node_details(action, action_xml);
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET),
strdup(action->node->details->uname));
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET_UUID),
strdup(action->node->details->id));
}
if (skip_details) {
return;
}
if ((action->rsc != NULL)
&& !pcmk_is_set(action->flags, pe_action_pseudo)) {
// This is a real resource action, so add resource details
add_resource_details(action, action_xml);
}
/* List any attributes in effect */
add_action_attributes(action, action_xml);
/* List any nodes this action is expected to make down */
if (needs_node_info && (action->node != NULL)) {
add_downed_nodes(action_xml, action);
}
if (needs_maintenance_info) {
add_maintenance_nodes(action_xml, data_set);
}
}
/*!
* \internal
* \brief Check whether an action should be added to the transition graph
*
* \param[in] action Action to check
*
* \return true if action should be added to graph, otherwise false
*/
static bool
should_add_action_to_graph(const pe_action_t *action)
{
if (!pcmk_is_set(action->flags, pe_action_runnable)) {
crm_trace("Ignoring action %s (%d): unrunnable",
action->uuid, action->id);
return false;
}
if (pcmk_is_set(action->flags, pe_action_optional)
&& !pcmk_is_set(action->flags, pe_action_print_always)) {
crm_trace("Ignoring action %s (%d): optional",
action->uuid, action->id);
return false;
}
/* Actions for unmanaged resources should be excluded from the graph,
* with the exception of monitors and cancellation of recurring monitors.
*/
if ((action->rsc != NULL)
&& !pcmk_is_set(action->rsc->flags, pe_rsc_managed)
&& !pcmk__str_eq(action->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
const char *interval_ms_s;
/* A cancellation of a recurring monitor will get here because the task
* is cancel rather than monitor, but the interval can still be used to
* recognize it. The interval has been normalized to milliseconds by
* this point, so a string comparison is sufficient.
*/
interval_ms_s = g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) {
crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)",
action->uuid, action->id, action->rsc->id);
return false;
}
}
/* Always add pseudo-actions, fence actions, and shutdown actions (already
* determined to be required and runnable by this point)
*/
if (pcmk_is_set(action->flags, pe_action_pseudo)
|| pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH,
PCMK_ACTION_DO_SHUTDOWN, NULL)) {
return true;
}
if (action->node == NULL) {
pe_err("Skipping action %s (%d) "
"because it was not assigned to a node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unassigned", action, false);
return false;
}
if (pcmk_is_set(action->flags, pe_action_dc)) {
crm_trace("Action %s (%d) should be dumped: "
"can run on DC instead of %s",
action->uuid, action->id, pe__node_name(action->node));
} else if (pe__is_guest_node(action->node)
&& !action->node->details->remote_requires_reset) {
crm_trace("Action %s (%d) should be dumped: "
"assuming will be runnable on guest %s",
action->uuid, action->id, pe__node_name(action->node));
} else if (!action->node->details->online) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for offline node (bug?)",
action->uuid, action->id);
pcmk__log_action("Offline node", action, false);
return false;
} else if (action->node->details->unclean) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for unclean node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unclean node", action, false);
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether an ordering's flags can change an action
*
* \param[in] ordering Ordering to check
*
* \return true if ordering has flags that can change an action, false otherwise
*/
static bool
ordering_can_change_actions(const pe_action_wrapper_t *ordering)
{
return pcmk_any_flags_set(ordering->type, ~(pe_order_implies_first_printed
|pe_order_implies_then_printed
|pe_order_optional));
}
/*!
* \internal
* \brief Check whether an action input should be in the transition graph
*
* \param[in] action Action to check
* \param[in,out] input Action input to check
*
* \return true if input should be in graph, false otherwise
* \note This function may not only check an input, but disable it under certian
* circumstances (load or anti-colocation orderings that are not needed).
*/
static bool
should_add_input_to_graph(const pe_action_t *action, pe_action_wrapper_t *input)
{
if (input->state == pe_link_dumped) {
return true;
}
if (input->type == pe_order_none) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering disabled",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, pe_action_runnable)
&& !ordering_can_change_actions(input)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, pe_action_runnable)
&& pcmk_is_set(input->type, pe_order_one_or_more)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"one-or-more and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_implies_first_migratable)
&& !pcmk_is_set(input->action->flags, pe_action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"implies input migratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_apply_first_non_migratable)
&& pcmk_is_set(input->action->flags,
pe_action_migrate_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"only if input unmigratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->type == pe_order_optional)
&& pcmk_is_set(input->action->flags, pe_action_migrate_runnable)
&& pcmk__ends_with(input->action->uuid, "_stop_0")) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional but stop in migration",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (input->type == pe_order_load) {
pe_node_t *input_node = input->action->node;
// load orderings are relevant only if actions are for same node
if ((action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
pe_node_t *assigned = action->rsc->allocated_to;
/* For load_stopped -> migrate_to orderings, we care about where it
* has been assigned to, not where it will be executed.
*/
if (!pe__same_node(input_node, assigned)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(assigned? assigned->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
}
} else if (!pe__same_node(input_node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->type == pe_order_anti_colocation) {
if (input->action->node && action->node
&& !pe__same_node(input->action->node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
pe__node_name(action->node),
pe__node_name(input->action->node));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& pcmk_is_set(input->action->rsc->flags, pe_rsc_failed)
&& !pcmk_is_set(input->action->rsc->flags, pe_rsc_managed)
&& pcmk__ends_with(input->action->uuid, "_stop_0")
&& action->rsc && pe_rsc_is_clone(action->rsc)) {
crm_warn("Ignoring requirement that %s complete before %s:"
" unmanaged failed resources cannot prevent clone shutdown",
input->action->uuid, action->uuid);
return false;
} else if (pcmk_is_set(input->action->flags, pe_action_optional)
&& !pcmk_any_flags_set(input->action->flags,
pe_action_print_always|pe_action_dumped)
&& !should_add_action_to_graph(input->action)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
}
crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x",
action->uuid, action->id, action_type_str(input->action->flags),
input->action->uuid, input->action->id,
action_node_str(input->action),
action_runnable_str(input->action->flags),
action_optional_str(input->action->flags), input->type);
return true;
}
/*!
* \internal
* \brief Check whether an ordering creates an ordering loop
*
* \param[in] init_action "First" action in ordering
* \param[in] action Callers should always set this the same as
* \p init_action (this function may use a different
* value for recursive calls)
* \param[in,out] input Action wrapper for "then" action in ordering
*
* \return true if the ordering creates a loop, otherwise false
*/
bool
pcmk__graph_has_loop(const pe_action_t *init_action, const pe_action_t *action,
pe_action_wrapper_t *input)
{
bool has_loop = false;
if (pcmk_is_set(input->action->flags, pe_action_tracking)) {
crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
return false;
}
// Don't need to check inputs that won't be used
if (!should_add_input_to_graph(action, input)) {
return false;
}
if (input->action == init_action) {
crm_debug("Input loop found in %s@%s ->...-> %s@%s",
action->uuid,
action->node? action->node->details->uname : "",
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
return true;
}
pe__set_action_flags(input->action, pe_action_tracking);
crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)"
"for graph loop with %s@%s ",
action->uuid,
action->node? action->node->details->uname : "",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
input->type,
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
// Recursively check input itself for loops
for (GList *iter = input->action->actions_before;
iter != NULL; iter = iter->next) {
if (pcmk__graph_has_loop(init_action, input->action,
(pe_action_wrapper_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
break;
}
}
pe__clear_action_flags(input->action, pe_action_tracking);
if (!has_loop) {
crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
}
return has_loop;
}
/*!
* \internal
* \brief Create a synapse XML element for a transition graph
*
* \param[in] action Action that synapse is for
* \param[in,out] data_set Cluster working set containing graph
*
* \return Newly added XML element for new graph synapse
*/
static xmlNode *
create_graph_synapse(const pe_action_t *action, pe_working_set_t *data_set)
{
int synapse_priority = 0;
xmlNode *syn = create_xml_node(data_set->graph, "synapse");
crm_xml_add_int(syn, XML_ATTR_ID, data_set->num_synapse);
data_set->num_synapse++;
if (action->rsc != NULL) {
synapse_priority = action->rsc->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, XML_CIB_ATTR_PRIORITY, synapse_priority);
}
return syn;
}
/*!
* \internal
* \brief Add an action to the transition graph XML if appropriate
*
* \param[in,out] data Action to possibly add
* \param[in,out] user_data Cluster working set
*
* \note This will de-duplicate the action inputs, meaning that the
* pe_action_wrapper_t:type flags can no longer be relied on to retain
* their original settings. That means this MUST be called after
* pcmk__apply_orderings() is complete, and nothing after this should rely
* on those type flags. (For example, some code looks for type equal to
* some flag rather than whether the flag is set, and some code looks for
* particular combinations of flags -- such code must be done before
* pcmk__create_graph().)
*/
static void
add_action_to_graph(gpointer data, gpointer user_data)
{
pe_action_t *action = (pe_action_t *) data;
pe_working_set_t *data_set = (pe_working_set_t *) user_data;
xmlNode *syn = NULL;
xmlNode *set = NULL;
xmlNode *in = NULL;
/* If we haven't already, de-duplicate inputs (even if we won't be adding
* the action to the graph, so that crm_simulate's dot graphs don't have
* duplicates).
*/
if (!pcmk_is_set(action->flags, pe_action_dedup)) {
pcmk__deduplicate_action_inputs(action);
pe__set_action_flags(action, pe_action_dedup);
}
if (pcmk_is_set(action->flags, pe_action_dumped) // Already added, or
|| !should_add_action_to_graph(action)) { // shouldn't be added
return;
}
pe__set_action_flags(action, pe_action_dumped);
crm_trace("Adding action %d (%s%s%s) to graph",
action->id, action->uuid,
((action->node == NULL)? "" : " on "),
((action->node == NULL)? "" : action->node->details->uname));
syn = create_graph_synapse(action, data_set);
set = create_xml_node(syn, "action_set");
in = create_xml_node(syn, "inputs");
create_graph_action(set, action, false, data_set);
for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) lpc->data;
if (should_add_input_to_graph(action, input)) {
xmlNode *input_xml = create_xml_node(in, "trigger");
input->state = pe_link_dumped;
create_graph_action(input_xml, input->action, true, data_set);
}
}
}
static int transition_id = -1;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const char *filename)
{
if (was_processing_error) {
crm_err("Calculated transition %d (with errors)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else if (was_processing_warning) {
crm_warn("Calculated transition %d (with warnings)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else {
crm_notice("Calculated transition %d%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
}
if (crm_config_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*!
* \internal
* \brief Add a resource's actions to the transition graph
*
* \param[in,out] rsc Resource whose actions should be added
*/
void
pcmk__add_rsc_actions_to_graph(pe_resource_t *rsc)
{
GList *iter = NULL;
CRM_ASSERT(rsc != NULL);
pe_rsc_trace(rsc, "Adding actions for %s to graph", rsc->id);
// First add the resource's own actions
g_list_foreach(rsc->actions, add_action_to_graph, rsc->cluster);
// Then recursively add its children's actions (appropriate to variant)
for (iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) iter->data;
child_rsc->cmds->add_actions_to_graph(child_rsc);
}
}
/*!
* \internal
* \brief Create a transition graph with all cluster actions needed
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__create_graph(pe_working_set_t *data_set)
{
GList *iter = NULL;
const char *value = NULL;
long long limit = 0LL;
transition_id++;
crm_trace("Creating transition graph %d", transition_id);
data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
value = pe_pref(data_set->config_hash, "cluster-delay");
crm_xml_add(data_set->graph, "cluster-delay", value);
value = pe_pref(data_set->config_hash, "stonith-timeout");
crm_xml_add(data_set->graph, "stonith-timeout", value);
crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY");
if (pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)) {
crm_xml_add(data_set->graph, "failed-start-offset", "INFINITY");
} else {
crm_xml_add(data_set->graph, "failed-start-offset", "1");
}
value = pe_pref(data_set->config_hash, "batch-limit");
crm_xml_add(data_set->graph, "batch-limit", value);
crm_xml_add_int(data_set->graph, "transition_id", transition_id);
value = pe_pref(data_set->config_hash, "migration-limit");
if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) {
crm_xml_add(data_set->graph, "migration-limit", value);
}
if (data_set->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu",
(long long) data_set->recheck_by);
crm_xml_add(data_set->graph, "recheck-by", recheck_epoch);
free(recheck_epoch);
}
/* The following code will de-duplicate action inputs, so nothing past this
* should rely on the action input type flags retaining their original
* values.
*/
// Add resource actions to graph
for (iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
pe_rsc_trace(rsc, "Processing actions for %s", rsc->id);
rsc->cmds->add_actions_to_graph(rsc);
}
// Add pseudo-action for list of nodes with maintenance state update
add_maintenance_update(data_set);
// Add non-resource (node) actions
for (iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = (pe_action_t *) iter->data;
if ((action->rsc != NULL)
&& (action->node != NULL)
&& action->node->details->shutdown
&& !pcmk_is_set(action->rsc->flags, pe_rsc_maintenance)
&& !pcmk_any_flags_set(action->flags,
pe_action_optional|pe_action_runnable)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* Eventually we should just ignore the 'fence' case, but for now
* it's the best way to detect (in CTS) when CIB resource updates
* are being lost.
*/
- if (pcmk_is_set(data_set->flags, pe_flag_have_quorum)
+ if (pcmk_is_set(data_set->flags, pcmk_sched_quorate)
|| (data_set->no_quorum_policy == pcmk_no_quorum_ignore)) {
const bool managed = pcmk_is_set(action->rsc->flags,
pe_rsc_managed);
const bool failed = pcmk_is_set(action->rsc->flags,
pe_rsc_failed);
crm_crit("Cannot %s %s because of %s:%s%s (%s)",
action->node->details->unclean? "fence" : "shut down",
pe__node_name(action->node), action->rsc->id,
(managed? " blocked" : " unmanaged"),
(failed? " failed" : ""), action->uuid);
}
}
add_action_to_graph((gpointer) action, (gpointer) data_set);
}
crm_log_xml_trace(data_set->graph, "graph");
}
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index 2c3f5f9fb4..9a6cd90520 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1650 +1,1650 @@
/*
* Copyright 2004-2023 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(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void promote_resource(pe_resource_t *rsc, pe_node_t *node,
bool optional);
static void assert_role_error(pe_resource_t *rsc, pe_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)(pe_resource_t *rsc, pe_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 pe_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(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail)
{
GList *nodes = NULL;
pe_node_t *chosen = NULL;
pe_node_t *best = NULL;
const pe_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) {
// 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 = (pe_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) {
pe_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) {
pe_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)) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pe__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pe_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.
*/
pe_node_t *running = pe__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pe_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) {
pe_node_t *allowed = (pe_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));
}
}
}
pe_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, pe_resource_t *rsc)
{
GHashTable *archive = NULL;
pe_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, pe_rsc_provisional)) {
pe_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)) {
pe_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 pe_resource_t *connection)
{
pe_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.
*/
pe_node_t *
pcmk__primitive_assign(pe_resource_t *rsc, const pe_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, pe_rsc_allocating)) {
pe_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, pe_rsc_provisional)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->allocated_to != NULL) {
node_name = pe__node_name(rsc->allocated_to);
}
pe_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, pe_rsc_allocating)) {
pe_rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pe__set_resource_flags(rsc, pe_rsc_allocating);
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) {
pe_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, pe_flag_have_quorum)
+ && !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 "
"no-quorum-policy=freeze",
rsc->id, role2text(rsc->role), role2text(rsc->next_role));
pe__set_next_role(rsc, rsc->role, "no-quorum-policy=freeze");
}
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_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, pe_flag_stonith_enabled)
&& !pcmk_is_set(rsc->cluster->flags, pe_flag_have_stonith_resource)) {
pe__clear_resource_flags(rsc, pe_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pe_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, pe_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pe_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, pe_flag_stop_everything)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
pe_rsc_debug(rsc, "Forcing %s to stop: 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, pe_rsc_orphan)) {
pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->running_on != NULL) && stop_if_fail) {
pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
}
pe__clear_resource_flags(rsc, pe_rsc_allocating);
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(pe_resource_t *rsc, pe_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, pe_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];
pe_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, pe_rsc_block)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pe_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, pe_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(pe_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(pe_resource_t *rsc)
{
pe_action_t *start = NULL;
pe_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, pe_action_print_always);
}
/*!
* \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(pe_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;
pe_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(pe_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pe_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);
pe_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)) {
pe_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.
*/
pe_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, pe_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
* 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
pe_proc_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, pe_rsc_stop_unexpected);
break;
default:
break;
}
} else {
pe__clear_resource_flags(rsc, pe_rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pe_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, pe_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pe_rsc_stop)) {
need_stop = true;
pe_rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pe_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, pe_rsc_block)) {
pe_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)) {
pe_action_t *start = NULL;
pe_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, pe_action_optional)) {
// Recovery of a promoted resource
pe_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 pe_resource_t *rsc)
{
GHashTableIter iter;
pe_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 pe_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(pe_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, pe_rsc_managed)) {
pe_rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource %s",
rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pe_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags,
pe_flag_enable_unfencing)
&& pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& !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,
pe_order_optional|pe_order_implies_then|pe_order_restart,
rsc->cluster);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pe_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,
pe_order_promoted_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,
pe_order_runnable_left, 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, pe_order_same_node|pe_order_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) {
pe_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, pe_rsc_allow_remote_remotes)) {
rsc_avoids_remote_nodes(rsc->container);
}
/* If someone cleans up a guest or bundle node's container, we will
* likely schedule a (re-)probe of the container and recovery of the
* connection. Order the connection stop after the container probe,
* so that if we detect the container running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pe_order_optional);
/* 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) {
pe_node_t *node = item->data;
if (node->details->remote_rsc != remote_rsc) {
node->weight = -INFINITY;
}
}
} else {
/* This resource is either a filler for a container that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its container %s",
rsc->id, rsc->container->id);
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,
pe_order_implies_then|pe_order_runnable_left,
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, pe_order_implies_first, rsc->cluster);
if (pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
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, pe_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(pe_resource_t *dependent,
const pe_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);
pe_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
* resource_alloc_functions_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pe_resource_t *rsc,
const pe_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
* resource_alloc_functions_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pe_resource_t *rsc,
const pe_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(pe_action_t *action, const pe_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 pe_resource_t *rsc, const pe_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pe_rsc_stop_unexpected|pe_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(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *current = (pe_node_t *) iter->data;
pe_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.
*/
pe_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)) {
pe_rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pe__node_name(current));
continue;
} else {
pe_rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pe__node_name(current));
optional = false;
}
}
pe_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, pe_rsc_restarting
|pe_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, pe_rsc_managed)) {
pe__clear_action_flags(stop, pe_action_runnable);
}
if (pcmk_is_set(rsc->cluster->flags, pe_flag_remove_after_stop)) {
pcmk__schedule_cleanup(rsc, current, optional);
}
if (pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing)) {
pe_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false, rsc->cluster);
order_actions(stop, unfence, pe_order_implies_first);
if (!pcmk__node_unfenced(current)) {
pe_proc_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(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
pe_action_t *start = NULL;
CRM_ASSERT(node != NULL);
pe_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, pe_order_implies_then);
if (pcmk_is_set(start->flags, pe_action_runnable) && !optional) {
pe__clear_action_flags(start, pe_action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pe_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, pe_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(pe_resource_t *rsc, pe_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) {
pe_action_t *start = (pe_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pe_action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pe_action_t *promote = promote_action(rsc, node, optional);
pe_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.
*/
pe_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, pe_action_pseudo);
}
} else {
pe_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) {
pe_action_t *promote = (pe_action_t *) iter->data;
pe__clear_action_flags(promote, pe_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(pe_resource_t *rsc, pe_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) {
pe_node_t *current = (pe_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pe_rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pe__node_name(current));
} else {
pe_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(pe_resource_t *rsc, pe_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(pe_resource_t *rsc, const pe_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? pe_order_implies_then : pe_order_optional;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pe_rsc_failed)) {
pe_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) {
pe_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 pe_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pe_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 resource_alloc_functions_t:add_utilization()
void
pcmk__primitive_add_utilization(const pe_resource_t *rsc,
const pe_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, pe_rsc_provisional)) {
return;
}
pe_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(pe_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 pe_node_t *node = (const pe_node_t *) data;
pe_resource_t *rsc = (pe_resource_t *) user_data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
XML_CONFIG_ATTR_SHUTDOWN_LOCK, rsc->cluster);
}
}
// Primitive implementation of resource_alloc_functions_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pe_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.
*/
pe_rsc_info(rsc,
"Cancelling shutdown lock because %s is already active",
rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node, rsc->cluster);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pe_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown",
rsc->id, 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;
pe_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);
} else {
pe_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/pacemaker/pcmk_sched_promotable.c b/lib/pacemaker/pcmk_sched_promotable.c
index 76bf87ad01..2da1c2686c 100644
--- a/lib/pacemaker/pcmk_sched_promotable.c
+++ b/lib/pacemaker/pcmk_sched_promotable.c
@@ -1,1299 +1,1299 @@
/*
* Copyright 2004-2023 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 <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Add implicit promotion ordering for a promotable instance
*
* \param[in,out] clone Clone resource
* \param[in,out] child Instance of \p clone being ordered
* \param[in,out] last Previous instance ordered (NULL if \p child is first)
*/
static void
order_instance_promotion(pe_resource_t *clone, pe_resource_t *child,
pe_resource_t *last)
{
// "Promote clone" -> promote instance -> "clone promoted"
pcmk__order_resource_actions(clone, PCMK_ACTION_PROMOTE,
child, PCMK_ACTION_PROMOTE,
pe_order_optional);
pcmk__order_resource_actions(child, PCMK_ACTION_PROMOTE,
clone, PCMK_ACTION_PROMOTED,
pe_order_optional);
// If clone is ordered, order this instance relative to last
if ((last != NULL) && pe__clone_is_ordered(clone)) {
pcmk__order_resource_actions(last, PCMK_ACTION_PROMOTE,
child, PCMK_ACTION_PROMOTE,
pe_order_optional);
}
}
/*!
* \internal
* \brief Add implicit demotion ordering for a promotable instance
*
* \param[in,out] clone Clone resource
* \param[in,out] child Instance of \p clone being ordered
* \param[in] last Previous instance ordered (NULL if \p child is first)
*/
static void
order_instance_demotion(pe_resource_t *clone, pe_resource_t *child,
pe_resource_t *last)
{
// "Demote clone" -> demote instance -> "clone demoted"
pcmk__order_resource_actions(clone, PCMK_ACTION_DEMOTE, child,
PCMK_ACTION_DEMOTE,
pe_order_implies_first_printed);
pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE,
clone, PCMK_ACTION_DEMOTED,
pe_order_implies_then_printed);
// If clone is ordered, order this instance relative to last
if ((last != NULL) && pe__clone_is_ordered(clone)) {
pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, last,
PCMK_ACTION_DEMOTE, pe_order_optional);
}
}
/*!
* \internal
* \brief Check whether an instance will be promoted or demoted
*
* \param[in] rsc Instance to check
* \param[out] demoting If \p rsc will be demoted, this will be set to true
* \param[out] promoting If \p rsc will be promoted, this will be set to true
*/
static void
check_for_role_change(const pe_resource_t *rsc, bool *demoting, bool *promoting)
{
const GList *iter = NULL;
// If this is a cloned group, check group members recursively
if (rsc->children != NULL) {
for (iter = rsc->children; iter != NULL; iter = iter->next) {
check_for_role_change((const pe_resource_t *) iter->data,
demoting, promoting);
}
return;
}
for (iter = rsc->actions; iter != NULL; iter = iter->next) {
const pe_action_t *action = (const pe_action_t *) iter->data;
if (*promoting && *demoting) {
return;
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
continue;
} else if (pcmk__str_eq(PCMK_ACTION_DEMOTE, action->task,
pcmk__str_none)) {
*demoting = true;
} else if (pcmk__str_eq(PCMK_ACTION_PROMOTE, action->task,
pcmk__str_none)) {
*promoting = true;
}
}
}
/*!
* \internal
* \brief Add promoted-role location constraint scores to an instance's priority
*
* Adjust a promotable clone instance's promotion priority by the scores of any
* location constraints in a list that are both limited to the promoted role and
* for the node where the instance will be placed.
*
* \param[in,out] child Promotable clone instance
* \param[in] location_constraints List of location constraints to apply
* \param[in] chosen Node where \p child will be placed
*/
static void
apply_promoted_locations(pe_resource_t *child,
const GList *location_constraints,
const pe_node_t *chosen)
{
for (const GList *iter = location_constraints; iter; iter = iter->next) {
const pe__location_t *location = iter->data;
const pe_node_t *constraint_node = NULL;
if (location->role_filter == pcmk_role_promoted) {
constraint_node = pe_find_node_id(location->node_list_rh,
chosen->details->id);
}
if (constraint_node != NULL) {
int new_priority = pcmk__add_scores(child->priority,
constraint_node->weight);
pe_rsc_trace(child,
"Applying location %s to %s promotion priority on %s: "
"%s + %s = %s",
location->id, child->id,
pe__node_name(constraint_node),
pcmk_readable_score(child->priority),
pcmk_readable_score(constraint_node->weight),
pcmk_readable_score(new_priority));
child->priority = new_priority;
}
}
}
/*!
* \internal
* \brief Get the node that an instance will be promoted on
*
* \param[in] rsc Promotable clone instance to check
*
* \return Node that \p rsc will be promoted on, or NULL if none
*/
static pe_node_t *
node_to_be_promoted_on(const pe_resource_t *rsc)
{
pe_node_t *node = NULL;
pe_node_t *local_node = NULL;
const pe_resource_t *parent = NULL;
// If this is a cloned group, bail if any group member can't be promoted
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child = (pe_resource_t *) iter->data;
if (node_to_be_promoted_on(child) == NULL) {
pe_rsc_trace(rsc,
"%s can't be promoted because member %s can't",
rsc->id, child->id);
return NULL;
}
}
node = rsc->fns->location(rsc, NULL, FALSE);
if (node == NULL) {
pe_rsc_trace(rsc, "%s can't be promoted because it won't be active",
rsc->id);
return NULL;
} else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) {
if (rsc->fns->state(rsc, TRUE) == pcmk_role_promoted) {
crm_notice("Unmanaged instance %s will be left promoted on %s",
rsc->id, pe__node_name(node));
} else {
pe_rsc_trace(rsc, "%s can't be promoted because it is unmanaged",
rsc->id);
return NULL;
}
} else if (rsc->priority < 0) {
pe_rsc_trace(rsc,
"%s can't be promoted because its promotion priority %d "
"is negative",
rsc->id, rsc->priority);
return NULL;
} else if (!pcmk__node_available(node, false, true)) {
pe_rsc_trace(rsc, "%s can't be promoted because %s can't run resources",
rsc->id, pe__node_name(node));
return NULL;
}
parent = pe__const_top_resource(rsc, false);
local_node = g_hash_table_lookup(parent->allowed_nodes, node->details->id);
if (local_node == NULL) {
/* It should not be possible for the scheduler to have assigned the
* instance to a node where its parent is not allowed, but it's good to
* have a fail-safe.
*/
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
crm_warn("%s can't be promoted because %s is not allowed on %s "
"(scheduler bug?)",
rsc->id, parent->id, pe__node_name(node));
} // else the instance is unmanaged and already promoted
return NULL;
} else if ((local_node->count >= pe__clone_promoted_node_max(parent))
&& pcmk_is_set(rsc->flags, pe_rsc_managed)) {
pe_rsc_trace(rsc,
"%s can't be promoted because %s has "
"maximum promoted instances already",
rsc->id, pe__node_name(node));
return NULL;
}
return local_node;
}
/*!
* \internal
* \brief Compare two promotable clone instances by promotion priority
*
* \param[in] a First instance to compare
* \param[in] b Second instance to compare
*
* \return A negative number if \p a has higher promotion priority,
* a positive number if \p b has higher promotion priority,
* or 0 if promotion priorities are equal
*/
static gint
cmp_promotable_instance(gconstpointer a, gconstpointer b)
{
const pe_resource_t *rsc1 = (const pe_resource_t *) a;
const pe_resource_t *rsc2 = (const pe_resource_t *) b;
enum rsc_role_e role1 = pcmk_role_unknown;
enum rsc_role_e role2 = pcmk_role_unknown;
CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL));
// Check sort index set by pcmk__set_instance_roles()
if (rsc1->sort_index > rsc2->sort_index) {
pe_rsc_trace(rsc1,
"%s has higher promotion priority than %s "
"(sort index %d > %d)",
rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index);
return -1;
} else if (rsc1->sort_index < rsc2->sort_index) {
pe_rsc_trace(rsc1,
"%s has lower promotion priority than %s "
"(sort index %d < %d)",
rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index);
return 1;
}
// If those are the same, prefer instance whose current role is higher
role1 = rsc1->fns->state(rsc1, TRUE);
role2 = rsc2->fns->state(rsc2, TRUE);
if (role1 > role2) {
pe_rsc_trace(rsc1,
"%s has higher promotion priority than %s "
"(higher current role)",
rsc1->id, rsc2->id);
return -1;
} else if (role1 < role2) {
pe_rsc_trace(rsc1,
"%s has lower promotion priority than %s "
"(lower current role)",
rsc1->id, rsc2->id);
return 1;
}
// Finally, do normal clone instance sorting
return pcmk__cmp_instance(a, b);
}
/*!
* \internal
* \brief Add a promotable clone instance's sort index to its node's score
*
* Add a promotable clone instance's sort index (which sums its promotion
* preferences and scores of relevant location constraints for the promoted
* role) to the node score of the instance's assigned node.
*
* \param[in] data Promotable clone instance
* \param[in,out] user_data Clone parent of \p data
*/
static void
add_sort_index_to_node_score(gpointer data, gpointer user_data)
{
const pe_resource_t *child = (const pe_resource_t *) data;
pe_resource_t *clone = (pe_resource_t *) user_data;
pe_node_t *node = NULL;
const pe_node_t *chosen = NULL;
if (child->sort_index < 0) {
pe_rsc_trace(clone, "Not adding sort index of %s: negative", child->id);
return;
}
chosen = child->fns->location(child, NULL, FALSE);
if (chosen == NULL) {
pe_rsc_trace(clone, "Not adding sort index of %s: inactive", child->id);
return;
}
node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id);
CRM_ASSERT(node != NULL);
node->weight = pcmk__add_scores(child->sort_index, node->weight);
pe_rsc_trace(clone,
"Added cumulative priority of %s (%s) to score on %s (now %s)",
child->id, pcmk_readable_score(child->sort_index),
pe__node_name(node), pcmk_readable_score(node->weight));
}
/*!
* \internal
* \brief Apply colocation to dependent's node scores if for promoted role
*
* \param[in,out] data Colocation constraint to apply
* \param[in,out] user_data Promotable clone that is constraint's dependent
*/
static void
apply_coloc_to_dependent(gpointer data, gpointer user_data)
{
pcmk__colocation_t *colocation = data;
pe_resource_t *clone = user_data;
pe_resource_t *primary = colocation->primary;
uint32_t flags = pcmk__coloc_select_default;
float factor = colocation->score / (float) INFINITY;
if (colocation->dependent_role != pcmk_role_promoted) {
return;
}
if (colocation->score < INFINITY) {
flags = pcmk__coloc_select_active;
}
pe_rsc_trace(clone, "Applying colocation %s (promoted %s with %s) @%s",
colocation->id, colocation->dependent->id,
colocation->primary->id,
pcmk_readable_score(colocation->score));
primary->cmds->add_colocated_node_scores(primary, clone, clone->id,
&clone->allowed_nodes, colocation,
factor, flags);
}
/*!
* \internal
* \brief Apply colocation to primary's node scores if for promoted role
*
* \param[in,out] data Colocation constraint to apply
* \param[in,out] user_data Promotable clone that is constraint's primary
*/
static void
apply_coloc_to_primary(gpointer data, gpointer user_data)
{
pcmk__colocation_t *colocation = data;
pe_resource_t *clone = user_data;
pe_resource_t *dependent = colocation->dependent;
const float factor = colocation->score / (float) INFINITY;
const uint32_t flags = pcmk__coloc_select_active
|pcmk__coloc_select_nonnegative;
if ((colocation->primary_role != pcmk_role_promoted)
|| !pcmk__colocation_has_influence(colocation, NULL)) {
return;
}
pe_rsc_trace(clone, "Applying colocation %s (%s with promoted %s) @%s",
colocation->id, colocation->dependent->id,
colocation->primary->id,
pcmk_readable_score(colocation->score));
dependent->cmds->add_colocated_node_scores(dependent, clone, clone->id,
&clone->allowed_nodes,
colocation, factor, flags);
}
/*!
* \internal
* \brief Set clone instance's sort index to its node's score
*
* \param[in,out] data Promotable clone instance
* \param[in] user_data Parent clone of \p data
*/
static void
set_sort_index_to_node_score(gpointer data, gpointer user_data)
{
pe_resource_t *child = (pe_resource_t *) data;
const pe_resource_t *clone = (const pe_resource_t *) user_data;
pe_node_t *chosen = child->fns->location(child, NULL, FALSE);
if (!pcmk_is_set(child->flags, pe_rsc_managed)
&& (child->next_role == pcmk_role_promoted)) {
child->sort_index = INFINITY;
pe_rsc_trace(clone,
"Final sort index for %s is INFINITY (unmanaged promoted)",
child->id);
} else if ((chosen == NULL) || (child->sort_index < 0)) {
pe_rsc_trace(clone,
"Final sort index for %s is %d (ignoring node score)",
child->id, child->sort_index);
} else {
const pe_node_t *node = g_hash_table_lookup(clone->allowed_nodes,
chosen->details->id);
CRM_ASSERT(node != NULL);
child->sort_index = node->weight;
pe_rsc_trace(clone,
"Adding scores for %s: final sort index for %s is %d",
clone->id, child->id, child->sort_index);
}
}
/*!
* \internal
* \brief Sort a promotable clone's instances by descending promotion priority
*
* \param[in,out] clone Promotable clone to sort
*/
static void
sort_promotable_instances(pe_resource_t *clone)
{
GList *colocations = NULL;
if (pe__set_clone_flag(clone, pe__clone_promotion_constrained)
== pcmk_rc_already) {
return;
}
pe__set_resource_flags(clone, pe_rsc_merging);
for (GList *iter = clone->children; iter != NULL; iter = iter->next) {
pe_resource_t *child = (pe_resource_t *) iter->data;
pe_rsc_trace(clone,
"Adding scores for %s: initial sort index for %s is %d",
clone->id, child->id, child->sort_index);
}
pe__show_node_scores(true, clone, "Before", clone->allowed_nodes,
clone->cluster);
g_list_foreach(clone->children, add_sort_index_to_node_score, clone);
colocations = pcmk__this_with_colocations(clone);
g_list_foreach(colocations, apply_coloc_to_dependent, clone);
g_list_free(colocations);
colocations = pcmk__with_this_colocations(clone);
g_list_foreach(colocations, apply_coloc_to_primary, clone);
g_list_free(colocations);
// Ban resource from all nodes if it needs a ticket but doesn't have it
pcmk__require_promotion_tickets(clone);
pe__show_node_scores(true, clone, "After", clone->allowed_nodes,
clone->cluster);
// Reset sort indexes to final node scores
g_list_foreach(clone->children, set_sort_index_to_node_score, clone);
// Finally, sort instances in descending order of promotion priority
clone->children = g_list_sort(clone->children, cmp_promotable_instance);
pe__clear_resource_flags(clone, pe_rsc_merging);
}
/*!
* \internal
* \brief Find the active instance (if any) of an anonymous clone on a node
*
* \param[in] clone Anonymous clone to check
* \param[in] id Instance ID (without instance number) to check
* \param[in] node Node to check
*
* \return
*/
static pe_resource_t *
find_active_anon_instance(const pe_resource_t *clone, const char *id,
const pe_node_t *node)
{
for (GList *iter = clone->children; iter; iter = iter->next) {
pe_resource_t *child = iter->data;
pe_resource_t *active = NULL;
// Use ->find_rsc() in case this is a cloned group
active = clone->fns->find_rsc(child, id, node,
pcmk_rsc_match_clone_only
|pcmk_rsc_match_current_node);
if (active != NULL) {
return active;
}
}
return NULL;
}
/*
* \brief Check whether an anonymous clone instance is known on a node
*
* \param[in] clone Anonymous clone to check
* \param[in] id Instance ID (without instance number) to check
* \param[in] node Node to check
*
* \return true if \p id instance of \p clone is known on \p node,
* otherwise false
*/
static bool
anonymous_known_on(const pe_resource_t *clone, const char *id,
const pe_node_t *node)
{
for (GList *iter = clone->children; iter; iter = iter->next) {
pe_resource_t *child = iter->data;
/* Use ->find_rsc() because this might be a cloned group, and knowing
* that other members of the group are known here implies nothing.
*/
child = clone->fns->find_rsc(child, id, NULL,
pcmk_rsc_match_clone_only);
CRM_LOG_ASSERT(child != NULL);
if (child != NULL) {
if (g_hash_table_lookup(child->known_on, node->details->id)) {
return true;
}
}
}
return false;
}
/*!
* \internal
* \brief Check whether a node is allowed to run a resource
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p node is allowed to run \p rsc, otherwise false
*/
static bool
is_allowed(const pe_resource_t *rsc, const pe_node_t *node)
{
pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes,
node->details->id);
return (allowed != NULL) && (allowed->weight >= 0);
}
/*!
* \brief Check whether a clone instance's promotion score should be considered
*
* \param[in] rsc Promotable clone instance to check
* \param[in] node Node where score would be applied
*
* \return true if \p rsc's promotion score should be considered on \p node,
* otherwise false
*/
static bool
promotion_score_applies(const pe_resource_t *rsc, const pe_node_t *node)
{
char *id = clone_strip(rsc->id);
const pe_resource_t *parent = pe__const_top_resource(rsc, false);
pe_resource_t *active = NULL;
const char *reason = "allowed";
// Some checks apply only to anonymous clone instances
if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) {
// If instance is active on the node, its score definitely applies
active = find_active_anon_instance(parent, id, node);
if (active == rsc) {
reason = "active";
goto check_allowed;
}
/* If *no* instance is active on this node, this instance's score will
* count if it has been probed on this node.
*/
if ((active == NULL) && anonymous_known_on(parent, id, node)) {
reason = "probed";
goto check_allowed;
}
}
/* If this clone's status is unknown on *all* nodes (e.g. cluster startup),
* take all instances' scores into account, to make sure we use any
* permanent promotion scores.
*/
if ((rsc->running_on == NULL) && (g_hash_table_size(rsc->known_on) == 0)) {
reason = "none probed";
goto check_allowed;
}
/* Otherwise, we've probed and/or started the resource *somewhere*, so
* consider promotion scores on nodes where we know the status.
*/
if ((g_hash_table_lookup(rsc->known_on, node->details->id) != NULL)
|| (pe_find_node_id(rsc->running_on, node->details->id) != NULL)) {
reason = "known";
} else {
pe_rsc_trace(rsc,
"Ignoring %s promotion score (for %s) on %s: not probed",
rsc->id, id, pe__node_name(node));
free(id);
return false;
}
check_allowed:
if (is_allowed(rsc, node)) {
pe_rsc_trace(rsc, "Counting %s promotion score (for %s) on %s: %s",
rsc->id, id, pe__node_name(node), reason);
free(id);
return true;
}
pe_rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: not allowed",
rsc->id, id, pe__node_name(node));
free(id);
return false;
}
/*!
* \internal
* \brief Get the value of a promotion score node attribute
*
* \param[in] rsc Promotable clone instance to get promotion score for
* \param[in] node Node to get promotion score for
* \param[in] name Resource name to use in promotion score attribute name
*
* \return Value of promotion score node attribute for \p rsc on \p node
*/
static const char *
promotion_attr_value(const pe_resource_t *rsc, const pe_node_t *node,
const char *name)
{
char *attr_name = NULL;
const char *attr_value = NULL;
enum pe__rsc_node node_type = pe__rsc_node_assigned;
if (pcmk_is_set(rsc->flags, pe_rsc_provisional)) {
// Not assigned yet
node_type = pe__rsc_node_current;
}
attr_name = pcmk_promotion_score_name(name);
attr_value = pe__node_attribute_calculated(node, attr_name, rsc, node_type,
false);
free(attr_name);
return attr_value;
}
/*!
* \internal
* \brief Get the promotion score for a clone instance on a node
*
* \param[in] rsc Promotable clone instance to get score for
* \param[in] node Node to get score for
* \param[out] is_default If non-NULL, will be set true if no score available
*
* \return Promotion score for \p rsc on \p node (or 0 if none)
*/
static int
promotion_score(const pe_resource_t *rsc, const pe_node_t *node,
bool *is_default)
{
char *name = NULL;
const char *attr_value = NULL;
if (is_default != NULL) {
*is_default = true;
}
CRM_CHECK((rsc != NULL) && (node != NULL), return 0);
/* If this is an instance of a cloned group, the promotion score is the sum
* of all members' promotion scores.
*/
if (rsc->children != NULL) {
int score = 0;
for (const GList *iter = rsc->children;
iter != NULL; iter = iter->next) {
const pe_resource_t *child = (const pe_resource_t *) iter->data;
bool child_default = false;
int child_score = promotion_score(child, node, &child_default);
if (!child_default && (is_default != NULL)) {
*is_default = false;
}
score += child_score;
}
return score;
}
if (!promotion_score_applies(rsc, node)) {
return 0;
}
/* For the promotion score attribute name, use the name the resource is
* known as in resource history, since that's what crm_attribute --promotion
* would have used.
*/
name = (rsc->clone_name == NULL)? rsc->id : rsc->clone_name;
attr_value = promotion_attr_value(rsc, node, name);
if (attr_value != NULL) {
pe_rsc_trace(rsc, "Promotion score for %s on %s = %s",
name, pe__node_name(node), pcmk__s(attr_value, "(unset)"));
} else if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) {
/* If we don't have any resource history yet, we won't have clone_name.
* In that case, for anonymous clones, try the resource name without
* any instance number.
*/
name = clone_strip(rsc->id);
if (strcmp(rsc->id, name) != 0) {
attr_value = promotion_attr_value(rsc, node, name);
pe_rsc_trace(rsc, "Promotion score for %s on %s (for %s) = %s",
name, pe__node_name(node), rsc->id,
pcmk__s(attr_value, "(unset)"));
}
free(name);
}
if (attr_value == NULL) {
return 0;
}
if (is_default != NULL) {
*is_default = false;
}
return char2score(attr_value);
}
/*!
* \internal
* \brief Include promotion scores in instances' node scores and priorities
*
* \param[in,out] rsc Promotable clone resource to update
*/
void
pcmk__add_promotion_scores(pe_resource_t *rsc)
{
if (pe__set_clone_flag(rsc, pe__clone_promotion_added) == pcmk_rc_already) {
return;
}
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) iter->data;
GHashTableIter iter;
pe_node_t *node = NULL;
int score, new_score;
g_hash_table_iter_init(&iter, child_rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (!pcmk__node_available(node, false, false)) {
/* This node will never be promoted, so don't apply the
* promotion score, as that may lead to clone shuffling.
*/
continue;
}
score = promotion_score(child_rsc, node, NULL);
if (score > 0) {
new_score = pcmk__add_scores(node->weight, score);
if (new_score != node->weight) { // Could remain INFINITY
node->weight = new_score;
pe_rsc_trace(rsc,
"Added %s promotion priority (%s) to score "
"on %s (now %s)",
child_rsc->id, pcmk_readable_score(score),
pe__node_name(node),
pcmk_readable_score(new_score));
}
}
if (score > child_rsc->priority) {
pe_rsc_trace(rsc,
"Updating %s priority to promotion score (%d->%d)",
child_rsc->id, child_rsc->priority, score);
child_rsc->priority = score;
}
}
}
}
/*!
* \internal
* \brief If a resource's current role is started, change it to unpromoted
*
* \param[in,out] data Resource to update
* \param[in] user_data Ignored
*/
static void
set_current_role_unpromoted(void *data, void *user_data)
{
pe_resource_t *rsc = (pe_resource_t *) data;
if (rsc->role == pcmk_role_started) {
// Promotable clones should use unpromoted role instead of started
rsc->role = pcmk_role_unpromoted;
}
g_list_foreach(rsc->children, set_current_role_unpromoted, NULL);
}
/*!
* \internal
* \brief Set a resource's next role to unpromoted (or stopped if unassigned)
*
* \param[in,out] data Resource to update
* \param[in] user_data Ignored
*/
static void
set_next_role_unpromoted(void *data, void *user_data)
{
pe_resource_t *rsc = (pe_resource_t *) data;
GList *assigned = NULL;
rsc->fns->location(rsc, &assigned, FALSE);
if (assigned == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "stopped instance");
} else {
pe__set_next_role(rsc, pcmk_role_unpromoted, "unpromoted instance");
g_list_free(assigned);
}
g_list_foreach(rsc->children, set_next_role_unpromoted, NULL);
}
/*!
* \internal
* \brief Set a resource's next role to promoted if not already set
*
* \param[in,out] data Resource to update
* \param[in] user_data Ignored
*/
static void
set_next_role_promoted(void *data, gpointer user_data)
{
pe_resource_t *rsc = (pe_resource_t *) data;
if (rsc->next_role == pcmk_role_unknown) {
pe__set_next_role(rsc, pcmk_role_promoted, "promoted instance");
}
g_list_foreach(rsc->children, set_next_role_promoted, NULL);
}
/*!
* \internal
* \brief Show instance's promotion score on node where it will be active
*
* \param[in,out] instance Promotable clone instance to show
*/
static void
show_promotion_score(pe_resource_t *instance)
{
pe_node_t *chosen = instance->fns->location(instance, NULL, FALSE);
if (pcmk_is_set(instance->cluster->flags, pe_flag_show_scores)
&& !pcmk__is_daemon && (instance->cluster->priv != NULL)) {
pcmk__output_t *out = instance->cluster->priv;
out->message(out, "promotion-score", instance, chosen,
pcmk_readable_score(instance->sort_index));
} else {
pe_rsc_debug(pe__const_top_resource(instance, false),
"%s promotion score on %s: sort=%s priority=%s",
instance->id,
((chosen == NULL)? "none" : pe__node_name(chosen)),
pcmk_readable_score(instance->sort_index),
pcmk_readable_score(instance->priority));
}
}
/*!
* \internal
* \brief Set a clone instance's promotion priority
*
* \param[in,out] data Promotable clone instance to update
* \param[in] user_data Instance's parent clone
*/
static void
set_instance_priority(gpointer data, gpointer user_data)
{
pe_resource_t *instance = (pe_resource_t *) data;
const pe_resource_t *clone = (const pe_resource_t *) user_data;
const pe_node_t *chosen = NULL;
enum rsc_role_e next_role = pcmk_role_unknown;
GList *list = NULL;
pe_rsc_trace(clone, "Assigning priority for %s: %s", instance->id,
role2text(instance->next_role));
if (instance->fns->state(instance, TRUE) == pcmk_role_started) {
set_current_role_unpromoted(instance, NULL);
}
// Only an instance that will be active can be promoted
chosen = instance->fns->location(instance, &list, FALSE);
if (pcmk__list_of_multiple(list)) {
pcmk__config_err("Cannot promote non-colocated child %s",
instance->id);
}
g_list_free(list);
if (chosen == NULL) {
return;
}
next_role = instance->fns->state(instance, FALSE);
switch (next_role) {
case pcmk_role_started:
case pcmk_role_unknown:
// Set instance priority to its promotion score (or -1 if none)
{
bool is_default = false;
instance->priority = promotion_score(instance, chosen,
&is_default);
if (is_default) {
/*
* Default to -1 if no value is set. This allows
* instances eligible for promotion to be specified
* based solely on rsc_location constraints, but
* prevents any instance from being promoted if neither
* a constraint nor a promotion score is present
*/
instance->priority = -1;
}
}
break;
case pcmk_role_unpromoted:
case pcmk_role_stopped:
// Instance can't be promoted
instance->priority = -INFINITY;
break;
case pcmk_role_promoted:
// Nothing needed (re-creating actions after scheduling fencing)
break;
default:
CRM_CHECK(FALSE, crm_err("Unknown resource role %d for %s",
next_role, instance->id));
}
// Add relevant location constraint scores for promoted role
apply_promoted_locations(instance, instance->rsc_location, chosen);
apply_promoted_locations(instance, clone->rsc_location, chosen);
// Consider instance's role-based colocations with other resources
list = pcmk__this_with_colocations(instance);
for (GList *iter = list; iter != NULL; iter = iter->next) {
pcmk__colocation_t *cons = (pcmk__colocation_t *) iter->data;
instance->cmds->apply_coloc_score(instance, cons->primary, cons, true);
}
g_list_free(list);
instance->sort_index = instance->priority;
if (next_role == pcmk_role_promoted) {
instance->sort_index = INFINITY;
}
pe_rsc_trace(clone, "Assigning %s priority = %d",
instance->id, instance->priority);
}
/*!
* \internal
* \brief Set a promotable clone instance's role
*
* \param[in,out] data Promotable clone instance to update
* \param[in,out] user_data Pointer to count of instances chosen for promotion
*/
static void
set_instance_role(gpointer data, gpointer user_data)
{
pe_resource_t *instance = (pe_resource_t *) data;
int *count = (int *) user_data;
const pe_resource_t *clone = pe__const_top_resource(instance, false);
pe_node_t *chosen = NULL;
show_promotion_score(instance);
if (instance->sort_index < 0) {
pe_rsc_trace(clone, "Not supposed to promote instance %s",
instance->id);
} else if ((*count < pe__clone_promoted_max(instance))
|| !pcmk_is_set(clone->flags, pe_rsc_managed)) {
chosen = node_to_be_promoted_on(instance);
}
if (chosen == NULL) {
set_next_role_unpromoted(instance, NULL);
return;
}
if ((instance->role < pcmk_role_promoted)
- && !pcmk_is_set(instance->cluster->flags, pe_flag_have_quorum)
+ && !pcmk_is_set(instance->cluster->flags, pcmk_sched_quorate)
&& (instance->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Clone instance %s cannot be promoted without quorum",
instance->id);
set_next_role_unpromoted(instance, NULL);
return;
}
chosen->count++;
pe_rsc_info(clone, "Choosing %s (%s) on %s for promotion",
instance->id, role2text(instance->role),
pe__node_name(chosen));
set_next_role_promoted(instance, NULL);
(*count)++;
}
/*!
* \internal
* \brief Set roles for all instances of a promotable clone
*
* \param[in,out] rsc Promotable clone resource to update
*/
void
pcmk__set_instance_roles(pe_resource_t *rsc)
{
int promoted = 0;
GHashTableIter iter;
pe_node_t *node = NULL;
// Repurpose count to track the number of promoted instances assigned
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
node->count = 0;
}
// Set instances' promotion priorities and sort by highest priority first
g_list_foreach(rsc->children, set_instance_priority, rsc);
sort_promotable_instances(rsc);
// Choose the first N eligible instances to be promoted
g_list_foreach(rsc->children, set_instance_role, &promoted);
pe_rsc_info(rsc, "%s: Promoted %d instances of a possible %d",
rsc->id, promoted, pe__clone_promoted_max(rsc));
}
/*!
*
* \internal
* \brief Create actions for promotable clone instances
*
* \param[in,out] clone Promotable clone to create actions for
* \param[out] any_promoting Will be set true if any instance is promoting
* \param[out] any_demoting Will be set true if any instance is demoting
*/
static void
create_promotable_instance_actions(pe_resource_t *clone,
bool *any_promoting, bool *any_demoting)
{
for (GList *iter = clone->children; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
instance->cmds->create_actions(instance);
check_for_role_change(instance, any_demoting, any_promoting);
}
}
/*!
* \internal
* \brief Reset each promotable instance's resource priority
*
* Reset the priority of each instance of a promotable clone to the clone's
* priority (after promotion actions are scheduled, when instance priorities
* were repurposed as promotion scores).
*
* \param[in,out] clone Promotable clone to reset
*/
static void
reset_instance_priorities(pe_resource_t *clone)
{
for (GList *iter = clone->children; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
instance->priority = clone->priority;
}
}
/*!
* \internal
* \brief Create actions specific to promotable clones
*
* \param[in,out] clone Promotable clone to create actions for
*/
void
pcmk__create_promotable_actions(pe_resource_t *clone)
{
bool any_promoting = false;
bool any_demoting = false;
// Create actions for each clone instance individually
create_promotable_instance_actions(clone, &any_promoting, &any_demoting);
// Create pseudo-actions for clone as a whole
pe__create_promotable_pseudo_ops(clone, any_promoting, any_demoting);
// Undo our temporary repurposing of resource priority for instances
reset_instance_priorities(clone);
}
/*!
* \internal
* \brief Create internal orderings for a promotable clone's instances
*
* \param[in,out] clone Promotable clone instance to order
*/
void
pcmk__order_promotable_instances(pe_resource_t *clone)
{
pe_resource_t *previous = NULL; // Needed for ordered clones
pcmk__promotable_restart_ordering(clone);
for (GList *iter = clone->children; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
// Demote before promote
pcmk__order_resource_actions(instance, PCMK_ACTION_DEMOTE,
instance, PCMK_ACTION_PROMOTE,
pe_order_optional);
order_instance_promotion(clone, instance, previous);
order_instance_demotion(clone, instance, previous);
previous = instance;
}
}
/*!
* \internal
* \brief Update dependent's allowed nodes for colocation with promotable
*
* \param[in,out] dependent Dependent resource to update
* \param[in] primary Primary resource
* \param[in] primary_node Node where an instance of the primary will be
* \param[in] colocation Colocation constraint to apply
*/
static void
update_dependent_allowed_nodes(pe_resource_t *dependent,
const pe_resource_t *primary,
const pe_node_t *primary_node,
const pcmk__colocation_t *colocation)
{
GHashTableIter iter;
pe_node_t *node = NULL;
const char *primary_value = NULL;
const char *attr = colocation->node_attribute;
if (colocation->score >= INFINITY) {
return; // Colocation is mandatory, so allowed node scores don't matter
}
primary_value = pcmk__colocation_node_attr(primary_node, attr, primary);
pe_rsc_trace(colocation->primary,
"Applying %s (%s with %s on %s by %s @%d) to %s",
colocation->id, colocation->dependent->id,
colocation->primary->id, pe__node_name(primary_node), attr,
colocation->score, dependent->id);
g_hash_table_iter_init(&iter, dependent->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
const char *dependent_value = pcmk__colocation_node_attr(node, attr,
dependent);
if (pcmk__str_eq(primary_value, dependent_value, pcmk__str_casei)) {
node->weight = pcmk__add_scores(node->weight, colocation->score);
pe_rsc_trace(colocation->primary,
"Added %s score (%s) to %s (now %s)",
colocation->id, pcmk_readable_score(colocation->score),
pe__node_name(node),
pcmk_readable_score(node->weight));
}
}
}
/*!
* \brief Update dependent for a colocation with a promotable clone
*
* \param[in] primary Primary resource in the colocation
* \param[in,out] dependent Dependent resource in the colocation
* \param[in] colocation Colocation constraint to apply
*/
void
pcmk__update_dependent_with_promotable(const pe_resource_t *primary,
pe_resource_t *dependent,
const pcmk__colocation_t *colocation)
{
GList *affected_nodes = NULL;
/* Build a list of all nodes where an instance of the primary will be, and
* (for optional colocations) update the dependent's allowed node scores for
* each one.
*/
for (GList *iter = primary->children; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
pe_node_t *node = instance->fns->location(instance, NULL, FALSE);
if (node == NULL) {
continue;
}
if (instance->fns->state(instance, FALSE) == colocation->primary_role) {
update_dependent_allowed_nodes(dependent, primary, node,
colocation);
affected_nodes = g_list_prepend(affected_nodes, node);
}
}
/* For mandatory colocations, add the primary's node score to the
* dependent's node score for each affected node, and ban the dependent
* from all other nodes.
*
* However, skip this for promoted-with-promoted colocations, otherwise
* inactive dependent instances can't start (in the unpromoted role).
*/
if ((colocation->score >= INFINITY)
&& ((colocation->dependent_role != pcmk_role_promoted)
|| (colocation->primary_role != pcmk_role_promoted))) {
pe_rsc_trace(colocation->primary,
"Applying %s (mandatory %s with %s) to %s",
colocation->id, colocation->dependent->id,
colocation->primary->id, dependent->id);
pcmk__colocation_intersect_nodes(dependent, primary, colocation,
affected_nodes, true);
}
g_list_free(affected_nodes);
}
/*!
* \internal
* \brief Update dependent priority for colocation with promotable
*
* \param[in] primary Primary resource in the colocation
* \param[in,out] dependent Dependent resource in the colocation
* \param[in] colocation Colocation constraint to apply
*/
void
pcmk__update_promotable_dependent_priority(const pe_resource_t *primary,
pe_resource_t *dependent,
const pcmk__colocation_t *colocation)
{
pe_resource_t *primary_instance = NULL;
// Look for a primary instance where dependent will be
primary_instance = pcmk__find_compatible_instance(dependent, primary,
colocation->primary_role,
false);
if (primary_instance != NULL) {
// Add primary instance's priority to dependent's
int new_priority = pcmk__add_scores(dependent->priority,
colocation->score);
pe_rsc_trace(colocation->primary,
"Applying %s (%s with %s) to %s priority (%s + %s = %s)",
colocation->id, colocation->dependent->id,
colocation->primary->id, dependent->id,
pcmk_readable_score(dependent->priority),
pcmk_readable_score(colocation->score),
pcmk_readable_score(new_priority));
dependent->priority = new_priority;
} else if (colocation->score >= INFINITY) {
// Mandatory colocation, but primary won't be here
pe_rsc_trace(colocation->primary,
"Applying %s (%s with %s) to %s: can't be promoted",
colocation->id, colocation->dependent->id,
colocation->primary->id, dependent->id);
dependent->priority = -INFINITY;
}
}
diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c
index 9bc9bd6987..c1c57195c6 100644
--- a/lib/pacemaker/pcmk_scheduler.c
+++ b/lib/pacemaker/pcmk_scheduler.c
@@ -1,816 +1,816 @@
/*
* Copyright 2004-2023 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 <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/scheduler_internal.h>
#include <glib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
CRM_TRACE_INIT_DATA(pacemaker);
/*!
* \internal
* \brief Do deferred action checks after assignment
*
* When unpacking the resource history, the scheduler checks for resource
* configurations that have changed since an action was run. However, at that
* time, bundles using the REMOTE_CONTAINER_HACK don't have their final
* parameter information, so instead they add a deferred check to a list. This
* function processes one entry in that list.
*
* \param[in,out] rsc Resource that action history is for
* \param[in,out] node Node that action history is for
* \param[in] rsc_op Action history entry
* \param[in] check Type of deferred check to do
*/
static void
check_params(pe_resource_t *rsc, pe_node_t *node, const xmlNode *rsc_op,
enum pcmk__check_parameters check)
{
const char *reason = NULL;
op_digest_cache_t *digest_data = NULL;
switch (check) {
case pcmk__check_active:
if (pcmk__check_action_config(rsc, node, rsc_op)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL)) {
reason = "action definition changed";
}
break;
case pcmk__check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node,
rsc->cluster);
switch (digest_data->rc) {
case RSC_DIGEST_UNKNOWN:
crm_trace("Resource %s history entry %s on %s has "
"no digest to compare",
rsc->id, ID(rsc_op), node->details->id);
break;
case RSC_DIGEST_MATCH:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason != NULL) {
pe__clear_failcount(rsc, node, reason, rsc->cluster);
}
}
/*!
* \internal
* \brief Check whether a resource has failcount clearing scheduled on a node
*
* \param[in] node Node to check
* \param[in] rsc Resource to check
*
* \return true if \p rsc has failcount clearing scheduled on \p node,
* otherwise false
*/
static bool
failcount_clear_action_exists(const pe_node_t *node, const pe_resource_t *rsc)
{
GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT,
TRUE);
if (list != NULL) {
g_list_free(list);
return true;
}
return false;
}
/*!
* \internal
* \brief Ban a resource from a node if it reached its failure threshold there
*
* \param[in,out] data Resource to check failure threshold for
* \param[in] user_data Node to check resource on
*/
static void
check_failure_threshold(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
// If this is a collective resource, apply recursively to children instead
if (rsc->children != NULL) {
g_list_foreach(rsc->children, check_failure_threshold, user_data);
return;
}
if (!failcount_clear_action_exists(node, rsc)) {
/* Don't force the resource away from this node due to a failcount
* that's going to be cleared.
*
* @TODO Failcount clearing can be scheduled in
* pcmk__handle_rsc_config_changes() via process_rsc_history(), or in
* schedule_resource_actions() via check_params(). This runs well before
* then, so it cannot detect those, meaning we might check the migration
* threshold when we shouldn't. Worst case, we stop or move the
* resource, then move it back in the next transition.
*/
pe_resource_t *failed = NULL;
if (pcmk__threshold_reached(rsc, node, &failed)) {
resource_location(failed, node, -INFINITY, "__fail_limit__",
rsc->cluster);
}
}
}
/*!
* \internal
* \brief If resource has exclusive discovery, ban node if not allowed
*
* Location constraints have a resource-discovery option that allows users to
* specify where probes are done for the affected resource. If this is set to
* exclusive, probes will only be done on nodes listed in exclusive constraints.
* This function bans the resource from the node if the node is not listed.
*
* \param[in,out] data Resource to check
* \param[in] user_data Node to check resource on
*/
static void
apply_exclusive_discovery(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
if (rsc->exclusive_discover
|| pe__const_top_resource(rsc, false)->exclusive_discover) {
pe_node_t *match = NULL;
// If this is a collective resource, apply recursively to children
g_list_foreach(rsc->children, apply_exclusive_discovery, user_data);
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if ((match != NULL)
&& (match->rsc_discover_mode != pe_discover_exclusive)) {
match->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Apply stickiness to a resource if appropriate
*
* \param[in,out] data Resource to check for stickiness
* \param[in] user_data Ignored
*/
static void
apply_stickiness(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
pe_node_t *node = NULL;
// If this is a collective resource, apply recursively to children instead
if (rsc->children != NULL) {
g_list_foreach(rsc->children, apply_stickiness, NULL);
return;
}
/* A resource is sticky if it is managed, has stickiness configured, and is
* active on a single node.
*/
if (!pcmk_is_set(rsc->flags, pe_rsc_managed)
|| (rsc->stickiness < 1) || !pcmk__list_of_1(rsc->running_on)) {
return;
}
node = rsc->running_on->data;
/* In a symmetric cluster, stickiness can always be used. In an
* asymmetric cluster, we have to check whether the resource is still
* allowed on the node, so we don't keep the resource somewhere it is no
* longer explicitly enabled.
*/
if (!pcmk_is_set(rsc->cluster->flags, pe_flag_symmetric_cluster)
&& (g_hash_table_lookup(rsc->allowed_nodes,
node->details->id) == NULL)) {
pe_rsc_debug(rsc,
"Ignoring %s stickiness because the cluster is "
"asymmetric and %s is not explicitly allowed",
rsc->id, pe__node_name(node));
return;
}
pe_rsc_debug(rsc, "Resource %s has %d stickiness on %s",
rsc->id, rsc->stickiness, pe__node_name(node));
resource_location(rsc, node, rsc->stickiness, "stickiness", rsc->cluster);
}
/*!
* \internal
* \brief Apply shutdown locks for all resources as appropriate
*
* \param[in,out] data_set Cluster working set
*/
static void
apply_shutdown_locks(pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
return;
}
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->shutdown_lock(rsc);
}
}
/*!
* \internal
* \brief Calculate the number of available nodes in the cluster
*
* \param[in,out] data_set Cluster working set
*/
static void
count_available_nodes(pe_working_set_t *data_set)
{
if (pcmk_is_set(data_set->flags, pe_flag_no_compat)) {
return;
}
// @COMPAT for API backward compatibility only (cluster does not use value)
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
if ((node != NULL) && (node->weight >= 0) && node->details->online
&& (node->details->type != node_ping)) {
data_set->max_valid_nodes++;
}
}
crm_trace("Online node count: %d", data_set->max_valid_nodes);
}
/*
* \internal
* \brief Apply node-specific scheduling criteria
*
* After the CIB has been unpacked, process node-specific scheduling criteria
* including shutdown locks, location constraints, resource stickiness,
* migration thresholds, and exclusive resource discovery.
*/
static void
apply_node_criteria(pe_working_set_t *data_set)
{
crm_trace("Applying node-specific scheduling criteria");
apply_shutdown_locks(data_set);
count_available_nodes(data_set);
pcmk__apply_locations(data_set);
g_list_foreach(data_set->resources, apply_stickiness, NULL);
for (GList *node_iter = data_set->nodes; node_iter != NULL;
node_iter = node_iter->next) {
for (GList *rsc_iter = data_set->resources; rsc_iter != NULL;
rsc_iter = rsc_iter->next) {
check_failure_threshold(rsc_iter->data, node_iter->data);
apply_exclusive_discovery(rsc_iter->data, node_iter->data);
}
}
}
/*!
* \internal
* \brief Assign resources to nodes
*
* \param[in,out] data_set Cluster working set
*/
static void
assign_resources(pe_working_set_t *data_set)
{
GList *iter = NULL;
crm_trace("Assigning resources to nodes");
if (!pcmk__str_eq(data_set->placement_strategy, "default",
pcmk__str_casei)) {
pcmk__sort_resources(data_set);
}
pcmk__show_node_capacities("Original", data_set);
if (pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) {
/* Assign remote connection resources first (which will also assign any
* colocation dependencies). If the connection is migrating, always
* prefer the partial migration target.
*/
for (iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (rsc->is_remote_node) {
pe_rsc_trace(rsc, "Assigning remote connection resource '%s'",
rsc->id);
rsc->cmds->assign(rsc, rsc->partial_migration_target, true);
}
}
}
/* now do the rest of the resources */
for (iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (!rsc->is_remote_node) {
pe_rsc_trace(rsc, "Assigning %s resource '%s'",
rsc->xml->name, rsc->id);
rsc->cmds->assign(rsc, NULL, true);
}
}
pcmk__show_node_capacities("Remaining", data_set);
}
/*!
* \internal
* \brief Schedule fail count clearing on online nodes if resource is orphaned
*
* \param[in,out] data Resource to check
* \param[in] user_data Ignored
*/
static void
clear_failcounts_if_orphaned(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
return;
}
crm_trace("Clear fail counts for orphaned resource %s", rsc->id);
/* There's no need to recurse into rsc->children because those
* should just be unassigned clone instances.
*/
for (GList *iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
pe_action_t *clear_op = NULL;
if (!node->details->online) {
continue;
}
if (pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL) == 0) {
continue;
}
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
rsc->cluster);
/* We can't use order_action_then_stop() here because its
* pe_order_preserve breaks things
*/
pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc),
NULL, pe_order_optional, rsc->cluster);
}
}
/*!
* \internal
* \brief Schedule any resource actions needed
*
* \param[in,out] data_set Cluster working set
*/
static void
schedule_resource_actions(pe_working_set_t *data_set)
{
// Process deferred action checks
pe__foreach_param_check(data_set, check_params);
pe__free_param_checks(data_set);
if (pcmk_is_set(data_set->flags, pe_flag_startup_probes)) {
crm_trace("Scheduling probes");
pcmk__schedule_probes(data_set);
}
if (pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) {
g_list_foreach(data_set->resources, clear_failcounts_if_orphaned, NULL);
}
crm_trace("Scheduling resource actions");
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->create_actions(rsc);
}
}
/*!
* \internal
* \brief Check whether a resource or any of its descendants are managed
*
* \param[in] rsc Resource to check
*
* \return true if resource or any descendant is managed, otherwise false
*/
static bool
is_managed(const pe_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pe_rsc_managed)) {
return true;
}
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
if (is_managed((pe_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether any resources in the cluster are managed
*
* \param[in] data_set Cluster working set
*
* \return true if any resource is managed, otherwise false
*/
static bool
any_managed_resources(const pe_working_set_t *data_set)
{
for (const GList *iter = data_set->resources;
iter != NULL; iter = iter->next) {
if (is_managed((const pe_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a node requires fencing
*
* \param[in] node Node to check
* \param[in] have_managed Whether any resource in cluster is managed
*
* \return true if \p node should be fenced, otherwise false
*/
static bool
needs_fencing(const pe_node_t *node, bool have_managed)
{
return have_managed && node->details->unclean
&& pe_can_fence(node->details->data_set, node);
}
/*!
* \internal
* \brief Check whether a node requires shutdown
*
* \param[in] node Node to check
*
* \return true if \p node should be shut down, otherwise false
*/
static bool
needs_shutdown(const pe_node_t *node)
{
if (pe__is_guest_or_remote_node(node)) {
/* Do not send shutdown actions for Pacemaker Remote nodes.
* @TODO We might come up with a good use for this in the future.
*/
return false;
}
return node->details->online && node->details->shutdown;
}
/*!
* \internal
* \brief Track and order non-DC fencing
*
* \param[in,out] list List of existing non-DC fencing actions
* \param[in,out] action Fencing action to prepend to \p list
* \param[in] data_set Cluster working set
*
* \return (Possibly new) head of \p list
*/
static GList *
add_nondc_fencing(GList *list, pe_action_t *action,
const pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)
&& (list != NULL)) {
/* Concurrent fencing is disabled, so order each non-DC
* fencing in a chain. If there is any DC fencing or
* shutdown, it will be ordered after the last action in the
* chain later.
*/
order_actions((pe_action_t *) list->data, action, pe_order_optional);
}
return g_list_prepend(list, action);
}
/*!
* \internal
* \brief Schedule a node for fencing
*
* \param[in,out] node Node that requires fencing
*/
static pe_action_t *
schedule_fencing(pe_node_t *node)
{
pe_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean",
FALSE, node->details->data_set);
pe_warn("Scheduling node %s for fencing", pe__node_name(node));
pcmk__order_vs_fence(fencing, node->details->data_set);
return fencing;
}
/*!
* \internal
* \brief Create and order node fencing and shutdown actions
*
* \param[in,out] data_set Cluster working set
*/
static void
schedule_fencing_and_shutdowns(pe_working_set_t *data_set)
{
pe_action_t *dc_down = NULL;
bool integrity_lost = false;
bool have_managed = any_managed_resources(data_set);
GList *fencing_ops = NULL;
GList *shutdown_ops = NULL;
crm_trace("Scheduling fencing and shutdowns as needed");
if (!have_managed) {
crm_notice("No fencing will be done until there are resources "
"to manage");
}
// Check each node for whether it needs fencing or shutdown
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
pe_action_t *fencing = NULL;
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pe__is_guest_node(node)) {
if (node->details->remote_requires_reset && have_managed
&& pe_can_fence(data_set, node)) {
pcmk__fence_guest(node);
}
continue;
}
if (needs_fencing(node, have_managed)) {
fencing = schedule_fencing(node);
// Track DC and non-DC fence actions separately
if (node->details->is_dc) {
dc_down = fencing;
} else {
fencing_ops = add_nondc_fencing(fencing_ops, fencing, data_set);
}
} else if (needs_shutdown(node)) {
pe_action_t *down_op = pcmk__new_shutdown_action(node);
// Track DC and non-DC shutdown actions separately
if (node->details->is_dc) {
dc_down = down_op;
} else {
shutdown_ops = g_list_prepend(shutdown_ops, down_op);
}
}
if ((fencing == NULL) && node->details->unclean) {
integrity_lost = true;
pe_warn("Node %s is unclean but cannot be fenced",
pe__node_name(node));
}
}
if (integrity_lost) {
if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
pe_warn("Resource functionality and data integrity cannot be "
"guaranteed (configure, enable, and test fencing to "
"correct this)");
- } else if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
+ } else if (!pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
crm_notice("Unclean nodes will not be fenced until quorum is "
"attained or no-quorum-policy is set to ignore");
}
}
if (dc_down != NULL) {
/* Order any non-DC shutdowns before any DC shutdown, to avoid repeated
* DC elections. However, we don't want to order non-DC shutdowns before
* a DC *fencing*, because even though we don't want a node that's
* shutting down to become DC, the DC fencing could be ordered before a
* clone stop that's also ordered before the shutdowns, thus leading to
* a graph loop.
*/
if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
pcmk__order_after_each(dc_down, shutdown_ops);
}
// Order any non-DC fencing before any DC fencing or shutdown
if (pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)) {
/* With concurrent fencing, order each non-DC fencing action
* separately before any DC fencing or shutdown.
*/
pcmk__order_after_each(dc_down, fencing_ops);
} else if (fencing_ops != NULL) {
/* Without concurrent fencing, the non-DC fencing actions are
* already ordered relative to each other, so we just need to order
* the DC fencing after the last action in the chain (which is the
* first item in the list).
*/
order_actions((pe_action_t *) fencing_ops->data, dc_down,
pe_order_optional);
}
}
g_list_free(fencing_ops);
g_list_free(shutdown_ops);
}
static void
log_resource_details(pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
GList *all = NULL;
/* Due to the `crm_mon --node=` feature, out->message() for all the
* resource-related messages expects a list of nodes that we are allowed to
* output information for. Here, we create a wildcard to match all nodes.
*/
all = g_list_prepend(all, (gpointer) "*");
for (GList *item = data_set->resources; item != NULL; item = item->next) {
pe_resource_t *rsc = (pe_resource_t *) item->data;
// Log all resources except inactive orphans
if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)
|| (rsc->role != pcmk_role_stopped)) {
out->message(out, crm_map_element_name(rsc->xml), 0, rsc, all, all);
}
}
g_list_free(all);
}
static void
log_all_actions(pe_working_set_t *data_set)
{
/* This only ever outputs to the log, so ignore whatever output object was
* previously set and just log instead.
*/
pcmk__output_t *prev_out = data_set->priv;
pcmk__output_t *out = NULL;
if (pcmk__log_output_new(&out) != pcmk_rc_ok) {
return;
}
pe__register_messages(out);
pcmk__register_lib_messages(out);
pcmk__output_set_log_level(out, LOG_NOTICE);
data_set->priv = out;
out->begin_list(out, NULL, NULL, "Actions");
pcmk__output_actions(data_set);
out->end_list(out);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
data_set->priv = prev_out;
}
/*!
* \internal
* \brief Log all required but unrunnable actions at trace level
*
* \param[in] data_set Cluster working set
*/
static void
log_unrunnable_actions(const pe_working_set_t *data_set)
{
const uint64_t flags = pe_action_optional
|pe_action_runnable
|pe_action_pseudo;
crm_trace("Required but unrunnable actions:");
for (const GList *iter = data_set->actions;
iter != NULL; iter = iter->next) {
const pe_action_t *action = (const pe_action_t *) iter->data;
if (!pcmk_any_flags_set(action->flags, flags)) {
pcmk__log_action("\t", action, true);
}
}
}
/*!
* \internal
* \brief Unpack the CIB for scheduling
*
* \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked)
* \param[in] flags Working set flags to set in addition to defaults
* \param[in,out] data_set Cluster working set
*/
static void
unpack_cib(xmlNode *cib, unsigned long long flags, pe_working_set_t *data_set)
{
const char* localhost_save = NULL;
if (pcmk_is_set(data_set->flags, pe_flag_have_status)) {
crm_trace("Reusing previously calculated cluster status");
pe__set_working_set_flags(data_set, flags);
return;
}
if (data_set->localhost) {
localhost_save = data_set->localhost;
}
CRM_ASSERT(cib != NULL);
crm_trace("Calculating cluster status");
/* This will zero the entire struct without freeing anything first, so
* callers should never call pcmk__schedule_actions() with a populated data
* set unless pe_flag_have_status is set (i.e. cluster_status() was
* previously called, whether directly or via pcmk__schedule_actions()).
*/
set_working_set_defaults(data_set);
if (localhost_save) {
data_set->localhost = localhost_save;
}
pe__set_working_set_flags(data_set, flags);
data_set->input = cib;
cluster_status(data_set); // Sets pe_flag_have_status
}
/*!
* \internal
* \brief Run the scheduler for a given CIB
*
* \param[in,out] cib CIB XML to use as scheduler input
* \param[in] flags Working set flags to set in addition to defaults
* \param[in,out] data_set Cluster working set
*/
void
pcmk__schedule_actions(xmlNode *cib, unsigned long long flags,
pe_working_set_t *data_set)
{
unpack_cib(cib, flags, data_set);
pcmk__set_assignment_methods(data_set);
pcmk__apply_node_health(data_set);
pcmk__unpack_constraints(data_set);
if (pcmk_is_set(data_set->flags, pe_flag_check_config)) {
return;
}
if (!pcmk_is_set(data_set->flags, pe_flag_quick_location) &&
pcmk__is_daemon) {
log_resource_details(data_set);
}
apply_node_criteria(data_set);
if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
return;
}
pcmk__create_internal_constraints(data_set);
pcmk__handle_rsc_config_changes(data_set);
assign_resources(data_set);
schedule_resource_actions(data_set);
/* Remote ordering constraints need to happen prior to calculating fencing
* because it is one more place we can mark nodes as needing fencing.
*/
pcmk__order_remote_connection_actions(data_set);
schedule_fencing_and_shutdowns(data_set);
pcmk__apply_orderings(data_set);
log_all_actions(data_set);
pcmk__create_graph(data_set);
if (get_crm_log_level() == LOG_TRACE) {
log_unrunnable_actions(data_set);
}
}
diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c
index 46605e1989..983c322813 100644
--- a/lib/pengine/pe_actions.c
+++ b/lib/pengine/pe_actions.c
@@ -1,1743 +1,1743 @@
/*
* Copyright 2004-2023 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 <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/scheduler_internal.h>
#include <crm/pengine/internal.h>
#include <crm/common/xml_internal.h>
#include "pe_status_private.h"
static void unpack_operation(pe_action_t *action, const xmlNode *xml_obj,
const pe_resource_t *container, guint interval_ms);
static void
add_singleton(pe_working_set_t *data_set, pe_action_t *action)
{
if (data_set->singletons == NULL) {
data_set->singletons = pcmk__strkey_table(NULL, NULL);
}
g_hash_table_insert(data_set->singletons, action->uuid, action);
}
static pe_action_t *
lookup_singleton(pe_working_set_t *data_set, const char *action_uuid)
{
if (data_set->singletons == NULL) {
return NULL;
}
return g_hash_table_lookup(data_set->singletons, action_uuid);
}
/*!
* \internal
* \brief Find an existing action that matches arguments
*
* \param[in] key Action key to match
* \param[in] rsc Resource to match (if any)
* \param[in] node Node to match (if any)
* \param[in] data_set Cluster working set
*
* \return Existing action that matches arguments (or NULL if none)
*/
static pe_action_t *
find_existing_action(const char *key, const pe_resource_t *rsc,
const pe_node_t *node, const pe_working_set_t *data_set)
{
GList *matches = NULL;
pe_action_t *action = NULL;
/* When rsc is NULL, it would be quicker to check data_set->singletons,
* but checking all data_set->actions takes the node into account.
*/
matches = find_actions(((rsc == NULL)? data_set->actions : rsc->actions),
key, node);
if (matches == NULL) {
return NULL;
}
CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches));
action = matches->data;
g_list_free(matches);
return action;
}
static xmlNode *
find_rsc_op_entry_helper(const pe_resource_t *rsc, const char *key,
gboolean include_disabled)
{
guint interval_ms = 0;
gboolean do_retry = TRUE;
char *local_key = NULL;
const char *name = NULL;
const char *interval_spec = NULL;
char *match_key = NULL;
xmlNode *op = NULL;
xmlNode *operation = NULL;
retry:
for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL;
operation = pcmk__xe_next(operation)) {
if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) {
bool enabled = false;
name = crm_element_value(operation, "name");
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (!include_disabled && pcmk__xe_get_bool_attr(operation, "enabled", &enabled) == pcmk_rc_ok &&
!enabled) {
continue;
}
interval_ms = crm_parse_interval_spec(interval_spec);
match_key = pcmk__op_key(rsc->id, name, interval_ms);
if (pcmk__str_eq(key, match_key, pcmk__str_casei)) {
op = operation;
}
free(match_key);
if (rsc->clone_name) {
match_key = pcmk__op_key(rsc->clone_name, name, interval_ms);
if (pcmk__str_eq(key, match_key, pcmk__str_casei)) {
op = operation;
}
free(match_key);
}
if (op != NULL) {
free(local_key);
return op;
}
}
}
free(local_key);
if (do_retry == FALSE) {
return NULL;
}
do_retry = FALSE;
if ((strstr(key, PCMK_ACTION_MIGRATE_TO) != NULL)
|| (strstr(key, PCMK_ACTION_MIGRATE_FROM) != NULL)) {
local_key = pcmk__op_key(rsc->id, "migrate", 0);
key = local_key;
goto retry;
} else if (strstr(key, "_notify_")) {
local_key = pcmk__op_key(rsc->id, PCMK_ACTION_NOTIFY, 0);
key = local_key;
goto retry;
}
return NULL;
}
xmlNode *
find_rsc_op_entry(const pe_resource_t *rsc, const char *key)
{
return find_rsc_op_entry_helper(rsc, key, FALSE);
}
/*!
* \internal
* \brief Create a new action object
*
* \param[in] key Action key
* \param[in] task Action name
* \param[in,out] rsc Resource that action is for (if any)
* \param[in] node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in] for_graph Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \return Newly allocated action
* \note This function takes ownership of \p key. It is the caller's
* responsibility to free the return value with pe_free_action().
*/
static pe_action_t *
new_action(char *key, const char *task, pe_resource_t *rsc,
const pe_node_t *node, bool optional, bool for_graph,
pe_working_set_t *data_set)
{
pe_action_t *action = calloc(1, sizeof(pe_action_t));
CRM_ASSERT(action != NULL);
action->rsc = rsc;
action->task = strdup(task); CRM_ASSERT(action->task != NULL);
action->uuid = key;
action->extra = pcmk__strkey_table(free, free);
action->meta = pcmk__strkey_table(free, free);
if (node) {
action->node = pe__copy_node(node);
}
if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_casei)) {
// Resource history deletion for a node can be done on the DC
pe__set_action_flags(action, pe_action_dc);
}
pe__set_action_flags(action, pe_action_runnable);
if (optional) {
pe__set_action_flags(action, pe_action_optional);
} else {
pe__clear_action_flags(action, pe_action_optional);
}
if (rsc != NULL) {
guint interval_ms = 0;
action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE);
parse_op_key(key, NULL, NULL, &interval_ms);
unpack_operation(action, action->op_entry, rsc->container, interval_ms);
}
if (for_graph) {
pe_rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s",
(optional? "optional" : "required"),
data_set->action_id, key, task,
((rsc == NULL)? "no resource" : rsc->id),
pe__node_name(node));
action->id = data_set->action_id++;
data_set->actions = g_list_prepend(data_set->actions, action);
if (rsc == NULL) {
add_singleton(data_set, action);
} else {
rsc->actions = g_list_prepend(rsc->actions, action);
}
}
return action;
}
/*!
* \internal
* \brief Evaluate node attribute values for an action
*
* \param[in,out] action Action to unpack attributes for
* \param[in,out] data_set Cluster working set
*/
static void
unpack_action_node_attributes(pe_action_t *action, pe_working_set_t *data_set)
{
if (!pcmk_is_set(action->flags, pe_action_have_node_attrs)
&& (action->op_entry != NULL)) {
pe_rule_eval_data_t rule_data = {
.node_hash = action->node->details->attrs,
.role = pcmk_role_unknown,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
pe__set_action_flags(action, pe_action_have_node_attrs);
pe__unpack_dataset_nvpairs(action->op_entry, XML_TAG_ATTR_SETS,
&rule_data, action->extra, NULL,
FALSE, data_set);
}
}
/*!
* \internal
* \brief Update an action's optional flag
*
* \param[in,out] action Action to update
* \param[in] optional Requested optional status
*/
static void
update_action_optional(pe_action_t *action, gboolean optional)
{
// Force a non-recurring action to be optional if its resource is unmanaged
if ((action->rsc != NULL) && (action->node != NULL)
&& !pcmk_is_set(action->flags, pe_action_pseudo)
&& !pcmk_is_set(action->rsc->flags, pe_rsc_managed)
&& (g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS) == NULL)) {
pe_rsc_debug(action->rsc, "%s on %s is optional (%s is unmanaged)",
action->uuid, pe__node_name(action->node),
action->rsc->id);
pe__set_action_flags(action, pe_action_optional);
// We shouldn't clear runnable here because ... something
// Otherwise require the action if requested
} else if (!optional) {
pe__clear_action_flags(action, pe_action_optional);
}
}
static enum pe_quorum_policy
effective_quorum_policy(pe_resource_t *rsc, pe_working_set_t *data_set)
{
enum pe_quorum_policy policy = data_set->no_quorum_policy;
- if (pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
+ if (pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
policy = pcmk_no_quorum_ignore;
} else if (data_set->no_quorum_policy == pcmk_no_quorum_demote) {
switch (rsc->role) {
case pcmk_role_promoted:
case pcmk_role_unpromoted:
if (rsc->next_role > pcmk_role_unpromoted) {
pe__set_next_role(rsc, pcmk_role_unpromoted,
"no-quorum-policy=demote");
}
policy = pcmk_no_quorum_ignore;
break;
default:
policy = pcmk_no_quorum_stop;
break;
}
}
return policy;
}
/*!
* \internal
* \brief Update a resource action's runnable flag
*
* \param[in,out] action Action to update
* \param[in] for_graph Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \note This may also schedule fencing if a stop is unrunnable.
*/
static void
update_resource_action_runnable(pe_action_t *action, bool for_graph,
pe_working_set_t *data_set)
{
if (pcmk_is_set(action->flags, pe_action_pseudo)) {
return;
}
if (action->node == NULL) {
pe_rsc_trace(action->rsc, "%s is unrunnable (unallocated)",
action->uuid);
pe__clear_action_flags(action, pe_action_runnable);
} else if (!pcmk_is_set(action->flags, pe_action_dc)
&& !(action->node->details->online)
&& (!pe__is_guest_node(action->node)
|| action->node->details->remote_requires_reset)) {
pe__clear_action_flags(action, pe_action_runnable);
do_crm_log((for_graph? LOG_WARNING: LOG_TRACE),
"%s on %s is unrunnable (node is offline)",
action->uuid, pe__node_name(action->node));
if (pcmk_is_set(action->rsc->flags, pe_rsc_managed)
&& for_graph
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)
&& !(action->node->details->unclean)) {
pe_fence_node(data_set, action->node, "stop is unrunnable", false);
}
} else if (!pcmk_is_set(action->flags, pe_action_dc)
&& action->node->details->pending) {
pe__clear_action_flags(action, pe_action_runnable);
do_crm_log((for_graph? LOG_WARNING: LOG_TRACE),
"Action %s on %s is unrunnable (node is pending)",
action->uuid, pe__node_name(action->node));
} else if (action->needs == pcmk_requires_nothing) {
pe_action_set_reason(action, NULL, TRUE);
if (pe__is_guest_node(action->node)
&& !pe_can_fence(data_set, action->node)) {
/* An action that requires nothing usually does not require any
* fencing in order to be runnable. However, there is an exception:
* such an action cannot be completed if it is on a guest node whose
* host is unclean and cannot be fenced.
*/
pe_rsc_debug(action->rsc, "%s on %s is unrunnable "
"(node's host cannot be fenced)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
} else {
pe_rsc_trace(action->rsc,
"%s on %s does not require fencing or quorum",
action->uuid, pe__node_name(action->node));
pe__set_action_flags(action, pe_action_runnable);
}
} else {
switch (effective_quorum_policy(action->rsc, data_set)) {
case pcmk_no_quorum_stop:
pe_rsc_debug(action->rsc, "%s on %s is unrunnable (no quorum)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
pe_action_set_reason(action, "no quorum", true);
break;
case pcmk_no_quorum_freeze:
if (!action->rsc->fns->active(action->rsc, TRUE)
|| (action->rsc->next_role > action->rsc->role)) {
pe_rsc_debug(action->rsc,
"%s on %s is unrunnable (no quorum)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
pe_action_set_reason(action, "quorum freeze", true);
}
break;
default:
//pe_action_set_reason(action, NULL, TRUE);
pe__set_action_flags(action, pe_action_runnable);
break;
}
}
}
/*!
* \internal
* \brief Update a resource object's flags for a new action on it
*
* \param[in,out] rsc Resource that action is for (if any)
* \param[in] action New action
*/
static void
update_resource_flags_for_action(pe_resource_t *rsc, const pe_action_t *action)
{
/* @COMPAT pe_rsc_starting and pe_rsc_stopping are not actually used
* within Pacemaker, and should be deprecated and eventually removed
*/
if (pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)) {
pe__set_resource_flags(rsc, pe_rsc_stopping);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)) {
if (pcmk_is_set(action->flags, pe_action_runnable)) {
pe__set_resource_flags(rsc, pe_rsc_starting);
} else {
pe__clear_resource_flags(rsc, pe_rsc_starting);
}
}
}
static bool
valid_stop_on_fail(const char *value)
{
return !pcmk__strcase_any_of(value, "standby", "demote", "stop", NULL);
}
/*!
* \internal
* \brief Validate (and possibly reset) resource action's on_fail meta-attribute
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Action name
* \param[in] action_config Action configuration XML from CIB (if any)
* \param[in,out] meta Table of action meta-attributes
*
* \return (Possibly new) value of on-fail meta-attribute
*/
static const char *
validate_on_fail(const pe_resource_t *rsc, const char *action_name,
const xmlNode *action_config, GHashTable *meta)
{
const char *name = NULL;
const char *role = NULL;
const char *interval_spec = NULL;
const char *value = g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
char *key = NULL;
char *new_value = NULL;
// Stop actions can only use certain on-fail values
if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)
&& !valid_stop_on_fail(value)) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s stop "
"action to default value because '%s' is not "
"allowed for stop", rsc->id, value);
g_hash_table_remove(meta, XML_OP_ATTR_ON_FAIL);
return NULL;
}
/* Demote actions default on-fail to the on-fail value for the first
* recurring monitor for the promoted role (if any).
*/
if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none)
&& (value == NULL)) {
/* @TODO This does not consider promote options set in a meta-attribute
* block (which may have rules that need to be evaluated) rather than
* XML properties.
*/
for (xmlNode *operation = first_named_child(rsc->ops_xml, XML_ATTR_OP);
operation != NULL; operation = crm_next_same_xml(operation)) {
bool enabled = false;
const char *promote_on_fail = NULL;
/* We only care about explicit on-fail (if promote uses default, so
* can demote)
*/
promote_on_fail = crm_element_value(operation, XML_OP_ATTR_ON_FAIL);
if (promote_on_fail == NULL) {
continue;
}
// We only care about recurring monitors for the promoted role
name = crm_element_value(operation, "name");
role = crm_element_value(operation, "role");
if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)) {
continue;
}
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (crm_parse_interval_spec(interval_spec) == 0) {
continue;
}
// We only care about enabled monitors
if ((pcmk__xe_get_bool_attr(operation, "enabled",
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
// Demote actions can't default to on-fail="demote"
if (pcmk__str_eq(promote_on_fail, "demote", pcmk__str_casei)) {
continue;
}
// Use value from first applicable promote action found
key = strdup(XML_OP_ATTR_ON_FAIL);
new_value = strdup(promote_on_fail);
CRM_ASSERT((key != NULL) && (new_value != NULL));
g_hash_table_insert(meta, key, new_value);
return g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
}
return NULL;
}
if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none)
&& !pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
key = strdup(XML_OP_ATTR_ON_FAIL);
new_value = strdup("ignore");
CRM_ASSERT((key != NULL) && (new_value != NULL));
g_hash_table_insert(meta, key, new_value);
return g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
}
// on-fail="demote" is allowed only for certain actions
if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
name = crm_element_value(action_config, "name");
role = crm_element_value(action_config, "role");
interval_spec = crm_element_value(action_config,
XML_LRM_ATTR_INTERVAL);
if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none)
&& (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)
|| (crm_parse_interval_spec(interval_spec) == 0))) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s %s "
"action to default value because 'demote' is not "
"allowed for it", rsc->id, name);
g_hash_table_remove(meta, XML_OP_ATTR_ON_FAIL);
return NULL;
}
}
return value;
}
static int
unpack_timeout(const char *value)
{
int timeout_ms = crm_get_msec(value);
if (timeout_ms < 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
return timeout_ms;
}
// true if value contains valid, non-NULL interval origin for recurring op
static bool
unpack_interval_origin(const char *value, const xmlNode *xml_obj,
guint interval_ms, const crm_time_t *now,
long long *start_delay)
{
long long result = 0;
guint interval_sec = interval_ms / 1000;
crm_time_t *origin = NULL;
// Ignore unspecified values and non-recurring operations
if ((value == NULL) || (interval_ms == 0) || (now == NULL)) {
return false;
}
// Parse interval origin from text
origin = crm_time_new(value);
if (origin == NULL) {
pcmk__config_err("Ignoring '" XML_OP_ATTR_ORIGIN "' for operation "
"'%s' because '%s' is not valid",
(ID(xml_obj)? ID(xml_obj) : "(missing ID)"), value);
return false;
}
// Get seconds since origin (negative if origin is in the future)
result = crm_time_get_seconds(now) - crm_time_get_seconds(origin);
crm_time_free(origin);
// Calculate seconds from closest interval to now
result = result % interval_sec;
// Calculate seconds remaining until next interval
result = ((result <= 0)? 0 : interval_sec) - result;
crm_info("Calculated a start delay of %llds for operation '%s'",
result,
(ID(xml_obj)? ID(xml_obj) : "(unspecified)"));
if (start_delay != NULL) {
*start_delay = result * 1000; // milliseconds
}
return true;
}
static int
unpack_start_delay(const char *value, GHashTable *meta)
{
int start_delay = 0;
if (value != NULL) {
start_delay = crm_get_msec(value);
if (start_delay < 0) {
start_delay = 0;
}
if (meta) {
g_hash_table_replace(meta, strdup(XML_OP_ATTR_START_DELAY),
pcmk__itoa(start_delay));
}
}
return start_delay;
}
static xmlNode *
find_min_interval_mon(pe_resource_t * rsc, gboolean include_disabled)
{
guint interval_ms = 0;
guint min_interval_ms = G_MAXUINT;
const char *name = NULL;
const char *interval_spec = NULL;
xmlNode *op = NULL;
xmlNode *operation = NULL;
for (operation = pcmk__xe_first_child(rsc->ops_xml);
operation != NULL;
operation = pcmk__xe_next(operation)) {
if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) {
bool enabled = false;
name = crm_element_value(operation, "name");
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (!include_disabled && pcmk__xe_get_bool_attr(operation, "enabled", &enabled) == pcmk_rc_ok &&
!enabled) {
continue;
}
if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
continue;
}
interval_ms = crm_parse_interval_spec(interval_spec);
if (interval_ms && (interval_ms < min_interval_ms)) {
min_interval_ms = interval_ms;
op = operation;
}
}
}
return op;
}
/*!
* \internal
* \brief Unpack action configuration
*
* Unpack a resource action's meta-attributes (normalizing the interval,
* timeout, and start delay values as integer milliseconds), requirements, and
* failure policy from its CIB XML configuration (including defaults).
*
* \param[in,out] action Resource action to unpack into
* \param[in] xml_obj Action configuration XML (NULL for defaults only)
* \param[in] container Resource that contains affected resource, if any
* \param[in] interval_ms How frequently to perform the operation
*/
static void
unpack_operation(pe_action_t *action, const xmlNode *xml_obj,
const pe_resource_t *container, guint interval_ms)
{
int timeout_ms = 0;
const char *value = NULL;
bool is_probe = false;
pe_rsc_eval_data_t rsc_rule_data = {
.standard = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_CLASS),
.provider = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_PROVIDER),
.agent = crm_element_value(action->rsc->xml, XML_EXPR_ATTR_TYPE)
};
pe_op_eval_data_t op_rule_data = {
.op_name = action->task,
.interval = interval_ms
};
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = action->rsc->cluster->now,
.match_data = NULL,
.rsc_data = &rsc_rule_data,
.op_data = &op_rule_data
};
CRM_CHECK(action && action->rsc, return);
is_probe = pcmk_is_probe(action->task, interval_ms);
// Cluster-wide <op_defaults> <meta_attributes>
pe__unpack_dataset_nvpairs(action->rsc->cluster->op_defaults,
XML_TAG_META_SETS, &rule_data, action->meta,
NULL, FALSE, action->rsc->cluster);
// Determine probe default timeout differently
if (is_probe) {
xmlNode *min_interval_mon = find_min_interval_mon(action->rsc, FALSE);
if (min_interval_mon) {
value = crm_element_value(min_interval_mon, XML_ATTR_TIMEOUT);
if (value) {
crm_trace("\t%s: Setting default timeout to minimum-interval "
"monitor's timeout '%s'", action->uuid, value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
strdup(value));
}
}
}
if (xml_obj) {
xmlAttrPtr xIter = NULL;
// <op> <meta_attributes> take precedence over defaults
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_META_SETS, &rule_data,
action->meta, NULL, TRUE,
action->rsc->cluster);
/* Anything set as an <op> XML property has highest precedence.
* This ensures we use the name and interval from the <op> tag.
*/
for (xIter = xml_obj->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
g_hash_table_replace(action->meta, strdup(prop_name), strdup(prop_value));
}
}
g_hash_table_remove(action->meta, "id");
// Normalize interval to milliseconds
if (interval_ms > 0) {
g_hash_table_replace(action->meta, strdup(XML_LRM_ATTR_INTERVAL),
crm_strdup_printf("%u", interval_ms));
} else {
g_hash_table_remove(action->meta, XML_LRM_ATTR_INTERVAL);
}
/*
* Timeout order of precedence:
* 1. pcmk_monitor_timeout (if rsc has pcmk_ra_cap_fence_params
* and task is start or a probe; pcmk_monitor_timeout works
* by default for a recurring monitor)
* 2. explicit op timeout on the primitive
* 3. default op timeout
* a. if probe, then min-interval monitor's timeout
* b. else, in XML_CIB_TAG_OPCONFIG
* 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS
*
* #1 overrides general rule of <op> XML property having highest
* precedence.
*/
if (pcmk_is_set(pcmk_get_ra_caps(rsc_rule_data.standard),
pcmk_ra_cap_fence_params)
&& (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)
|| is_probe)) {
GHashTable *params = pe_rsc_params(action->rsc, action->node,
action->rsc->cluster);
value = g_hash_table_lookup(params, "pcmk_monitor_timeout");
if (value) {
crm_trace("\t%s: Setting timeout to pcmk_monitor_timeout '%s', "
"overriding default", action->uuid, value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
strdup(value));
}
}
// Normalize timeout to positive milliseconds
value = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT);
timeout_ms = unpack_timeout(value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
pcmk__itoa(timeout_ms));
if (!pcmk__strcase_any_of(action->task, PCMK_ACTION_START,
PCMK_ACTION_PROMOTE, NULL)) {
action->needs = pcmk_requires_nothing;
value = "nothing (not start or promote)";
} else if (pcmk_is_set(action->rsc->flags, pe_rsc_needs_fencing)) {
action->needs = pcmk_requires_fencing;
value = "fencing";
} else if (pcmk_is_set(action->rsc->flags, pe_rsc_needs_quorum)) {
action->needs = pcmk_requires_quorum;
value = "quorum";
} else {
action->needs = pcmk_requires_nothing;
value = "nothing";
}
pe_rsc_trace(action->rsc, "%s requires %s", action->uuid, value);
value = validate_on_fail(action->rsc, action->task, xml_obj, action->meta);
if (value == NULL) {
} else if (pcmk__str_eq(value, "block", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_block;
} else if (pcmk__str_eq(value, "fence", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_fence_node;
value = "node fencing";
if (!pcmk_is_set(action->rsc->cluster->flags,
pe_flag_stonith_enabled)) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for "
"operation '%s' to 'stop' because 'fence' is not "
"valid when fencing is disabled", action->uuid);
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop resource";
}
} else if (pcmk__str_eq(value, "standby", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_standby_node;
value = "node standby";
} else if (pcmk__strcase_any_of(value, "ignore", PCMK__VALUE_NOTHING,
NULL)) {
action->on_fail = pcmk_on_fail_ignore;
value = "ignore";
} else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_ban;
value = "force migration";
} else if (pcmk__str_eq(value, "stop", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop resource";
} else if (pcmk__str_eq(value, "restart", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_restart;
value = "restart (and possibly migrate)";
} else if (pcmk__str_eq(value, "restart-container", pcmk__str_casei)) {
if (container) {
action->on_fail = pcmk_on_fail_restart_container;
value = "restart container (and possibly migrate)";
} else {
value = NULL;
}
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_demote;
value = "demote instance";
} else {
pe_err("Resource %s: Unknown failure type (%s)", action->rsc->id, value);
value = NULL;
}
/* defaults */
if (value == NULL && container) {
action->on_fail = pcmk_on_fail_restart_container;
value = "restart container (and possibly migrate) (default)";
/* For remote nodes, ensure that any failure that results in dropping an
* active connection to the node results in fencing of the node.
*
* There are only two action failures that don't result in fencing.
* 1. probes - probe failures are expected.
* 2. start - a start failure indicates that an active connection does not already
* exist. The user can set op on-fail=fence if they really want to fence start
* failures. */
} else if (((value == NULL) || !pcmk_is_set(action->rsc->flags, pe_rsc_managed))
&& pe__resource_is_remote_conn(action->rsc)
&& !(pcmk__str_eq(action->task, PCMK_ACTION_MONITOR,
pcmk__str_casei)
&& (interval_ms == 0))
&& !pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)) {
if (!pcmk_is_set(action->rsc->flags, pe_rsc_managed)) {
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop unmanaged remote node (enforcing default)";
} else {
if (pcmk_is_set(action->rsc->cluster->flags,
pe_flag_stonith_enabled)) {
value = "fence remote node (default)";
} else {
value = "recover remote node connection (default)";
}
if (action->rsc->remote_reconnect_ms) {
action->fail_role = pcmk_role_stopped;
}
action->on_fail = pcmk_on_fail_reset_remote;
}
} else if ((value == NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_casei)) {
if (pcmk_is_set(action->rsc->cluster->flags, pe_flag_stonith_enabled)) {
action->on_fail = pcmk_on_fail_fence_node;
value = "resource fence (default)";
} else {
action->on_fail = pcmk_on_fail_block;
value = "resource block (default)";
}
} else if (value == NULL) {
action->on_fail = pcmk_on_fail_restart;
value = "restart (and possibly migrate) (default)";
}
pe_rsc_trace(action->rsc, "%s failure handling: %s",
action->uuid, value);
value = NULL;
if (xml_obj != NULL) {
value = g_hash_table_lookup(action->meta, "role_after_failure");
if (value) {
pe_warn_once(pe_wo_role_after,
"Support for role_after_failure is deprecated and will be removed in a future release");
}
}
if (value != NULL && action->fail_role == pcmk_role_unknown) {
action->fail_role = text2role(value);
}
/* defaults */
if (action->fail_role == pcmk_role_unknown) {
if (pcmk__str_eq(action->task, PCMK_ACTION_PROMOTE, pcmk__str_casei)) {
action->fail_role = pcmk_role_unpromoted;
} else {
action->fail_role = pcmk_role_started;
}
}
pe_rsc_trace(action->rsc, "%s failure results in: %s",
action->uuid, role2text(action->fail_role));
value = g_hash_table_lookup(action->meta, XML_OP_ATTR_START_DELAY);
if (value) {
unpack_start_delay(value, action->meta);
} else {
long long start_delay = 0;
value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ORIGIN);
if (unpack_interval_origin(value, xml_obj, interval_ms,
action->rsc->cluster->now, &start_delay)) {
g_hash_table_replace(action->meta, strdup(XML_OP_ATTR_START_DELAY),
crm_strdup_printf("%lld", start_delay));
}
}
}
/*!
* \brief Create or update an action object
*
* \param[in,out] rsc Resource that action is for (if any)
* \param[in,out] key Action key (must be non-NULL)
* \param[in] task Action name (must be non-NULL)
* \param[in] on_node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in] save_action Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \return Action object corresponding to arguments
* \note This function takes ownership of (and might free) \p key. If
* \p save_action is true, \p data_set will own the returned action,
* otherwise it is the caller's responsibility to free the return value
* with pe_free_action().
*/
pe_action_t *
custom_action(pe_resource_t *rsc, char *key, const char *task,
const pe_node_t *on_node, gboolean optional, gboolean save_action,
pe_working_set_t *data_set)
{
pe_action_t *action = NULL;
CRM_ASSERT((key != NULL) && (task != NULL) && (data_set != NULL));
if (save_action) {
action = find_existing_action(key, rsc, on_node, data_set);
}
if (action == NULL) {
action = new_action(key, task, rsc, on_node, optional, save_action,
data_set);
} else {
free(key);
}
update_action_optional(action, optional);
if (rsc != NULL) {
if (action->node != NULL) {
unpack_action_node_attributes(action, data_set);
}
update_resource_action_runnable(action, save_action, data_set);
if (save_action) {
update_resource_flags_for_action(rsc, action);
}
}
return action;
}
pe_action_t *
get_pseudo_op(const char *name, pe_working_set_t * data_set)
{
pe_action_t *op = lookup_singleton(data_set, name);
if (op == NULL) {
op = custom_action(NULL, strdup(name), name, NULL, TRUE, TRUE, data_set);
pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable);
}
return op;
}
static GList *
find_unfencing_devices(GList *candidates, GList *matches)
{
for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) {
pe_resource_t *candidate = gIter->data;
if (candidate->children != NULL) {
matches = find_unfencing_devices(candidate->children, matches);
} else if (!pcmk_is_set(candidate->flags, pe_rsc_fence_device)) {
continue;
} else if (pcmk_is_set(candidate->flags, pe_rsc_needs_unfencing)) {
matches = g_list_prepend(matches, candidate);
} else if (pcmk__str_eq(g_hash_table_lookup(candidate->meta,
PCMK_STONITH_PROVIDES),
PCMK__VALUE_UNFENCING,
pcmk__str_casei)) {
matches = g_list_prepend(matches, candidate);
}
}
return matches;
}
static int
node_priority_fencing_delay(const pe_node_t *node,
const pe_working_set_t *data_set)
{
int member_count = 0;
int online_count = 0;
int top_priority = 0;
int lowest_priority = 0;
GList *gIter = NULL;
// `priority-fencing-delay` is disabled
if (data_set->priority_fencing_delay <= 0) {
return 0;
}
/* No need to request a delay if the fencing target is not a normal cluster
* member, for example if it's a remote node or a guest node. */
if (node->details->type != pcmk_node_variant_cluster) {
return 0;
}
// No need to request a delay if the fencing target is in our partition
if (node->details->online) {
return 0;
}
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *n = gIter->data;
if (n->details->type != pcmk_node_variant_cluster) {
continue;
}
member_count ++;
if (n->details->online) {
online_count++;
}
if (member_count == 1
|| n->details->priority > top_priority) {
top_priority = n->details->priority;
}
if (member_count == 1
|| n->details->priority < lowest_priority) {
lowest_priority = n->details->priority;
}
}
// No need to delay if we have more than half of the cluster members
if (online_count > member_count / 2) {
return 0;
}
/* All the nodes have equal priority.
* Any configured corresponding `pcmk_delay_base/max` will be applied. */
if (lowest_priority == top_priority) {
return 0;
}
if (node->details->priority < top_priority) {
return 0;
}
return data_set->priority_fencing_delay;
}
pe_action_t *
pe_fence_op(pe_node_t *node, const char *op, bool optional,
const char *reason, bool priority_delay, pe_working_set_t *data_set)
{
char *op_key = NULL;
pe_action_t *stonith_op = NULL;
if(op == NULL) {
op = data_set->stonith_action;
}
op_key = crm_strdup_printf("%s-%s-%s",
PCMK_ACTION_STONITH, node->details->uname, op);
stonith_op = lookup_singleton(data_set, op_key);
if(stonith_op == NULL) {
stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node,
TRUE, TRUE, data_set);
add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET, node->details->uname);
add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET_UUID, node->details->id);
add_hash_param(stonith_op->meta, "stonith_action", op);
if (pcmk_is_set(data_set->flags, pe_flag_enable_unfencing)) {
/* Extra work to detect device changes
*/
GString *digests_all = g_string_sized_new(1024);
GString *digests_secure = g_string_sized_new(1024);
GList *matches = find_unfencing_devices(data_set->resources, NULL);
char *key = NULL;
char *value = NULL;
for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) {
pe_resource_t *match = gIter->data;
const char *agent = g_hash_table_lookup(match->meta,
XML_ATTR_TYPE);
op_digest_cache_t *data = NULL;
data = pe__compare_fencing_digest(match, agent, node, data_set);
if(data->rc == RSC_DIGEST_ALL) {
optional = FALSE;
crm_notice("Unfencing node %s because the definition of "
"%s changed", pe__node_name(node), match->id);
if (!pcmk__is_daemon && data_set->priv != NULL) {
pcmk__output_t *out = data_set->priv;
out->info(out,
"notice: Unfencing node %s because the "
"definition of %s changed",
pe__node_name(node), match->id);
}
}
pcmk__g_strcat(digests_all,
match->id, ":", agent, ":",
data->digest_all_calc, ",", NULL);
pcmk__g_strcat(digests_secure,
match->id, ":", agent, ":",
data->digest_secure_calc, ",", NULL);
}
key = strdup(XML_OP_ATTR_DIGESTS_ALL);
value = strdup((const char *) digests_all->str);
CRM_ASSERT((key != NULL) && (value != NULL));
g_hash_table_insert(stonith_op->meta, key, value);
g_string_free(digests_all, TRUE);
key = strdup(XML_OP_ATTR_DIGESTS_SECURE);
value = strdup((const char *) digests_secure->str);
CRM_ASSERT((key != NULL) && (value != NULL));
g_hash_table_insert(stonith_op->meta, key, value);
g_string_free(digests_secure, TRUE);
}
} else {
free(op_key);
}
if (data_set->priority_fencing_delay > 0
/* It's a suitable case where `priority-fencing-delay` applies.
* At least add `priority-fencing-delay` field as an indicator. */
&& (priority_delay
/* The priority delay needs to be recalculated if this function has
* been called by schedule_fencing_and_shutdowns() after node
* priority has already been calculated by native_add_running().
*/
|| g_hash_table_lookup(stonith_op->meta,
XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY) != NULL)) {
/* Add `priority-fencing-delay` to the fencing op even if it's 0 for
* the targeting node. So that it takes precedence over any possible
* `pcmk_delay_base/max`.
*/
char *delay_s = pcmk__itoa(node_priority_fencing_delay(node, data_set));
g_hash_table_insert(stonith_op->meta,
strdup(XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY),
delay_s);
}
if(optional == FALSE && pe_can_fence(data_set, node)) {
pe__clear_action_flags(stonith_op, pe_action_optional);
pe_action_set_reason(stonith_op, reason, false);
} else if(reason && stonith_op->reason == NULL) {
stonith_op->reason = strdup(reason);
}
return stonith_op;
}
void
pe_free_action(pe_action_t * action)
{
if (action == NULL) {
return;
}
g_list_free_full(action->actions_before, free); /* pe_action_wrapper_t* */
g_list_free_full(action->actions_after, free); /* pe_action_wrapper_t* */
if (action->extra) {
g_hash_table_destroy(action->extra);
}
if (action->meta) {
g_hash_table_destroy(action->meta);
}
free(action->cancel_task);
free(action->reason);
free(action->task);
free(action->uuid);
free(action->node);
free(action);
}
int
pe_get_configured_timeout(pe_resource_t *rsc, const char *action, pe_working_set_t *data_set)
{
xmlNode *child = NULL;
GHashTable *action_meta = NULL;
const char *timeout_spec = NULL;
int timeout_ms = 0;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
for (child = first_named_child(rsc->ops_xml, XML_ATTR_OP);
child != NULL; child = crm_next_same_xml(child)) {
if (pcmk__str_eq(action, crm_element_value(child, XML_NVPAIR_ATTR_NAME),
pcmk__str_casei)) {
timeout_spec = crm_element_value(child, XML_ATTR_TIMEOUT);
break;
}
}
if (timeout_spec == NULL && data_set->op_defaults) {
action_meta = pcmk__strkey_table(free, free);
pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS,
&rule_data, action_meta, NULL, FALSE, data_set);
timeout_spec = g_hash_table_lookup(action_meta, XML_ATTR_TIMEOUT);
}
// @TODO check meta-attributes
// @TODO maybe use min-interval monitor timeout as default for monitors
timeout_ms = crm_get_msec(timeout_spec);
if (timeout_ms < 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
if (action_meta != NULL) {
g_hash_table_destroy(action_meta);
}
return timeout_ms;
}
enum action_tasks
get_complex_task(const pe_resource_t *rsc, const char *name)
{
enum action_tasks task = text2task(name);
if ((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)) {
switch (task) {
case pcmk_action_stopped:
case pcmk_action_started:
case pcmk_action_demoted:
case pcmk_action_promoted:
crm_trace("Folding %s back into its atomic counterpart for %s",
name, rsc->id);
--task;
break;
default:
break;
}
}
return task;
}
/*!
* \internal
* \brief Find first matching action in a list
*
* \param[in] input List of actions to search
* \param[in] uuid If not NULL, action must have this UUID
* \param[in] task If not NULL, action must have this action name
* \param[in] on_node If not NULL, action must be on this node
*
* \return First action in list that matches criteria, or NULL if none
*/
pe_action_t *
find_first_action(const GList *input, const char *uuid, const char *task,
const pe_node_t *on_node)
{
CRM_CHECK(uuid || task, return NULL);
for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) {
continue;
} else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
return action;
} else if (action->node == NULL) {
continue;
} else if (on_node->details == action->node->details) {
return action;
}
}
return NULL;
}
GList *
find_actions(GList *input, const char *key, const pe_node_t *on_node)
{
GList *gIter = input;
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
crm_trace("Action %s matches (ignoring node)", key);
result = g_list_prepend(result, action);
} else if (action->node == NULL) {
crm_trace("Action %s matches (unallocated, assigning to %s)",
key, pe__node_name(on_node));
action->node = pe__copy_node(on_node);
result = g_list_prepend(result, action);
} else if (on_node->details == action->node->details) {
crm_trace("Action %s on %s matches", key, pe__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
GList *
find_actions_exact(GList *input, const char *key, const pe_node_t *on_node)
{
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
if (on_node == NULL) {
return NULL;
}
for (GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if ((action->node != NULL)
&& pcmk__str_eq(key, action->uuid, pcmk__str_casei)
&& pcmk__str_eq(on_node->details->id, action->node->details->id,
pcmk__str_casei)) {
crm_trace("Action %s on %s matches", key, pe__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
/*!
* \brief Find all actions of given type for a resource
*
* \param[in] rsc Resource to search
* \param[in] node Find only actions scheduled on this node
* \param[in] task Action name to search for
* \param[in] require_node If TRUE, NULL node or action node will not match
*
* \return List of actions found (or NULL if none)
* \note If node is not NULL and require_node is FALSE, matching actions
* without a node will be assigned to node.
*/
GList *
pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node,
const char *task, bool require_node)
{
GList *result = NULL;
char *key = pcmk__op_key(rsc->id, task, 0);
if (require_node) {
result = find_actions_exact(rsc->actions, key, node);
} else {
result = find_actions(rsc->actions, key, node);
}
free(key);
return result;
}
/*!
* \internal
* \brief Create an action reason string based on the action itself
*
* \param[in] action Action to create reason string for
* \param[in] flag Action flag that was cleared
*
* \return Newly allocated string suitable for use as action reason
* \note It is the caller's responsibility to free() the result.
*/
char *
pe__action2reason(const pe_action_t *action, enum pe_action_flags flag)
{
const char *change = NULL;
switch (flag) {
case pe_action_runnable:
change = "unrunnable";
break;
case pe_action_migrate_runnable:
change = "unmigrateable";
break;
case pe_action_optional:
change = "required";
break;
default:
// Bug: caller passed unsupported flag
CRM_CHECK(change != NULL, change = "");
break;
}
return crm_strdup_printf("%s%s%s %s", change,
(action->rsc == NULL)? "" : " ",
(action->rsc == NULL)? "" : action->rsc->id,
action->task);
}
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite)
{
if (action->reason != NULL && overwrite) {
pe_rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'",
action->uuid, action->reason, pcmk__s(reason, "(none)"));
} else if (action->reason == NULL) {
pe_rsc_trace(action->rsc, "Set %s reason to '%s'",
action->uuid, pcmk__s(reason, "(none)"));
} else {
// crm_assert(action->reason != NULL && !overwrite);
return;
}
pcmk__str_update(&action->reason, reason);
}
/*!
* \internal
* \brief Create an action to clear a resource's history from CIB
*
* \param[in,out] rsc Resource to clear
* \param[in] node Node to clear history on
* \param[in,out] data_set Cluster working set
*
* \return New action to clear resource history
*/
pe_action_t *
pe__clear_resource_history(pe_resource_t *rsc, const pe_node_t *node,
pe_working_set_t *data_set)
{
char *key = NULL;
CRM_ASSERT(rsc && node);
key = pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0);
return custom_action(rsc, key, PCMK_ACTION_LRM_DELETE, node, FALSE, TRUE,
data_set);
}
#define sort_return(an_int, why) do { \
free(a_uuid); \
free(b_uuid); \
crm_trace("%s (%d) %c %s (%d) : %s", \
a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \
b_xml_id, b_call_id, why); \
return an_int; \
} while(0)
int
pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b,
bool same_node_default)
{
int a_call_id = -1;
int b_call_id = -1;
char *a_uuid = NULL;
char *b_uuid = NULL;
const char *a_xml_id = crm_element_value(xml_a, XML_ATTR_ID);
const char *b_xml_id = crm_element_value(xml_b, XML_ATTR_ID);
const char *a_node = crm_element_value(xml_a, XML_LRM_ATTR_TARGET);
const char *b_node = crm_element_value(xml_b, XML_LRM_ATTR_TARGET);
bool same_node = true;
/* @COMPAT The on_node attribute was added to last_failure as of 1.1.13 (via
* 8b3ca1c) and the other entries as of 1.1.12 (via 0b07b5c).
*
* In case that any of the lrm_rsc_op entries doesn't have on_node
* attribute, we need to explicitly tell whether the two operations are on
* the same node.
*/
if (a_node == NULL || b_node == NULL) {
same_node = same_node_default;
} else {
same_node = pcmk__str_eq(a_node, b_node, pcmk__str_casei);
}
if (same_node && pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_none)) {
/* We have duplicate lrm_rsc_op entries in the status
* section which is unlikely to be a good thing
* - we can handle it easily enough, but we need to get
* to the bottom of why it's happening.
*/
pe_err("Duplicate lrm_rsc_op entries named %s", a_xml_id);
sort_return(0, "duplicate");
}
crm_element_value_int(xml_a, XML_LRM_ATTR_CALLID, &a_call_id);
crm_element_value_int(xml_b, XML_LRM_ATTR_CALLID, &b_call_id);
if (a_call_id == -1 && b_call_id == -1) {
/* both are pending ops so it doesn't matter since
* stops are never pending
*/
sort_return(0, "pending");
} else if (same_node && a_call_id >= 0 && a_call_id < b_call_id) {
sort_return(-1, "call id");
} else if (same_node && b_call_id >= 0 && a_call_id > b_call_id) {
sort_return(1, "call id");
} else if (a_call_id >= 0 && b_call_id >= 0
&& (!same_node || a_call_id == b_call_id)) {
/*
* The op and last_failed_op are the same
* Order on last-rc-change
*/
time_t last_a = -1;
time_t last_b = -1;
crm_element_value_epoch(xml_a, XML_RSC_OP_LAST_CHANGE, &last_a);
crm_element_value_epoch(xml_b, XML_RSC_OP_LAST_CHANGE, &last_b);
crm_trace("rc-change: %lld vs %lld",
(long long) last_a, (long long) last_b);
if (last_a >= 0 && last_a < last_b) {
sort_return(-1, "rc-change");
} else if (last_b >= 0 && last_a > last_b) {
sort_return(1, "rc-change");
}
sort_return(0, "rc-change");
} else {
/* One of the inputs is a pending operation
* Attempt to use XML_ATTR_TRANSITION_MAGIC to determine its age relative to the other
*/
int a_id = -1;
int b_id = -1;
const char *a_magic = crm_element_value(xml_a, XML_ATTR_TRANSITION_MAGIC);
const char *b_magic = crm_element_value(xml_b, XML_ATTR_TRANSITION_MAGIC);
CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic"));
if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic a");
}
if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic b");
}
/* try to determine the relative age of the operation...
* some pending operations (e.g. a start) may have been superseded
* by a subsequent stop
*
* [a|b]_id == -1 means it's a shutdown operation and _always_ comes last
*/
if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) {
/*
* some of the logic in here may be redundant...
*
* if the UUID from the TE doesn't match then one better
* be a pending operation.
* pending operations don't survive between elections and joins
* because we query the LRM directly
*/
if (b_call_id == -1) {
sort_return(-1, "transition + call");
} else if (a_call_id == -1) {
sort_return(1, "transition + call");
}
} else if ((a_id >= 0 && a_id < b_id) || b_id == -1) {
sort_return(-1, "transition");
} else if ((b_id >= 0 && a_id > b_id) || a_id == -1) {
sort_return(1, "transition");
}
}
/* we should never end up here */
CRM_CHECK(FALSE, sort_return(0, "default"));
}
gint
sort_op_by_callid(gconstpointer a, gconstpointer b)
{
const xmlNode *xml_a = a;
const xmlNode *xml_b = b;
return pe__is_newer_op(xml_a, xml_b, true);
}
/*!
* \internal
* \brief Create a new pseudo-action for a resource
*
* \param[in,out] rsc Resource to create action for
* \param[in] task Action name
* \param[in] optional Whether action should be considered optional
* \param[in] runnable Whethe action should be considered runnable
*
* \return New action object corresponding to arguments
*/
pe_action_t *
pe__new_rsc_pseudo_action(pe_resource_t *rsc, const char *task, bool optional,
bool runnable)
{
pe_action_t *action = NULL;
CRM_ASSERT((rsc != NULL) && (task != NULL));
action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL,
optional, TRUE, rsc->cluster);
pe__set_action_flags(action, pe_action_pseudo);
if (runnable) {
pe__set_action_flags(action, pe_action_runnable);
}
return action;
}
/*!
* \internal
* \brief Add the expected result to an action
*
* \param[in,out] action Action to add expected result to
* \param[in] expected_result Expected result to add
*
* \note This is more efficient than calling add_hash_param().
*/
void
pe__add_action_expected_result(pe_action_t *action, int expected_result)
{
char *name = NULL;
CRM_ASSERT((action != NULL) && (action->meta != NULL));
name = strdup(XML_ATTR_TE_TARGET_RC);
CRM_ASSERT (name != NULL);
g_hash_table_insert(action->meta, name, pcmk__itoa(expected_result));
}
diff --git a/lib/pengine/status.c b/lib/pengine/status.c
index 2d1577e7e2..bd5a9a7bbd 100644
--- a/lib/pengine/status.c
+++ b/lib/pengine/status.c
@@ -1,483 +1,483 @@
/*
* Copyright 2004-2023 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 <sys/param.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <glib.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
/*!
* \brief Create a new working set
*
* \return New, initialized working set on success, else NULL (and set errno)
* \note Only pe_working_set_t objects created with this function (as opposed
* to statically declared or directly allocated) should be used with the
* functions in this library, to allow for future extensions to the
* data type. The caller is responsible for freeing the memory with
* pe_free_working_set() when the instance is no longer needed.
*/
pe_working_set_t *
pe_new_working_set(void)
{
pe_working_set_t *data_set = calloc(1, sizeof(pe_working_set_t));
if (data_set != NULL) {
set_working_set_defaults(data_set);
}
return data_set;
}
/*!
* \brief Free a working set
*
* \param[in,out] data_set Working set to free
*/
void
pe_free_working_set(pe_working_set_t *data_set)
{
if (data_set != NULL) {
pe_reset_working_set(data_set);
data_set->priv = NULL;
free(data_set);
}
}
/*
* Unpack everything
* At the end you'll have:
* - A list of nodes
* - A list of resources (each with any dependencies on other resources)
* - A list of constraints between resources and nodes
* - A list of constraints between start/stop actions
* - A list of nodes that need to be stonith'd
* - A list of nodes that need to be shutdown
* - A list of the possible stop/start actions (without dependencies)
*/
gboolean
cluster_status(pe_working_set_t * data_set)
{
xmlNode *section = NULL;
if ((data_set == NULL) || (data_set->input == NULL)) {
return FALSE;
}
crm_trace("Beginning unpack");
if (data_set->failed != NULL) {
free_xml(data_set->failed);
}
data_set->failed = create_xml_node(NULL, "failed-ops");
if (data_set->now == NULL) {
data_set->now = crm_time_new(NULL);
}
if (data_set->dc_uuid == NULL) {
data_set->dc_uuid = crm_element_value_copy(data_set->input,
XML_ATTR_DC_UUID);
}
if (pcmk__xe_attr_is_true(data_set->input, XML_ATTR_HAVE_QUORUM)) {
- pe__set_working_set_flags(data_set, pe_flag_have_quorum);
+ pe__set_working_set_flags(data_set, pcmk_sched_quorate);
} else {
- pe__clear_working_set_flags(data_set, pe_flag_have_quorum);
+ pe__clear_working_set_flags(data_set, pcmk_sched_quorate);
}
data_set->op_defaults = get_xpath_object("//" XML_CIB_TAG_OPCONFIG,
data_set->input, LOG_NEVER);
data_set->rsc_defaults = get_xpath_object("//" XML_CIB_TAG_RSCCONFIG,
data_set->input, LOG_NEVER);
section = get_xpath_object("//" XML_CIB_TAG_CRMCONFIG, data_set->input,
LOG_TRACE);
unpack_config(section, data_set);
if (!pcmk_any_flags_set(data_set->flags,
- pe_flag_quick_location|pe_flag_have_quorum)
+ pe_flag_quick_location|pcmk_sched_quorate)
&& (data_set->no_quorum_policy != pcmk_no_quorum_ignore)) {
crm_warn("Fencing and resource management disabled due to lack of quorum");
}
section = get_xpath_object("//" XML_CIB_TAG_NODES, data_set->input,
LOG_TRACE);
unpack_nodes(section, data_set);
section = get_xpath_object("//" XML_CIB_TAG_RESOURCES, data_set->input,
LOG_TRACE);
if (!pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
unpack_remote_nodes(section, data_set);
}
unpack_resources(section, data_set);
section = get_xpath_object("//" XML_CIB_TAG_TAGS, data_set->input,
LOG_NEVER);
unpack_tags(section, data_set);
if (!pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
section = get_xpath_object("//"XML_CIB_TAG_STATUS, data_set->input,
LOG_TRACE);
unpack_status(section, data_set);
}
if (!pcmk_is_set(data_set->flags, pe_flag_no_counts)) {
for (GList *item = data_set->resources; item != NULL;
item = item->next) {
((pe_resource_t *) (item->data))->fns->count(item->data);
}
crm_trace("Cluster resource count: %d (%d disabled, %d blocked)",
data_set->ninstances, data_set->disabled_resources,
data_set->blocked_resources);
}
pe__set_working_set_flags(data_set, pe_flag_have_status);
return TRUE;
}
/*!
* \internal
* \brief Free a list of pe_resource_t
*
* \param[in,out] resources List to free
*
* \note When a working set's resource list is freed, that includes the original
* storage for the uname and id of any Pacemaker Remote nodes in the
* working set's node list, so take care not to use those afterward.
* \todo Refactor pe_node_t to strdup() the node name.
*/
static void
pe_free_resources(GList *resources)
{
pe_resource_t *rsc = NULL;
GList *iterator = resources;
while (iterator != NULL) {
rsc = (pe_resource_t *) iterator->data;
iterator = iterator->next;
rsc->fns->free(rsc);
}
if (resources != NULL) {
g_list_free(resources);
}
}
static void
pe_free_actions(GList *actions)
{
GList *iterator = actions;
while (iterator != NULL) {
pe_free_action(iterator->data);
iterator = iterator->next;
}
if (actions != NULL) {
g_list_free(actions);
}
}
static void
pe_free_nodes(GList *nodes)
{
for (GList *iterator = nodes; iterator != NULL; iterator = iterator->next) {
pe_node_t *node = (pe_node_t *) iterator->data;
// Shouldn't be possible, but to be safe ...
if (node == NULL) {
continue;
}
if (node->details == NULL) {
free(node);
continue;
}
/* This is called after pe_free_resources(), which means that we can't
* use node->details->uname for Pacemaker Remote nodes.
*/
crm_trace("Freeing node %s", (pe__is_guest_or_remote_node(node)?
"(guest or remote)" : pe__node_name(node)));
if (node->details->attrs != NULL) {
g_hash_table_destroy(node->details->attrs);
}
if (node->details->utilization != NULL) {
g_hash_table_destroy(node->details->utilization);
}
if (node->details->digest_cache != NULL) {
g_hash_table_destroy(node->details->digest_cache);
}
g_list_free(node->details->running_rsc);
g_list_free(node->details->allocated_rsc);
free(node->details);
free(node);
}
if (nodes != NULL) {
g_list_free(nodes);
}
}
static void
pe__free_ordering(GList *constraints)
{
GList *iterator = constraints;
while (iterator != NULL) {
pe__ordering_t *order = iterator->data;
iterator = iterator->next;
free(order->lh_action_task);
free(order->rh_action_task);
free(order);
}
if (constraints != NULL) {
g_list_free(constraints);
}
}
static void
pe__free_location(GList *constraints)
{
GList *iterator = constraints;
while (iterator != NULL) {
pe__location_t *cons = iterator->data;
iterator = iterator->next;
g_list_free_full(cons->node_list_rh, free);
free(cons->id);
free(cons);
}
if (constraints != NULL) {
g_list_free(constraints);
}
}
/*!
* \brief Reset working set to default state without freeing it or constraints
*
* \param[in,out] data_set Working set to reset
*
* \deprecated This function is deprecated as part of the API;
* pe_reset_working_set() should be used instead.
*/
void
cleanup_calculations(pe_working_set_t * data_set)
{
if (data_set == NULL) {
return;
}
pe__clear_working_set_flags(data_set, pe_flag_have_status);
if (data_set->config_hash != NULL) {
g_hash_table_destroy(data_set->config_hash);
}
if (data_set->singletons != NULL) {
g_hash_table_destroy(data_set->singletons);
}
if (data_set->tickets) {
g_hash_table_destroy(data_set->tickets);
}
if (data_set->template_rsc_sets) {
g_hash_table_destroy(data_set->template_rsc_sets);
}
if (data_set->tags) {
g_hash_table_destroy(data_set->tags);
}
free(data_set->dc_uuid);
crm_trace("deleting resources");
pe_free_resources(data_set->resources);
crm_trace("deleting actions");
pe_free_actions(data_set->actions);
crm_trace("deleting nodes");
pe_free_nodes(data_set->nodes);
pe__free_param_checks(data_set);
g_list_free(data_set->stop_needed);
free_xml(data_set->graph);
crm_time_free(data_set->now);
free_xml(data_set->input);
free_xml(data_set->failed);
set_working_set_defaults(data_set);
CRM_CHECK(data_set->ordering_constraints == NULL,;
);
CRM_CHECK(data_set->placement_constraints == NULL,;
);
}
/*!
* \brief Reset a working set to default state without freeing it
*
* \param[in,out] data_set Working set to reset
*/
void
pe_reset_working_set(pe_working_set_t *data_set)
{
if (data_set == NULL) {
return;
}
crm_trace("Deleting %d ordering constraints",
g_list_length(data_set->ordering_constraints));
pe__free_ordering(data_set->ordering_constraints);
data_set->ordering_constraints = NULL;
crm_trace("Deleting %d location constraints",
g_list_length(data_set->placement_constraints));
pe__free_location(data_set->placement_constraints);
data_set->placement_constraints = NULL;
crm_trace("Deleting %d colocation constraints",
g_list_length(data_set->colocation_constraints));
g_list_free_full(data_set->colocation_constraints, free);
data_set->colocation_constraints = NULL;
crm_trace("Deleting %d ticket constraints",
g_list_length(data_set->ticket_constraints));
g_list_free_full(data_set->ticket_constraints, free);
data_set->ticket_constraints = NULL;
cleanup_calculations(data_set);
}
void
set_working_set_defaults(pe_working_set_t * data_set)
{
void *priv = data_set->priv;
memset(data_set, 0, sizeof(pe_working_set_t));
data_set->priv = priv;
data_set->order_id = 1;
data_set->action_id = 1;
data_set->no_quorum_policy = pcmk_no_quorum_stop;
data_set->flags = 0x0ULL;
pe__set_working_set_flags(data_set,
pe_flag_stop_rsc_orphans
|pe_flag_symmetric_cluster
|pe_flag_stop_action_orphans);
if (!strcmp(PCMK__CONCURRENT_FENCING_DEFAULT, "true")) {
pe__set_working_set_flags(data_set, pe_flag_concurrent_fencing);
}
}
pe_resource_t *
pe_find_resource(GList *rsc_list, const char *id)
{
return pe_find_resource_with_flags(rsc_list, id, pcmk_rsc_match_history);
}
pe_resource_t *
pe_find_resource_with_flags(GList *rsc_list, const char *id, enum pe_find flags)
{
GList *rIter = NULL;
for (rIter = rsc_list; id && rIter; rIter = rIter->next) {
pe_resource_t *parent = rIter->data;
pe_resource_t *match =
parent->fns->find_rsc(parent, id, NULL, flags);
if (match != NULL) {
return match;
}
}
crm_trace("No match for %s", id);
return NULL;
}
/*!
* \brief Find a node by name or ID in a list of nodes
*
* \param[in] nodes List of nodes (as pe_node_t*)
* \param[in] id If not NULL, ID of node to find
* \param[in] node_name If not NULL, name of node to find
*
* \return Node from \p nodes that matches \p id if any,
* otherwise node from \p nodes that matches \p uname if any,
* otherwise NULL
*/
pe_node_t *
pe_find_node_any(const GList *nodes, const char *id, const char *uname)
{
pe_node_t *match = NULL;
if (id != NULL) {
match = pe_find_node_id(nodes, id);
}
if ((match == NULL) && (uname != NULL)) {
match = pe_find_node(nodes, uname);
}
return match;
}
/*!
* \brief Find a node by ID in a list of nodes
*
* \param[in] nodes List of nodes (as pe_node_t*)
* \param[in] id ID of node to find
*
* \return Node from \p nodes that matches \p id if any, otherwise NULL
*/
pe_node_t *
pe_find_node_id(const GList *nodes, const char *id)
{
for (const GList *iter = nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
/* @TODO Whether node IDs should be considered case-sensitive should
* probably depend on the node type, so functionizing the comparison
* would be worthwhile
*/
if (pcmk__str_eq(node->details->id, id, pcmk__str_casei)) {
return node;
}
}
return NULL;
}
/*!
* \brief Find a node by name in a list of nodes
*
* \param[in] nodes List of nodes (as pe_node_t*)
* \param[in] node_name Name of node to find
*
* \return Node from \p nodes that matches \p node_name if any, otherwise NULL
*/
pe_node_t *
pe_find_node(const GList *nodes, const char *node_name)
{
for (const GList *iter = nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
if (pcmk__str_eq(node->details->uname, node_name, pcmk__str_casei)) {
return node;
}
}
return NULL;
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index d0fefad910..2f6f07abf6 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,4923 +1,4923 @@
/*
* Copyright 2004-2023 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 {
pe_resource_t *rsc; // Resource that history is for
pe_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(data_set, option, flag) do { \
const char *scf_value = pe_pref((data_set)->config_hash, (option)); \
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(data_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Working set", \
crm_system_name, (data_set)->flags, \
(flag), #flag); \
} else { \
(data_set)->flags = pcmk__clear_flags_as(__func__, __LINE__,\
LOG_TRACE, "Working set", \
crm_system_name, (data_set)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
xmlNode **last_failure,
enum action_fail_response *failed);
static void determine_remote_online_status(pe_working_set_t *data_set,
pe_node_t *this_node);
static void add_node_attrs(const xmlNode *xml_obj, pe_node_t *node,
bool overwrite, pe_working_set_t *data_set);
static void determine_online_status(const xmlNode *node_state,
pe_node_t *this_node,
pe_working_set_t *data_set);
static void unpack_node_lrm(pe_node_t *node, const xmlNode *xml,
pe_working_set_t *data_set);
// Bitmask for warnings we only want to print once
uint32_t pe_wo = 0;
static gboolean
is_dangling_guest_node(pe_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,
pe_rsc_orphan_container_filler)) {
return TRUE;
}
return FALSE;
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] data_set Current working set of cluster
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider `priority-fencing-delay`
*/
void
pe_fence_node(pe_working_set_t * data_set, pe_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)) {
pe_resource_t *rsc = node->details->remote_rsc->container;
if (!pcmk_is_set(rsc->flags, pe_rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pe_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 {
crm_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, pe_rsc_failed|pe_rsc_stop);
}
}
} 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,
pe_rsc_failed|pe_rsc_stop);
} else if (pe__is_remote_node(node)) {
pe_resource_t *rsc = node->details->remote_rsc;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_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;
crm_warn("Remote node %s %s: %s",
pe__node_name(node),
pe_can_fence(data_set, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply `priority-fencing-delay` for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, data_set);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pe__node_name(node),
pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean",
reason);
} else {
crm_warn("Cluster node %s %s: %s",
pe__node_name(node),
pe_can_fence(data_set, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, data_set);
}
}
// @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, pe_working_set_t *data_set)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(data_set->flags, flag)) {
result = xpath_search(data_set->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pe__set_working_set_flags(data_set, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode * config, pe_working_set_t * data_set)
{
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 = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
data_set->config_hash = config_hash;
pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash,
CIB_OPTIONS_FIRST, FALSE, data_set);
verify_pe_options(data_set->config_hash);
set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes);
if (!pcmk_is_set(data_set->flags, pe_flag_startup_probes)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and stonith-watchdog-timeout is nonzero");
pe__set_working_set_flags(data_set, pe_flag_have_stonith_resource);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set);
value = pe_pref(data_set->config_hash, "stonith-timeout");
data_set->stonith_timeout = (int) crm_parse_interval_spec(value);
crm_debug("STONITH timeout: %d", data_set->stonith_timeout);
set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled);
crm_debug("STONITH of failed nodes is %s",
pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)? "enabled" : "disabled");
data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action");
if (!strcmp(data_set->stonith_action, "poweroff")) {
pe_warn_once(pe_wo_poweroff,
"Support for stonith-action of 'poweroff' is deprecated "
"and will be removed in a future release (use 'off' instead)");
data_set->stonith_action = PCMK_ACTION_OFF;
}
crm_trace("STONITH will %s nodes", data_set->stonith_action);
set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing);
crm_debug("Concurrent fencing is %s",
pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)? "enabled" : "disabled");
value = pe_pref(data_set->config_hash,
XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY);
if (value) {
data_set->priority_fencing_delay = crm_parse_interval_spec(value) / 1000;
crm_trace("Priority fencing delay is %ds", data_set->priority_fencing_delay);
}
set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything);
crm_debug("Stop all active resources: %s",
pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_stop_everything)));
set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster);
if (pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pe_pref(data_set->config_hash, "no-quorum-policy");
if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
data_set->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) {
data_set->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
data_set->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) {
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
int do_panic = 0;
crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC,
&do_panic);
- if (do_panic || pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
+ if (do_panic || pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
data_set->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum");
data_set->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting no-quorum-policy to 'stop' because "
"fencing is disabled");
data_set->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
data_set->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (data_set->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(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans);
crm_trace("Orphan resources are %s",
pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)? "stopped" : "ignored");
set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans);
crm_trace("Orphan resource actions are %s",
pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)? "stopped" : "ignored");
value = pe_pref(data_set->config_hash, "remove-after-stop");
if (value != NULL) {
if (crm_is_true(value)) {
pe__set_working_set_flags(data_set, pe_flag_remove_after_stop);
#ifndef PCMK__COMPAT_2_0
pe_warn_once(pe_wo_remove_after,
"Support for the remove-after-stop cluster property is"
" deprecated and will be removed in a future release");
#endif
} else {
pe__clear_working_set_flags(data_set, pe_flag_remove_after_stop);
}
}
set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode);
crm_trace("Maintenance mode: %s",
pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_maintenance_mode)));
set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal);
crm_trace("Start failures are %s",
pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)? "always fatal" : "handled by failcount");
if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing);
}
if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(data_set);
data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy");
crm_trace("Placement strategy: %s", data_set->placement_strategy);
set_config_flag(data_set, "shutdown-lock", pe_flag_shutdown_lock);
crm_trace("Resources will%s be locked to cleanly shut down nodes",
(pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)? "" : " not"));
if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
value = pe_pref(data_set->config_hash,
XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT);
data_set->shutdown_lock = crm_parse_interval_spec(value) / 1000;
crm_trace("Shutdown locks expire after %us", data_set->shutdown_lock);
}
value = pe_pref(data_set->config_hash,
XML_CONFIG_ATTR_NODE_PENDING_TIMEOUT);
data_set->node_pending_timeout = crm_parse_interval_spec(value) / 1000;
crm_trace("Node pending timeout is %us", data_set->node_pending_timeout);
return TRUE;
}
pe_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pe_working_set_t * data_set)
{
pe_node_t *new_node = NULL;
if (pe_find_node(data_set->nodes, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
new_node = calloc(1, sizeof(pe_node_t));
if (new_node == NULL) {
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);
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 = data_set;
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(data_set, pe_flag_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);
}
pe_warn_once(pe_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);
data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_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(pe_working_set_t *data_set, pe_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(data_set->flags, pe_flag_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, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
pe_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, data_set);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(data_set, new_node);
add_node_attrs(xml_obj, new_node, FALSE, data_set);
crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME));
}
}
if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) {
crm_info("Creating a fake local node");
pe_create_node(data_set->localhost, data_set->localhost, NULL, 0,
data_set);
}
return TRUE;
}
static void
setup_container(pe_resource_t * rsc, pe_working_set_t * data_set)
{
const char *container_id = NULL;
if (rsc->children) {
g_list_foreach(rsc->children, (GFunc) setup_container, data_set);
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)) {
pe_resource_t *container = pe_find_resource(data_set->resources, container_id);
if (container) {
rsc->container = container;
pe__set_resource_flags(container, pe_rsc_is_container);
container->fillers = g_list_append(container->fillers, rsc);
pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id);
} else {
pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set)
{
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(data_set->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,
data_set);
}
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, data_set);
if (new_node_id && pe_find_node(data_set->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,
data_set);
}
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, data_set);
if (new_node_id && pe_find_node(data_set->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,
data_set);
}
}
}
}
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(pe_working_set_t *data_set, pe_resource_t *new_rsc)
{
pe_node_t *remote_node = NULL;
if (new_rsc->is_remote_node == FALSE) {
return;
}
if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pe_find_node(data_set->nodes, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pe_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(data_set, 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)
{
pe_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] data_set Where to put resource information
*
* \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, pe_working_set_t * data_set)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
data_set->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)) {
pe_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(data_set->template_rsc_sets, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
g_hash_table_insert(data_set->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,
data_set) == pcmk_rc_ok) {
data_set->resources = g_list_append(data_set->resources, new_rsc);
pe_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 = data_set->resources; gIter != NULL; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
setup_container(rsc, data_set);
link_rsc2remotenode(data_set, rsc);
}
data_set->resources = g_list_sort(data_set->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) {
/* Ignore */
} else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)
&& !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) {
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 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, pe_working_set_t * data_set)
{
xmlNode *xml_tag = NULL;
data_set->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(data_set->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, pe_working_set_t * data_set)
{
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pe_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(data_set->tickets, ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, data_set);
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, pe_working_set_t * data_set)
{
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, data_set);
}
return TRUE;
}
static void
unpack_handle_remote_attrs(pe_node_t *this_node, const xmlNode *state,
pe_working_set_t *data_set)
{
const char *resource_discovery_enabled = NULL;
const xmlNode *attrs = NULL;
pe_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, data_set);
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, pe_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(data_set->flags, pe_flag_stonith_enabled)) {
crm_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] data_set Cluster working set
*/
static void
unpack_transient_attributes(const xmlNode *state, pe_node_t *node,
pe_working_set_t *data_set)
{
const char *discovery = NULL;
const xmlNode *attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS,
FALSE);
add_node_attrs(attrs, node, TRUE, data_set);
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)) {
crm_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] data_set Cluster working set
*/
static void
unpack_node_state(const xmlNode *state, pe_working_set_t *data_set)
{
const char *id = NULL;
const char *uname = NULL;
pe_node_t *this_node = NULL;
id = crm_element_value(state, XML_ATTR_ID);
if (id == NULL) {
crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without "
XML_ATTR_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(data_set->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, data_set);
/* 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, data_set);
- if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)
+ if (!pcmk_is_set(data_set->flags, pcmk_sched_quorate)
&& this_node->details->online
&& (data_set->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(data_set, 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] data_set Cluster working set
*
* \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,
pe_working_set_t *data_set)
{
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);
pe_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(data_set->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.
*/
pe_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.
*/
pe_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL)
|| (!pcmk_is_set(data_set->flags, pe_flag_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(data_set->flags, pe_flag_stonith_enabled
|pe_flag_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(data_set, this_node);
unpack_handle_remote_attrs(this_node, state, data_set);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
this_node->details->unpacked = TRUE;
unpack_node_lrm(this_node, state, data_set);
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, pe_working_set_t * data_set)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (data_set->tickets == NULL) {
data_set->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, data_set);
} else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
unpack_node_state(state, data_set);
}
}
while (unpack_node_history(status, FALSE, data_set) == 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(data_set->flags, pe_flag_stonith_enabled),
data_set);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (data_set->stop_needed != NULL) {
for (GList *item = data_set->stop_needed; item; item = item->next) {
pe_resource_t *container = item->data;
pe_node_t *node = pe__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(data_set->stop_needed);
data_set->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 = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_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(data_set, this_node);
}
}
return TRUE;
}
static gboolean
determine_online_status_no_fencing(pe_working_set_t *data_set,
const xmlNode *node_state,
pe_node_t *this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE);
const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER);
const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
int member = false;
bool crmd_online = false;
long long when_member = 0;
long long when_online = 0;
// @COMPAT DCs < 2.1.7 use boolean instead of time for cluster membership
if (crm_str_to_boolean(in_cluster, &member) != 1) {
pcmk__scan_ll(in_cluster, &when_member, 0LL);
member = (when_member > 0) ? true : false;
}
if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) {
crmd_online = true;
} else if (pcmk__str_eq(is_peer, OFFLINESTATUS, pcmk__str_casei)) {
crmd_online = false;
} else {
pcmk__scan_ll(is_peer, &when_online, 0LL);
crmd_online = (when_online > 0) ? true : false;
}
if (!member) {
crm_trace("Node is down: in_cluster=%s",
pcmk__s(in_cluster, "<null>"));
} else if (crmd_online) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node is not ready to run resources: %s", join);
}
} else if (this_node->details->expected_up == FALSE) {
crm_trace("Controller is down: "
"in_cluster=%s is_peer=%s join=%s expected=%s",
pcmk__s(in_cluster, "<null>"), pcmk__s(is_peer, "<null>"),
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(data_set, this_node, "peer is unexpectedly down", FALSE);
crm_info("in_cluster=%s is_peer=%s join=%s expected=%s",
pcmk__s(in_cluster, "<null>"), pcmk__s(is_peer, "<null>"),
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
static gboolean
determine_online_status_fencing(pe_working_set_t *data_set,
const xmlNode *node_state, pe_node_t *this_node)
{
gboolean online = FALSE;
gboolean do_terminate = FALSE;
bool crmd_online = FALSE;
const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE);
const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER);
const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
const char *terminate = pe_node_attribute_raw(this_node, "terminate");
int member = false;
long long when_member = 0;
long long when_online = 0;
/*
- XML_NODE_JOIN_STATE ::= member|down|pending|banned
- XML_NODE_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- XML_NODE_IN_CLUSTER ::= true|false
- XML_NODE_IS_PEER ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- XML_NODE_IN_CLUSTER ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- XML_NODE_IS_PEER ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
if (crm_is_true(terminate)) {
do_terminate = TRUE;
} else if (terminate != NULL && strlen(terminate) > 0) {
/* could be a time() value */
char t = terminate[0];
if (t != '0' && isdigit(t)) {
do_terminate = TRUE;
}
}
crm_trace("%s: in_cluster=%s is_peer=%s join=%s expected=%s term=%d",
pe__node_name(this_node), pcmk__s(in_cluster, "<null>"),
pcmk__s(is_peer, "<null>"), pcmk__s(join, "<null>"),
pcmk__s(exp_state, "<null>"), do_terminate);
/* @COMPAT with boolean values of XML_NODE_IN_CLUSTER recorded for
* DCs < 2.1.7
*/
if (crm_str_to_boolean(in_cluster, &member) != 1) {
pcmk__scan_ll(in_cluster, &when_member, 0LL);
member = (when_member > 0) ? true : false;
}
online = member;
/* @COMPAT with "online"/"offline" values of XML_NODE_IS_PEER recorded for
* DCs < 2.1.7
*/
if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) {
crmd_online = true;
} else if (pcmk__str_eq(is_peer, OFFLINESTATUS, pcmk__str_casei)) {
crmd_online = false;
} else {
pcmk__scan_ll(is_peer, &when_online, 0LL);
crmd_online = (when_online > 0) ? true : false;
}
if (exp_state == NULL) {
exp_state = CRMD_JOINSTATE_DOWN;
}
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 */
online = crmd_online;
} else if (in_cluster == NULL) {
pe_fence_node(data_set, this_node, "peer has not been seen by the cluster", FALSE);
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_casei)) {
pe_fence_node(data_set, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (do_terminate == FALSE && pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_casei)) {
if (when_member > 0
&& when_online == 0
&& (get_effective_time(data_set) - when_member
>= data_set->node_pending_timeout)) {
pe_fence_node(data_set, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if (member || crmd_online) {
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 (do_terminate && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_casei)
&& !member && !crmd_online) {
crm_info("%s was just shot", pe__node_name(this_node));
online = FALSE;
} else if (!member) {
// Consider `priority-fencing-delay` for lost nodes
pe_fence_node(data_set, this_node, "peer is no longer part of the cluster", TRUE);
} else if (!crmd_online) {
pe_fence_node(data_set, this_node, "peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (do_terminate) {
pe_fence_node(data_set, this_node, "termination was requested", FALSE);
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
crm_info("%s is active", pe__node_name(this_node));
} else if (pcmk__strcase_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(data_set, this_node, "peer was in an unknown state", FALSE);
crm_warn("%s: in-cluster=%s is-peer=%s join=%s expected=%s term=%d shutdown=%d",
pe__node_name(this_node), pcmk__s(in_cluster, "<null>"),
pcmk__s(is_peer, "<null>"), pcmk__s(join, "<null>"),
pcmk__s(exp_state, "<null>"), do_terminate,
this_node->details->shutdown);
}
return online;
}
static void
determine_remote_online_status(pe_working_set_t * data_set, pe_node_t * this_node)
{
pe_resource_t *rsc = this_node->details->remote_rsc;
pe_resource_t *container = NULL;
pe_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, pe_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, pe_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, pe_node_t *this_node,
pe_working_set_t *data_set)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, XML_NODE_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(data_set->flags, pe_flag_stonith_enabled)) {
online = determine_online_status_no_fencing(data_set, node_state, this_node);
} else {
online = determine_online_status_fencing(data_set, 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) {
pe_proc_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 pe_resource_t *
create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry,
pe_working_set_t *data_set)
{
pe_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, data_set) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pe_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pe_find_node(data_set->nodes, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set);
}
link_rsc2remotenode(data_set, 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, pe_rsc_orphan_container_filler);
}
pe__set_resource_flags(rsc, pe_rsc_orphan);
data_set->resources = g_list_append(data_set->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] data_set Cluster working set
*
* \return Newly added orphaned instance of \p parent
*/
static pe_resource_t *
create_anonymous_orphan(pe_resource_t *parent, const char *rsc_id,
const pe_node_t *node, pe_working_set_t *data_set)
{
pe_resource_t *top = pe__create_clone_child(parent, data_set);
// find_rsc() because we might be a cloned group
pe_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pe_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] data_set Cluster information
* \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 (without instance)
*/
static pe_resource_t *
find_anonymous_clone(pe_working_set_t *data_set, const pe_node_t *node,
pe_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pe_resource_t *rsc = NULL;
pe_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, pe_rsc_unique));
// Check for active (or partially active, for cloned groups) instance
pe_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;
pe_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 data set 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 (((pe_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 {
pe_rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pe_rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pe_rsc_block)) {
// 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)) {
pe_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, pe_rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pe__is_guest_node(node)
&& !pe__is_universal_clone(parent, data_set)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, data_set);
pe_rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pe_resource_t *
unpack_find_resource(pe_working_set_t *data_set, const pe_node_t *node,
const char *rsc_id)
{
pe_resource_t *rsc = NULL;
pe_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(data_set->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);
pe_resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pe_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(data_set, 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);
pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pe__node_name(node), rsc->id,
(pcmk_is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : ""));
}
return rsc;
}
static pe_resource_t *
process_orphan_resource(const xmlNode *rsc_entry, const pe_node_t *node,
pe_working_set_t *data_set)
{
pe_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, data_set);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) {
pe__clear_resource_flags(rsc, pe_rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pe_rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", data_set);
}
return rsc;
}
static void
process_rsc_state(pe_resource_t * rsc, pe_node_t * node,
enum action_fail_response on_fail)
{
pe_node_t *tmpnode = NULL;
char *reason = NULL;
enum action_fail_response save_on_fail = pcmk_on_fail_ignore;
CRM_ASSERT(rsc);
pe_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) {
pe_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) {
pe_node_t *n = pe__copy_node(node);
pe_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, pe_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, pe_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, pe_rsc_failed|pe_rsc_stop);
should_fence = TRUE;
} else if (pcmk_is_set(rsc->cluster->flags, pe_flag_stonith_enabled)) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& !pcmk_is_set(node->details->remote_rsc->flags, pe_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, pe_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, pe_rsc_managed);
pe__set_resource_flags(rsc, pe_rsc_block);
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, pe_rsc_failed|pe_rsc_stop);
stop_action(rsc, node, FALSE);
}
break;
case pcmk_on_fail_restart_container:
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
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, pe_rsc_failed|pe_rsc_stop);
if (pcmk_is_set(rsc->cluster->flags, pe_flag_stonith_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, pe_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, pe_rsc_orphan)) {
if (pcmk_is_set(rsc->flags, pe_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, pe_rsc_failed);
break;
default:
pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop);
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
*/
pe_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) {
pe_action_t *stop = (pe_action_t *) gIter->data;
pe__set_action_flags(stop, pe_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(pe_node_t * node, pe_resource_t * rsc,
int start_index, int stop_index,
GList *sorted_op_list, pe_working_set_t * data_set)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
CRM_ASSERT(rsc);
pe_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) {
pe_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) {
pe_rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pe__node_name(node));
continue;
} else if (counter < start_index) {
pe_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) {
pe_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)) {
pe_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);
pe_rsc_trace(rsc, "Creating %s on %s", key, pe__node_name(node));
custom_action(rsc, key, task, node, TRUE, TRUE, data_set);
}
}
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, pe_resource_t *rsc,
const pe_node_t *node, pe_working_set_t *data_set)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((data_set->shutdown_lock > 0)
&& (get_effective_time(data_set)
> (lock_time + data_set->shutdown_lock))) {
pe_rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pe__node_name(node));
pe__clear_resource_history(rsc, node, data_set);
} 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 = (pe_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] data_set Cluster working set
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pe_resource_t *
unpack_lrm_resource(pe_node_t *node, const xmlNode *lrm_resource,
pe_working_set_t *data_set)
{
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);
pe_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) {
crm_warn("Ignoring malformed " XML_LRM_TAG_RESOURCE
" entry without 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(data_set->flags, pe_flag_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(data_set, 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, data_set);
}
}
CRM_ASSERT(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, data_set);
}
/* 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, data_set);
/* 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) {
pe_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,
pe_working_set_t *data_set)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list);
rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) {
pe_resource_t *rsc;
pe_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(data_set->resources, container_id);
if (container == NULL) {
continue;
}
rsc = pe_find_resource(data_set->resources, rsc_id);
if (rsc == NULL ||
!pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler) ||
rsc->container != NULL) {
continue;
}
pe_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] data_set Cluster working set
*/
static void
unpack_node_lrm(pe_node_t *node, const xmlNode *xml, pe_working_set_t *data_set)
{
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)) {
pe_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, data_set);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pe_rsc_orphan_container_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, data_set);
}
}
static void
set_active(pe_resource_t * rsc)
{
const pe_resource_t *top = pe__const_top_resource(rsc, false);
if (top && pcmk_is_set(top->flags, pe_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)
{
pe_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, pe_working_set_t *data_set)
{
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, data_set->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,
pe_working_set_t *data_set)
{
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, data_set->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(pe_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] data_set Cluster working set
*
* \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,
pe_working_set_t *data_set)
{
/* 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, data_set);
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] data_set Cluster working set
*
* \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,
pe_working_set_t *data_set)
{
xmlNode *lrm_resource = NULL;
lrm_resource = find_lrm_resource(rsc_id, node_name, data_set);
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] data_set Cluster working set
*
* \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,
pe_working_set_t *data_set)
{
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, data_set)
|| monitor_not_running_after(rsc_id, node_name, xml_op, same_node,
data_set);
}
/*!
* \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 pe_node_t *source_node,
const pe_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)) {
crm_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)) {
crm_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)) {
crm_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(pe_resource_t *rsc, const pe_node_t *node)
{
pe_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;
pe_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, pe_rsc_failed|pe_rsc_stop);
pe__clear_resource_flags(history->rsc, pe_rsc_allow_migrate);
}
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) {
pe_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, pe_rsc_failed|pe_rsc_stop);
pe__clear_resource_flags(history->rsc, pe_rsc_allow_migrate);
}
}
/*!
* \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).
*/
pe_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).
*/
pe_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(pe_resource_t *rsc)
{
int score = -INFINITY;
pe_resource_t *fail_rsc = rsc;
if (fail_rsc->parent != NULL) {
pe_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 Update resource role, failure handling, etc., after a failed action
*
* \param[in,out] history Parsed action result history
* \param[out] last_failure Set this to action XML
* \param[in,out] on_fail What should be done about the result
*/
static void
unpack_rsc_op_failure(struct action_history *history, xmlNode **last_failure,
enum action_fail_response *on_fail)
{
bool is_probe = false;
pe_action_t *action = NULL;
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, pe_flag_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 {
crm_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);
action = custom_action(history->rsc, strdup(history->key), history->task,
NULL, TRUE, FALSE, history->rsc->cluster);
if (cmp_on_fail(*on_fail, action->on_fail) < 0) {
pe_rsc_trace(history->rsc, "on-fail %s -> %s for %s (%s)",
fail2text(*on_fail), fail2text(action->on_fail),
action->uuid, history->key);
*on_fail = action->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 (action->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 */
pe_rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id);
history->rsc->role = pcmk_role_stopped;
} else if (history->rsc->role < pcmk_role_started) {
pe_rsc_trace(history->rsc, "Setting %s active", history->rsc->id);
set_active(history->rsc);
}
pe_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(action->on_fail), role2text(action->fail_role));
if ((action->fail_role != pcmk_role_started)
&& (history->rsc->next_role < action->fail_role)) {
pe__set_next_role(history->rsc, action->fail_role, "failure");
}
if (action->fail_role == pcmk_role_stopped) {
ban_from_all_nodes(history->rsc);
}
pe_free_action(action);
}
/*!
* \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);
pe_proc_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, pe_rsc_managed);
pe__set_resource_flags(history->rsc, pe_rsc_block);
}
/*!
* \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 data set'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");
crm_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");
pe_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");
pe_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, pe_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");
pe_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) {
pe_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,
pe_resource_t *rsc, pe_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 {
op_digest_cache_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->cluster);
switch (digest_data->rc) {
case RSC_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 RSC_DIGEST_MATCH:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pe_action_t *action, pe_resource_t *remote_conn,
pe_working_set_t *data_set)
{
pe_node_t *remote_node = pe_find_node(data_set->nodes, remote_conn->id);
if (remote_node) {
pe_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, data_set);
order_actions(fence, action, pe_order_implies_then);
}
}
static bool
should_ignore_failure_timeout(const pe_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, pe_flag_stonith_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pe_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) {
pe_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, pe_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);
}
}
if (expired) {
if (pe_get_failcount(history->node, history->rsc, NULL, pe_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.
*/
pe_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) {
// Schedule clearing of the fail count
pe_action_t *clear_op = pe__clear_failcount(history->rsc, history->node,
clear_reason,
history->rsc->cluster);
if (pcmk_is_set(history->rsc->cluster->flags, pe_flag_stonith_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
pe_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 Get the failure handling for an action
*
* \param[in,out] history Parsed action history entry
*
* \return Failure handling appropriate to action
*/
static enum action_fail_response
get_action_on_fail(struct action_history *history)
{
enum action_fail_response result = pcmk_on_fail_restart;
pe_action_t *action = custom_action(history->rsc, strdup(history->key),
history->task, NULL, TRUE, FALSE,
history->rsc->cluster);
result = action->on_fail;
pe_free_action(action);
return result;
}
/*!
* \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) {
pe_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:
pe_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)) {
crm_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.
*/
crm_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)
{
pe_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pe_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, pe_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;
pe_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(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum action_fail_response *on_fail)
{
int old_rc = 0;
bool expired = false;
pe_resource_t *parent = rsc;
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) {
crm_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) {
crm_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)) {
pe_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));
pe_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) {
pe_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, pe_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:
failure_strategy = get_action_on_fail(&history);
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 pe_rsc_failed later */
*on_fail = pcmk_on_fail_ban;
}
resource_location(parent, node, -INFINITY, "hard-error",
rsc->cluster);
unpack_rsc_op_failure(&history, 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, pe_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,
pe_rsc_failed|pe_rsc_stop);
}
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;
}
failure_strategy = get_action_on_fail(&history);
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, pe_rsc_failure_ignored);
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, 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) {
crm_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:
pe_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, pe_node_t *node, bool overwrite,
pe_working_set_t *data_set)
{
const char *cluster_name = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = data_set->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, data_set->dc_uuid, pcmk__str_casei)) {
data_set->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(data_set->config_hash, "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, data_set);
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data,
node->details->utilization, NULL,
FALSE, data_set);
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,
pe_working_set_t * data_set)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE);
pe_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(data_set->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(data_set, this_node);
} else {
determine_online_status(node_state, this_node, data_set);
}
if (this_node->details->online
|| pcmk_is_set(data_set->flags, pe_flag_stonith_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;
}
diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c
index e244f70b09..9f4818425d 100644
--- a/lib/pengine/utils.c
+++ b/lib/pengine/utils.c
@@ -1,900 +1,900 @@
/*
* Copyright 2004-2023 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 <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
extern bool pcmk__is_daemon;
gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data);
/*!
* \internal
* \brief Check whether we can fence a particular node
*
* \param[in] data_set Working set for cluster
* \param[in] node Name of node to check
*
* \return true if node can be fenced, false otherwise
*/
bool
pe_can_fence(const pe_working_set_t *data_set, const pe_node_t *node)
{
if (pe__is_guest_node(node)) {
/* Guest nodes are fenced by stopping their container resource. We can
* do that if the container's host is either online or fenceable.
*/
pe_resource_t *rsc = node->details->remote_rsc->container;
for (GList *n = rsc->running_on; n != NULL; n = n->next) {
pe_node_t *container_node = n->data;
if (!container_node->details->online
&& !pe_can_fence(data_set, container_node)) {
return false;
}
}
return true;
} else if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) {
return false; /* Turned off */
} else if (!pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) {
return false; /* No devices */
- } else if (pcmk_is_set(data_set->flags, pe_flag_have_quorum)) {
+ } else if (pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
return true;
} else if (data_set->no_quorum_policy == pcmk_no_quorum_ignore) {
return true;
} else if(node == NULL) {
return false;
} else if(node->details->online) {
crm_notice("We can fence %s without quorum because they're in our membership",
pe__node_name(node));
return true;
}
crm_trace("Cannot fence %s", pe__node_name(node));
return false;
}
/*!
* \internal
* \brief Copy a node object
*
* \param[in] this_node Node object to copy
*
* \return Newly allocated shallow copy of this_node
* \note This function asserts on errors and is guaranteed to return non-NULL.
*/
pe_node_t *
pe__copy_node(const pe_node_t *this_node)
{
pe_node_t *new_node = NULL;
CRM_ASSERT(this_node != NULL);
new_node = calloc(1, sizeof(pe_node_t));
CRM_ASSERT(new_node != NULL);
new_node->rsc_discover_mode = this_node->rsc_discover_mode;
new_node->weight = this_node->weight;
new_node->fixed = this_node->fixed; // @COMPAT deprecated and unused
new_node->count = this_node->count;
new_node->details = this_node->details;
return new_node;
}
/*!
* \internal
* \brief Create a node hash table from a node list
*
* \param[in] list Node list
*
* \return Hash table equivalent of node list
*/
GHashTable *
pe__node_list2table(const GList *list)
{
GHashTable *result = NULL;
result = pcmk__strkey_table(NULL, free);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
pe_node_t *new_node = pe__copy_node((const pe_node_t *) gIter->data);
g_hash_table_insert(result, (gpointer) new_node->details->id, new_node);
}
return result;
}
/*!
* \internal
* \brief Compare two nodes by name, with numeric portions sorted numerically
*
* Sort two node names case-insensitively like strcasecmp(), but with any
* numeric portions of the name sorted numerically. For example, "node10" will
* sort higher than "node9" but lower than "remotenode9".
*
* \param[in] a First node to compare (can be \c NULL)
* \param[in] b Second node to compare (can be \c NULL)
*
* \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not)
* \retval 0 \c a and \c b are equal (or both are \c NULL)
* \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not)
*/
gint
pe__cmp_node_name(gconstpointer a, gconstpointer b)
{
const pe_node_t *node1 = (const pe_node_t *) a;
const pe_node_t *node2 = (const pe_node_t *) b;
if ((node1 == NULL) && (node2 == NULL)) {
return 0;
}
if (node1 == NULL) {
return -1;
}
if (node2 == NULL) {
return 1;
}
return pcmk__numeric_strcasecmp(node1->details->uname,
node2->details->uname);
}
/*!
* \internal
* \brief Output node weights to stdout
*
* \param[in] rsc Use allowed nodes for this resource
* \param[in] comment Text description to prefix lines with
* \param[in] nodes If rsc is not specified, use these nodes
* \param[in,out] data_set Cluster working set
*/
static void
pe__output_node_weights(const pe_resource_t *rsc, const char *comment,
GHashTable *nodes, pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
// Sort the nodes so the output is consistent for regression tests
GList *list = g_list_sort(g_hash_table_get_values(nodes),
pe__cmp_node_name);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
const pe_node_t *node = (const pe_node_t *) gIter->data;
out->message(out, "node-weight", rsc, comment, node->details->uname,
pcmk_readable_score(node->weight));
}
g_list_free(list);
}
/*!
* \internal
* \brief Log node weights at trace level
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] rsc If not NULL, include this resource's ID in logs
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be logged
*/
static void
pe__log_node_weights(const char *file, const char *function, int line,
const pe_resource_t *rsc, const char *comment,
GHashTable *nodes)
{
GHashTableIter iter;
pe_node_t *node = NULL;
// Don't waste time if we're not tracing at this point
pcmk__if_tracing({}, return);
g_hash_table_iter_init(&iter, nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (rsc) {
qb_log_from_external_source(function, file,
"%s: %s allocation score on %s: %s",
LOG_TRACE, line, 0,
comment, rsc->id,
pe__node_name(node),
pcmk_readable_score(node->weight));
} else {
qb_log_from_external_source(function, file, "%s: %s = %s",
LOG_TRACE, line, 0,
comment, pe__node_name(node),
pcmk_readable_score(node->weight));
}
}
}
/*!
* \internal
* \brief Log or output node weights
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] to_log Log if true, otherwise output
* \param[in] rsc If not NULL, use this resource's ID in logs,
* and show scores recursively for any children
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be shown
* \param[in,out] data_set Cluster working set
*/
void
pe__show_node_scores_as(const char *file, const char *function, int line,
bool to_log, const pe_resource_t *rsc,
const char *comment, GHashTable *nodes,
pe_working_set_t *data_set)
{
if (rsc != NULL && pcmk_is_set(rsc->flags, pe_rsc_orphan)) {
// Don't show allocation scores for orphans
return;
}
if (nodes == NULL) {
// Nothing to show
return;
}
if (to_log) {
pe__log_node_weights(file, function, line, rsc, comment, nodes);
} else {
pe__output_node_weights(rsc, comment, nodes, data_set);
}
// If this resource has children, repeat recursively for each
if (rsc && rsc->children) {
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child = (pe_resource_t *) gIter->data;
pe__show_node_scores_as(file, function, line, to_log, child,
comment, child->allowed_nodes, data_set);
}
}
}
/*!
* \internal
* \brief Compare two resources by priority
*
* \param[in] a First resource to compare (can be \c NULL)
* \param[in] b Second resource to compare (can be \c NULL)
*
* \retval -1 \c a->priority > \c b->priority (or \c b is \c NULL and \c a is
* not)
* \retval 0 \c a->priority == \c b->priority (or both \c a and \c b are
* \c NULL)
* \retval 1 \c a->priority < \c b->priority (or \c a is \c NULL and \c b is
* not)
*/
gint
pe__cmp_rsc_priority(gconstpointer a, gconstpointer b)
{
const pe_resource_t *resource1 = (const pe_resource_t *)a;
const pe_resource_t *resource2 = (const pe_resource_t *)b;
if (a == NULL && b == NULL) {
return 0;
}
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (resource1->priority > resource2->priority) {
return -1;
}
if (resource1->priority < resource2->priority) {
return 1;
}
return 0;
}
static void
resource_node_score(pe_resource_t *rsc, const pe_node_t *node, int score,
const char *tag)
{
pe_node_t *match = NULL;
if ((rsc->exclusive_discover || (node->rsc_discover_mode == pe_discover_never))
&& pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) {
/* This string comparision may be fragile, but exclusive resources and
* exclusive nodes should not have the symmetric_default constraint
* applied to them.
*/
return;
} else if (rsc->children) {
GList *gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) gIter->data;
resource_node_score(child_rsc, node, score, tag);
}
}
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (match == NULL) {
match = pe__copy_node(node);
g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match);
}
match->weight = pcmk__add_scores(match->weight, score);
pe_rsc_trace(rsc,
"Enabling %s preference (%s) for %s on %s (now %s)",
tag, pcmk_readable_score(score), rsc->id, pe__node_name(node),
pcmk_readable_score(match->weight));
}
void
resource_location(pe_resource_t *rsc, const pe_node_t *node, int score,
const char *tag, pe_working_set_t *data_set)
{
if (node != NULL) {
resource_node_score(rsc, node, score, tag);
} else if (data_set != NULL) {
GList *gIter = data_set->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pe_node_t *node_iter = (pe_node_t *) gIter->data;
resource_node_score(rsc, node_iter, score, tag);
}
} else {
GHashTableIter iter;
pe_node_t *node_iter = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) {
resource_node_score(rsc, node_iter, score, tag);
}
}
if (node == NULL && score == -INFINITY) {
if (rsc->allocated_to) {
crm_info("Deallocating %s from %s",
rsc->id, pe__node_name(rsc->allocated_to));
free(rsc->allocated_to);
rsc->allocated_to = NULL;
}
}
}
time_t
get_effective_time(pe_working_set_t * data_set)
{
if(data_set) {
if (data_set->now == NULL) {
crm_trace("Recording a new 'now'");
data_set->now = crm_time_new(NULL);
}
return crm_time_get_seconds_since_epoch(data_set->now);
}
crm_trace("Defaulting to 'now'");
return time(NULL);
}
gboolean
get_target_role(const pe_resource_t *rsc, enum rsc_role_e *role)
{
enum rsc_role_e local_role = pcmk_role_unknown;
const char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
CRM_CHECK(role != NULL, return FALSE);
if (pcmk__str_eq(value, "started", pcmk__str_null_matches | pcmk__str_casei)
|| pcmk__str_eq("default", value, pcmk__str_casei)) {
return FALSE;
}
local_role = text2role(value);
if (local_role == pcmk_role_unknown) {
pcmk__config_err("Ignoring '" XML_RSC_ATTR_TARGET_ROLE "' for %s "
"because '%s' is not valid", rsc->id, value);
return FALSE;
} else if (local_role > pcmk_role_started) {
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pe_rsc_promotable)) {
if (local_role > pcmk_role_unpromoted) {
/* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */
return FALSE;
}
} else {
pcmk__config_err("Ignoring '" XML_RSC_ATTR_TARGET_ROLE "' for %s "
"because '%s' only makes sense for promotable "
"clones", rsc->id, value);
return FALSE;
}
}
*role = local_role;
return TRUE;
}
gboolean
order_actions(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order)
{
GList *gIter = NULL;
pe_action_wrapper_t *wrapper = NULL;
GList *list = NULL;
if (order == pe_order_none) {
return FALSE;
}
if (lh_action == NULL || rh_action == NULL) {
return FALSE;
}
crm_trace("Creating action wrappers for ordering: %s then %s",
lh_action->uuid, rh_action->uuid);
/* Ensure we never create a dependency on ourselves... it's happened */
CRM_ASSERT(lh_action != rh_action);
/* Filter dups, otherwise update_action_states() has too much work to do */
gIter = lh_action->actions_after;
for (; gIter != NULL; gIter = gIter->next) {
pe_action_wrapper_t *after = (pe_action_wrapper_t *) gIter->data;
if (after->action == rh_action && (after->type & order)) {
return FALSE;
}
}
wrapper = calloc(1, sizeof(pe_action_wrapper_t));
wrapper->action = rh_action;
wrapper->type = order;
list = lh_action->actions_after;
list = g_list_prepend(list, wrapper);
lh_action->actions_after = list;
wrapper = calloc(1, sizeof(pe_action_wrapper_t));
wrapper->action = lh_action;
wrapper->type = order;
list = rh_action->actions_before;
list = g_list_prepend(list, wrapper);
rh_action->actions_before = list;
return TRUE;
}
void
destroy_ticket(gpointer data)
{
pe_ticket_t *ticket = data;
if (ticket->state) {
g_hash_table_destroy(ticket->state);
}
free(ticket->id);
free(ticket);
}
pe_ticket_t *
ticket_new(const char *ticket_id, pe_working_set_t * data_set)
{
pe_ticket_t *ticket = NULL;
if (pcmk__str_empty(ticket_id)) {
return NULL;
}
if (data_set->tickets == NULL) {
data_set->tickets = pcmk__strkey_table(free, destroy_ticket);
}
ticket = g_hash_table_lookup(data_set->tickets, ticket_id);
if (ticket == NULL) {
ticket = calloc(1, sizeof(pe_ticket_t));
if (ticket == NULL) {
crm_err("Cannot allocate ticket '%s'", ticket_id);
return NULL;
}
crm_trace("Creaing ticket entry for %s", ticket_id);
ticket->id = strdup(ticket_id);
ticket->granted = FALSE;
ticket->last_granted = -1;
ticket->standby = FALSE;
ticket->state = pcmk__strkey_table(free, free);
g_hash_table_insert(data_set->tickets, strdup(ticket->id), ticket);
}
return ticket;
}
const char *
rsc_printable_id(const pe_resource_t *rsc)
{
return pcmk_is_set(rsc->flags, pe_rsc_unique)? rsc->id : ID(rsc->xml);
}
void
pe__clear_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags)
{
pe__clear_resource_flags(rsc, flags);
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe__clear_resource_flags_recursive((pe_resource_t *) gIter->data, flags);
}
}
void
pe__clear_resource_flags_on_all(pe_working_set_t *data_set, uint64_t flag)
{
for (GList *lpc = data_set->resources; lpc != NULL; lpc = lpc->next) {
pe_resource_t *r = (pe_resource_t *) lpc->data;
pe__clear_resource_flags_recursive(r, flag);
}
}
void
pe__set_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags)
{
pe__set_resource_flags(rsc, flags);
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe__set_resource_flags_recursive((pe_resource_t *) gIter->data, flags);
}
}
void
trigger_unfencing(pe_resource_t *rsc, pe_node_t *node, const char *reason,
pe_action_t *dependency, pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pe_flag_enable_unfencing)) {
/* No resources require it */
return;
} else if ((rsc != NULL)
&& !pcmk_is_set(rsc->flags, pe_rsc_fence_device)) {
/* Wasn't a stonith device */
return;
} else if(node
&& node->details->online
&& node->details->unclean == FALSE
&& node->details->shutdown == FALSE) {
pe_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE, reason,
FALSE, data_set);
if(dependency) {
order_actions(unfence, dependency, pe_order_optional);
}
} else if(rsc) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) {
trigger_unfencing(rsc, node, reason, dependency, data_set);
}
}
}
}
gboolean
add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref)
{
pe_tag_t *tag = NULL;
GList *gIter = NULL;
gboolean is_existing = FALSE;
CRM_CHECK(tags && tag_name && obj_ref, return FALSE);
tag = g_hash_table_lookup(tags, tag_name);
if (tag == NULL) {
tag = calloc(1, sizeof(pe_tag_t));
if (tag == NULL) {
return FALSE;
}
tag->id = strdup(tag_name);
tag->refs = NULL;
g_hash_table_insert(tags, strdup(tag_name), tag);
}
for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) {
const char *existing_ref = (const char *) gIter->data;
if (pcmk__str_eq(existing_ref, obj_ref, pcmk__str_none)){
is_existing = TRUE;
break;
}
}
if (is_existing == FALSE) {
tag->refs = g_list_append(tag->refs, strdup(obj_ref));
crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref);
}
return TRUE;
}
/*!
* \internal
* \brief Check whether shutdown has been requested for a node
*
* \param[in] node Node to check
*
* \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise
* \note This differs from simply using node->details->shutdown in that it can
* be used before that has been determined (and in fact to determine it),
* and it can also be used to distinguish requested shutdown from implicit
* shutdown of remote nodes by virtue of their connection stopping.
*/
bool
pe__shutdown_requested(const pe_node_t *node)
{
const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches);
}
/*!
* \internal
* \brief Update a data set's "recheck by" time
*
* \param[in] recheck Epoch time when recheck should happen
* \param[in,out] data_set Current working set
*/
void
pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set)
{
if ((recheck > get_effective_time(data_set))
&& ((data_set->recheck_by == 0)
|| (data_set->recheck_by > recheck))) {
data_set->recheck_by = recheck;
}
}
/*!
* \internal
* \brief Extract nvpair blocks contained by a CIB XML element into a hash table
*
* \param[in] xml_obj XML element containing blocks of nvpair elements
* \param[in] set_name If not NULL, only use blocks of this element
* \param[in] rule_data Matching parameters to use when unpacking
* \param[out] hash Where to store extracted name/value pairs
* \param[in] always_first If not NULL, process block with this ID first
* \param[in] overwrite Whether to replace existing values with same name
* \param[in,out] data_set Cluster working set containing \p xml_obj
*/
void
pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name,
const pe_rule_eval_data_t *rule_data,
GHashTable *hash, const char *always_first,
gboolean overwrite, pe_working_set_t *data_set)
{
crm_time_t *next_change = crm_time_new_undefined();
pe_eval_nvpairs(data_set->input, xml_obj, set_name, rule_data, hash,
always_first, overwrite, next_change);
if (crm_time_is_defined(next_change)) {
time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change);
pe__update_recheck_time(recheck, data_set);
}
crm_time_free(next_change);
}
bool
pe__resource_is_disabled(const pe_resource_t *rsc)
{
const char *target_role = NULL;
CRM_CHECK(rsc != NULL, return false);
target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE);
if (target_role) {
enum rsc_role_e target_role_e = text2role(target_role);
if ((target_role_e == pcmk_role_stopped)
|| ((target_role_e == pcmk_role_unpromoted)
&& pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pe_rsc_promotable))) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a resource is running only on given node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p rsc is running only on \p node, otherwise false
*/
bool
pe__rsc_running_on_only(const pe_resource_t *rsc, const pe_node_t *node)
{
return (rsc != NULL) && pcmk__list_of_1(rsc->running_on)
&& pe__same_node((const pe_node_t *) rsc->running_on->data, node);
}
bool
pe__rsc_running_on_any(pe_resource_t *rsc, GList *node_list)
{
for (GList *ele = rsc->running_on; ele; ele = ele->next) {
pe_node_t *node = (pe_node_t *) ele->data;
if (pcmk__str_in_list(node->details->uname, node_list,
pcmk__str_star_matches|pcmk__str_casei)) {
return true;
}
}
return false;
}
bool
pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GList *only_node)
{
return (rsc->fns->active(rsc, FALSE) && !pe__rsc_running_on_any(rsc, only_node));
}
GList *
pe__filter_rsc_list(GList *rscs, GList *filter)
{
GList *retval = NULL;
for (GList *gIter = rscs; gIter; gIter = gIter->next) {
pe_resource_t *rsc = (pe_resource_t *) gIter->data;
/* I think the second condition is safe here for all callers of this
* function. If not, it needs to move into pe__node_text.
*/
if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) ||
(rsc->parent && pcmk__str_in_list(rsc_printable_id(rsc->parent), filter, pcmk__str_star_matches))) {
retval = g_list_prepend(retval, rsc);
}
}
return retval;
}
GList *
pe__build_node_name_list(pe_working_set_t *data_set, const char *s) {
GList *nodes = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
/* Nothing was given so return a list of all node names. Or, '*' was
* given. This would normally fall into the pe__unames_with_tag branch
* where it will return an empty list. Catch it here instead.
*/
nodes = g_list_prepend(nodes, strdup("*"));
} else {
pe_node_t *node = pe_find_node(data_set->nodes, s);
if (node) {
/* The given string was a valid uname for a node. Return a
* singleton list containing just that uname.
*/
nodes = g_list_prepend(nodes, strdup(s));
} else {
/* The given string was not a valid uname. It's either a tag or
* it's a typo or something. In the first case, we'll return a
* list of all the unames of the nodes with the given tag. In the
* second case, we'll return a NULL pointer and nothing will
* get displayed.
*/
nodes = pe__unames_with_tag(data_set, s);
}
}
return nodes;
}
GList *
pe__build_rsc_list(pe_working_set_t *data_set, const char *s) {
GList *resources = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
resources = g_list_prepend(resources, strdup("*"));
} else {
const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename;
pe_resource_t *rsc = pe_find_resource_with_flags(data_set->resources, s,
flags);
if (rsc) {
/* A colon in the name we were given means we're being asked to filter
* on a specific instance of a cloned resource. Put that exact string
* into the filter list. Otherwise, use the printable ID of whatever
* resource was found that matches what was asked for.
*/
if (strstr(s, ":") != NULL) {
resources = g_list_prepend(resources, strdup(rsc->id));
} else {
resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc)));
}
} else {
/* The given string was not a valid resource name. It's a tag or a
* typo or something. See pe__build_node_name_list() for more
* detail.
*/
resources = pe__rscs_with_tag(data_set, s);
}
}
return resources;
}
xmlNode *
pe__failed_probe_for_rsc(const pe_resource_t *rsc, const char *name)
{
const pe_resource_t *parent = pe__const_top_resource(rsc, false);
const char *rsc_id = rsc->id;
if (parent->variant == pcmk_rsc_variant_clone) {
rsc_id = pe__clone_child_id(parent);
}
for (xmlNode *xml_op = pcmk__xml_first_child(rsc->cluster->failed); xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
const char *value = NULL;
char *op_id = NULL;
/* This resource operation is not a failed probe. */
if (!pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
/* This resource operation was not run on the given node. Note that if name is
* NULL, this will always succeed.
*/
value = crm_element_value(xml_op, XML_LRM_ATTR_TARGET);
if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) {
continue;
}
if (!parse_op_key(pe__xe_history_key(xml_op), &op_id, NULL, NULL)) {
continue; // This history entry is missing an operation key
}
/* This resource operation's ID does not match the rsc_id we are looking for. */
if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) {
free(op_id);
continue;
}
free(op_id);
return xml_op;
}
return NULL;
}