diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index ad9cc849b3..d4f416dafd 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1678 +1,1678 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <stdint.h> // uint8_t, uint32_t
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
#define RSC_ROLE_MAX (pcmk_role_promoted + 1)
static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the immediate next role when transitioning from one role
* to a target role. For example, when going from Stopped to Promoted, the
* next role is Unpromoted, because the resource must be started before it
* can be promoted. The current state then becomes Started, which is fed
* into this array again, giving a next role of Promoted.
*
* Current role Immediate next role Final target role
* ------------ ------------------- -----------------
*/
/* Unknown */ { pcmk_role_unknown, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_stopped, /* Unpromoted */
pcmk_role_stopped, /* Promoted */
},
/* Stopped */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_unpromoted, /* Promoted */
},
/* Started */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Unpromoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Promoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_unpromoted, /* Stopped */
pcmk_role_unpromoted, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
};
/*!
* \internal
* \brief Function to schedule actions needed for a role change
*
* \param[in,out] rsc Resource whose role is changing
* \param[in,out] node Node where resource will be in its next role
* \param[in] optional Whether scheduled actions should be optional
*/
typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the function needed to transition directly from one role
* to another. NULL indicates that nothing is needed.
*
* Current role Transition function Next role
* ------------ ------------------- ----------
*/
/* Unknown */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
assert_role_error, /* Started */
assert_role_error, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Stopped */ { assert_role_error, /* Unknown */
NULL, /* Stopped */
start_resource, /* Started */
start_resource, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Started */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
NULL, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Unpromoted */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
stop_resource, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Promoted */ { assert_role_error, /* Unknown */
demote_resource, /* Stopped */
demote_resource, /* Started */
demote_resource, /* Unpromoted */
NULL, /* Promoted */
},
};
/*!
* \internal
* \brief Get a list of a resource's allowed nodes sorted by node score
*
* \param[in] rsc Resource to check
*
* \return List of allowed nodes sorted by node score
*/
static GList *
sorted_allowed_nodes(const pcmk_resource_t *rsc)
{
if (rsc->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->allowed_nodes);
if (nodes != NULL) {
return pcmk__sort_nodes(nodes, pcmk__current_node(rsc));
}
}
return NULL;
}
/*!
* \internal
* \brief Assign a resource to its best allowed node, if possible
*
* \param[in,out] rsc Resource to choose a node for
* \param[in] prefer If not \c NULL, prefer this node when all else
* equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return true if \p rsc could be assigned to a node, otherwise false
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
static bool
assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *nodes = NULL;
pcmk_node_t *chosen = NULL;
pcmk_node_t *best = NULL;
const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// We've already finished assignment of resources to nodes
return rsc->allocated_to != NULL;
}
// Sort allowed nodes by score
nodes = sorted_allowed_nodes(rsc);
if (nodes != NULL) {
best = (pcmk_node_t *) nodes->data; // First node has best score
}
if ((prefer != NULL) && (nodes != NULL)) {
// Get the allowed node version of prefer
chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id);
if (chosen == NULL) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown",
pcmk__node_name(prefer), rsc->id);
/* Favor the preferred node as long as its score is at least as good as
* the best allowed node's.
*
* An alternative would be to favor the preferred node even if the best
* node is better, when the best node's score is less than INFINITY.
*/
} else if (chosen->weight < best->weight) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else if (!pcmk__node_available(chosen, true, false)) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pcmk__rsc_trace(rsc,
"Chose preferred node %s for %s "
"(ignoring %d candidates)",
pcmk__node_name(chosen), rsc->id,
g_list_length(nodes));
}
}
if ((chosen == NULL) && (best != NULL)) {
/* Either there is no preferred node, or the preferred node is not
* suitable, but another node is allowed to run the resource.
*/
chosen = best;
if (!pe_rsc_is_unique_clone(rsc->parent)
&& (chosen->weight > 0) // Zero not acceptable
&& pcmk__node_available(chosen, false, false)) {
/* If the resource is already running on a node, prefer that node if
* it is just as good as the chosen node.
*
* We don't do this for unique clone instances, because
* pcmk__assign_instances() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unassigned instances to prefer a node that's already
* running another instance.
*/
pcmk_node_t *running = pcmk__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pcmk__rsc_trace(rsc,
"Current node for %s (%s) can't run resources",
rsc->id, pcmk__node_name(running));
} else {
int nodes_with_best_score = 1;
for (GList *iter = nodes->next; iter; iter = iter->next) {
pcmk_node_t *allowed = (pcmk_node_t *) iter->data;
if (allowed->weight != chosen->weight) {
// The nodes are sorted by score, so no more are equal
break;
}
if (pcmk__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",
pcmk__node_name(chosen), rsc->id,
nodes_with_best_score,
pcmk_readable_score(chosen->weight));
}
}
}
pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates",
pcmk__node_name(chosen), rsc->id, g_list_length(nodes));
}
pcmk__assign_resource(rsc, chosen, false, stop_if_fail);
g_list_free(nodes);
return rsc->allocated_to != NULL;
}
/*!
* \internal
* \brief Apply a "this with" colocation to a node's allowed node scores
*
* \param[in,out] colocation Colocation to apply
* \param[in,out] rsc Resource being assigned
*/
static void
apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc)
{
GHashTable *archive = NULL;
pcmk_resource_t *other = colocation->primary;
// In certain cases, we will need to revert the node scores
if ((colocation->dependent_role >= pcmk_role_promoted)
|| ((colocation->score < 0) && (colocation->score > -INFINITY))) {
archive = pcmk__copy_node_table(rsc->allowed_nodes);
}
if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) {
pcmk__rsc_trace(rsc,
"%s: Assigning colocation %s primary %s first"
"(score=%d role=%s)",
rsc->id, colocation->id, other->id,
colocation->score,
pcmk_role_text(colocation->dependent_role));
other->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->allowed_nodes)) {
pcmk__rsc_info(rsc,
"%s: Reverting scores from colocation with %s "
"because no nodes allowed",
rsc->id, other->id);
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = archive;
archive = NULL;
}
if (archive != NULL) {
g_hash_table_destroy(archive);
}
}
/*!
* \internal
* \brief Update a Pacemaker Remote node once its connection has been assigned
*
* \param[in] connection Connection resource that has been assigned
*/
static void
remote_connection_assigned(const pcmk_resource_t *connection)
{
pcmk_node_t *remote_node = pe_find_node(connection->cluster->nodes,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->allocated_to != NULL)
&& (connection->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->details->id);
remote_node->details->online = TRUE;
if (remote_node->details->unseen) {
// Avoid unnecessary fence, since we will attempt connection
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Pacemaker Remote node %s will be shut down "
"(%sassigned connection's next role is %s)",
remote_node->details->id,
((connection->allocated_to == NULL)? "un" : ""),
pcmk_role_text(connection->next_role));
remote_node->details->shutdown = TRUE;
}
}
/*!
* \internal
* \brief Assign a primitive resource to a node
*
* \param[in,out] rsc Resource to assign to a node
* \param[in] prefer Node to prefer, if all else is equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pcmk_node_t *
pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *this_with_colocations = NULL;
GList *with_this_colocations = NULL;
GList *iter = NULL;
pcmk__colocation_t *colocation = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
// Never assign a child without parent being assigned first
if ((rsc->parent != NULL)
&& !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, rsc->parent->id);
rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->allocated_to != NULL) {
node_name = pcmk__node_name(rsc->allocated_to);
}
pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->allocated_to;
}
// Ensure we detect assignment loops
if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning);
pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes,
rsc->cluster);
this_with_colocations = pcmk__this_with_colocations(rsc);
with_this_colocations = pcmk__with_this_colocations(rsc);
// Apply mandatory colocations first, to satisfy as many as possible
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->allowed_nodes, rsc->cluster);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->next_role == pcmk_role_stopped) {
pcmk__rsc_trace(rsc,
"Banning %s from all nodes because it will be stopped",
rsc->id);
resource_location(rsc, NULL, -INFINITY, PCMK_META_TARGET_ROLE,
rsc->cluster);
} else if ((rsc->next_role > rsc->role)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate)
&& (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Resource %s cannot be elevated from %s to %s due to "
PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE,
rsc->id, pcmk_role_text(rsc->role),
pcmk_role_text(rsc->next_role));
pe__set_next_role(rsc, rsc->role,
PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE);
}
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->allowed_nodes, rsc->cluster);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pcmk_node_t *assign_to = NULL;
pe__set_next_role(rsc, rsc->role, "unmanaged");
assign_to = pcmk__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->role == pcmk_role_promoted) {
reason = "promoted";
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id,
(assign_to? assign_to->details->uname : "no node"),
reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
pcmk__rsc_debug(rsc,
"Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES,
rsc->id);
}
pcmk__assign_resource(rsc, NULL, true, stop_if_fail);
} else if (!assign_best_node(rsc, prefer, stop_if_fail)) {
// Assignment failed
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->running_on != NULL) && stop_if_fail) {
pcmk__rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
}
pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning);
if (rsc->is_remote_node) {
remote_connection_assigned(rsc);
}
return rsc->allocated_to;
}
/*!
* \internal
* \brief Schedule actions to bring resource down and back to current role
*
* \param[in,out] rsc Resource to restart
* \param[in,out] current Node that resource should be brought down on
* \param[in] need_stop Whether the resource must be stopped
* \param[in] need_promote Whether the resource must be promoted
*
* \return Role that resource would have after scheduled actions are taken
*/
static void
schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current,
bool need_stop, bool need_promote)
{
enum rsc_role_e role = rsc->role;
enum rsc_role_e next_role;
rsc_transition_fn fn = NULL;
pcmk__set_rsc_flags(rsc, pcmk_rsc_restarting);
// Bring resource down to a stop on its current node
while (role != pcmk_role_stopped) {
next_role = rsc_state_matrix[role][pcmk_role_stopped];
pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s",
(need_stop? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, current, !need_stop);
role = next_role;
}
// Bring resource up to its next role on its next node
while ((rsc->role <= rsc->next_role) && (role != rsc->role)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s",
(required? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, !required);
role = next_role;
}
pcmk__clear_rsc_flags(rsc, pcmk_rsc_restarting);
}
/*!
* \internal
* \brief If a resource's next role is not explicitly specified, set a default
*
* \param[in,out] rsc Resource to set next role for
*
* \return "explicit" if next role was explicitly set, otherwise "implicit"
*/
static const char *
set_default_next_role(pcmk_resource_t *rsc)
{
if (rsc->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->allocated_to == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "assignment");
} else {
pe__set_next_role(rsc, pcmk_role_started, "assignment");
}
return "implicit";
}
/*!
* \internal
* \brief Create an action to represent an already pending start
*
* \param[in,out] rsc Resource to create start action for
*/
static void
create_pending_start(pcmk_resource_t *rsc)
{
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc,
"Creating action for %s to represent already pending start",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
pcmk__set_action_flags(start, pcmk_action_always_in_graph);
}
/*!
* \internal
* \brief Schedule actions needed to take a resource to its next role
*
* \param[in,out] rsc Resource to schedule actions for
*/
static void
schedule_role_transition_actions(pcmk_resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
while (role != rsc->next_role) {
enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role];
rsc_transition_fn fn = NULL;
pcmk__rsc_trace(rsc,
"Creating action to take %s from %s to %s "
"(ending at %s)",
rsc->id, pcmk_role_text(role),
pcmk_role_text(next_role),
pcmk_role_text(rsc->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, false);
role = next_role;
}
}
/*!
* \internal
* \brief Create all actions needed for a given primitive resource
*
* \param[in,out] rsc Primitive resource to create actions for
*/
void
pcmk__primitive_create_actions(pcmk_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pcmk_node_t *current = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
next_role_source = set_default_next_role(rsc);
pcmk__rsc_trace(rsc,
"Creating all actions for %s transition from %s to %s "
"(%s) on %s",
rsc->id, pcmk_role_text(rsc->role),
pcmk_role_text(rsc->next_role), next_role_source,
pcmk__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)
&& !pcmk__same_node(current, rsc->allocated_to)
&& (rsc->next_role >= pcmk_role_started)) {
pcmk__rsc_trace(rsc, "Moving %s from %s to %s",
rsc->id, pcmk__node_name(current),
pcmk__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)
&& pcmk__same_node(current, rsc->partial_migration_source)
&& pcmk__same_node(rsc->allocated_to, rsc->partial_migration_target)) {
/* A partial migration is in progress, and the migration target remains
* the same as when the migration began.
*/
pcmk__rsc_trace(rsc,
"Partial migration of %s from %s to %s will continue",
rsc->id, pcmk__node_name(rsc->partial_migration_source),
pcmk__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, pcmk__node_name(rsc->partial_migration_source),
pcmk__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, pcmk__node_name(rsc->partial_migration_source),
pcmk__node_name(rsc->partial_migration_target));
}
need_stop = true;
rsc->partial_migration_source = rsc->partial_migration_target = NULL;
allow_migrate = false;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
multiply_active = (num_all_active > 1);
} else {
- /* If a resource has "requires" set to nothing or quorum, don't consider
- * it active on unclean nodes (similar to how all resources behave when
- * PCMK_OPT_STONITH_ENABLED is false). We can start such resources
- * elsewhere before fencing completes, and if we considered the resource
- * active on the failed node, we would attempt recovery for being active
- * on multiple nodes.
+ /* If a resource has PCMK_META_REQUIRES set to PCMK_VALUE_NOTHING or
+ * PCMK_VALUE_QUORUM, don't consider it active on unclean nodes (similar
+ * to how all resources behave when PCMK_OPT_STONITH_ENABLED is false).
+ * We can start such resources elsewhere before fencing completes, and
+ * if we considered the resource active on the failed node, we would
+ * attempt recovery for being active on multiple nodes.
*/
multiply_active = (num_clean_active > 1);
}
if (multiply_active) {
const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
// Resource was (possibly) incorrectly multiply active
pcmk__sched_err("%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), rsc->id, num_all_active,
pcmk_multiply_active_text(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
pcmk__set_rsc_flags(rsc, pcmk_rsc_stop_unexpected);
break;
default:
break;
}
} else {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
create_pending_start(rsc);
}
if (is_moving) {
// Remaining tests are only for resources staying where they are
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) {
need_stop = true;
pcmk__rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id);
if (rsc->next_role == pcmk_role_promoted) {
need_promote = true;
}
}
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id);
need_stop = true;
} else if ((rsc->role > pcmk_role_started) && (current != NULL)
&& (rsc->allocated_to != NULL)) {
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
if (!pcmk_is_set(start->flags, pcmk_action_optional)) {
// Recovery of a promoted resource
pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id);
need_stop = true;
}
}
// Create any actions needed to bring resource down and back up to same role
schedule_restart_actions(rsc, current, need_stop, need_promote);
// Create any actions needed to take resource from this role to the next
schedule_role_transition_actions(rsc);
pcmk__create_recurring_actions(rsc);
if (allow_migrate) {
pcmk__create_migration_actions(rsc, current);
}
}
/*!
* \internal
* \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes
*
* \param[in] rsc Resource to check
*/
static void
rsc_avoids_remote_nodes(const pcmk_resource_t *rsc)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (node->details->remote_rsc != NULL) {
node->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Return allowed nodes as (possibly sorted) list
*
* Convert a resource's hash table of allowed nodes to a list. If printing to
* stdout, sort the list, to keep action ID numbers consistent for regression
* test output (while avoiding the performance hit on a live cluster).
*
* \param[in] rsc Resource to check for allowed nodes
*
* \return List of resource's allowed nodes
* \note Callers should take care not to rely on the list being sorted.
*/
static GList *
allowed_nodes_as_list(const pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
if (rsc->allowed_nodes) {
allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
if (!pcmk__is_daemon) {
allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name);
}
return allowed_nodes;
}
/*!
* \internal
* \brief Create implicit constraints needed for a primitive resource
*
* \param[in,out] rsc Primitive resource to create implicit constraints for
*/
void
pcmk__primitive_internal_constraints(pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
bool check_unfencing = false;
bool check_utilization = false;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource "
"%s", rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags,
pcmk_sched_enable_unfencing)
&& pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& !pcmk__str_eq(rsc->cluster->placement_strategy,
PCMK_VALUE_DEFAULT, pcmk__str_casei);
// Order stops before starts (i.e. restart)
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL,
pcmk__ar_ordered
|pcmk__ar_first_implies_then
|pcmk__ar_intermediate_stop,
rsc->cluster);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable)
|| (rsc->role > pcmk_role_unpromoted)) {
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pcmk__ar_promoted_then_implies_first, rsc->cluster);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0),
NULL,
pcmk__ar_unrunnable_first_blocks, rsc->cluster);
}
// Don't clear resource history if probing on same node
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL,
pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first,
rsc->cluster);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization || (rsc->container != NULL)) {
allowed_nodes = allowed_nodes_as_list(rsc);
}
if (check_unfencing) {
g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc);
}
if (check_utilization) {
pcmk__create_utilization_constraints(rsc, allowed_nodes);
}
if (rsc->container != NULL) {
pcmk_resource_t *remote_rsc = NULL;
if (rsc->is_remote_node) {
// rsc is the implicit remote connection for a guest or bundle node
/* Guest resources are not allowed to run on Pacemaker Remote nodes,
* to avoid nesting remotes. However, bundles are allowed.
*/
if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
rsc_avoids_remote_nodes(rsc->container);
}
/* If someone cleans up a guest or bundle node's container, we will
* likely schedule a (re-)probe of the container and recovery of the
* connection. Order the connection stop after the container probe,
* so that if we detect the container running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pcmk__ar_ordered);
/* A user can specify that a resource must start on a Pacemaker Remote
* node by explicitly configuring it with the container=NODENAME
* meta-attribute. This is of questionable merit, since location
* constraints can accomplish the same thing. But we support it, so here
* we check whether a resource (that is not itself a remote connection)
* has container set to a remote node or guest node resource.
*/
} else if (rsc->container->is_remote_node) {
remote_rsc = rsc->container;
} else {
remote_rsc = pe__resource_contains_guest_node(rsc->cluster,
rsc->container);
}
if (remote_rsc != NULL) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the container resource.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pcmk_node_t *node = item->data;
if (node->details->remote_rsc != remote_rsc) {
node->weight = -INFINITY;
}
}
} else {
/* This resource is either a filler for a container that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its container %s",
rsc->id, rsc->container->id);
pcmk__new_ordering(rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_START, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
pcmk__ar_first_implies_then
|pcmk__ar_unrunnable_first_blocks,
rsc->cluster);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_STOP, 0),
NULL, pcmk__ar_then_implies_first, rsc->cluster);
if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->container, NULL, NULL,
pcmk__coloc_influence);
}
}
if (rsc->is_remote_node
|| pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
/* Remote connections and fencing devices are not allowed to run on
* Pacemaker Remote nodes
*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint, apply its score to the dependent's
* allowed node scores (if we are still placing resources) or priority (if
* we are choosing promotable clone instance roles).
*
* \param[in,out] dependent Dependent resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
* \param[in] for_dependent true if called on behalf of dependent
*/
void
pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent,
const pcmk_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
enum pcmk__coloc_affects filter_results;
CRM_ASSERT((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
// Always process on behalf of primary resource
primary->cmds->apply_coloc_score(dependent, primary, colocation, false);
return;
}
filter_results = pcmk__colocation_affects(dependent, primary, colocation,
false);
pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)",
((colocation->score > 0)? "Colocating" : "Anti-colocating"),
dependent->id, primary->id, colocation->id,
colocation->score,
filter_results);
switch (filter_results) {
case pcmk__coloc_affects_role:
pcmk__apply_coloc_to_priority(dependent, primary, colocation);
break;
case pcmk__coloc_affects_location:
pcmk__apply_coloc_to_scores(dependent, primary, colocation);
break;
default: // pcmk__coloc_affects_nothing
return;
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_with_this(list, colocation, orig_rsc);
}
}
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
}
}
/*!
* \internal
* \brief Return action flags for a given primitive resource action
*
* \param[in,out] action Action to get flags for
* \param[in] node If not NULL, limit effects to this node (ignored)
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node)
{
CRM_ASSERT(action != NULL);
return (uint32_t) action->flags;
}
/*!
* \internal
* \brief Check whether a node is a multiply active resource's expected node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return \c true if \p rsc is multiply active with
* \c PCMK_META_MULTIPLE_ACTIVE set to \c stop_unexpected, and \p node
* is the node where it will remain active
* \note This assumes that the resource's next role cannot be changed to stopped
* after this is called, which should be reasonable if status has already
* been unpacked and resources have been assigned to nodes.
*/
static bool
is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pcmk_rsc_stop_unexpected|pcmk_rsc_restarting)
&& (rsc->next_role > pcmk_role_stopped)
&& pcmk__same_node(rsc->allocated_to, node);
}
/*!
* \internal
* \brief Schedule actions needed to stop a resource wherever it is active
*
* \param[in,out] rsc Resource being stopped
* \param[in] node Node where resource is being stopped (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
pcmk_action_t *stop = NULL;
if (is_expected_node(rsc, current)) {
/* We are scheduling restart actions for a multiply active resource
* with PCMK_META_MULTIPLE_ACTIVE=stop_unexpected, and this is where
* it should not be stopped.
*/
pcmk__rsc_trace(rsc,
"Skipping stop of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
continue;
}
if (rsc->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pcmk__same_node(current, rsc->partial_migration_target)
&& pcmk__same_node(current, rsc->allocated_to)) {
pcmk__rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pcmk__node_name(current));
continue;
} else {
pcmk__rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pcmk__node_name(current));
optional = false;
}
}
pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s",
rsc->id, pcmk__node_name(current));
stop = stop_action(rsc, current, optional);
if (rsc->allocated_to == NULL) {
pe_action_set_reason(stop, "node availability", true);
} else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting
|pcmk_rsc_stop_unexpected)) {
/* We are stopping a multiply active resource on a node that is
* not its expected node, and we are still scheduling restart
* actions, so the stop is for being multiply active.
*/
pe_action_set_reason(stop, "being multiply active", true);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__clear_action_flags(stop, pcmk_action_runnable);
}
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) {
pcmk__schedule_cleanup(rsc, current, optional);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) {
pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false, rsc->cluster);
order_actions(stop, unfence, pcmk__ar_then_implies_first);
if (!pcmk__node_unfenced(current)) {
pcmk__sched_err("Stopping %s until %s can be unfenced",
rsc->id, pcmk__node_name(current));
}
}
}
}
/*!
* \internal
* \brief Schedule actions needed to start a resource on a node
*
* \param[in,out] rsc Resource being started
* \param[in,out] node Node where resource should be started
* \param[in] optional Whether actions should be optional
*/
static void
start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
pcmk_action_t *start = NULL;
CRM_ASSERT(node != NULL);
pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node), node->weight);
start = start_action(rsc, node, TRUE);
pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then);
if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) {
pcmk__clear_action_flags(start, pcmk_action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pcmk__rsc_trace(rsc,
"Start of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(start, pcmk_action_pseudo);
}
}
/*!
* \internal
* \brief Schedule actions needed to promote a resource on a node
*
* \param[in,out] rsc Resource being promoted
* \param[in] node Node where resource should be promoted
* \param[in] optional Whether actions should be optional
*/
static void
promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
GList *iter = NULL;
GList *action_list = NULL;
bool runnable = true;
CRM_ASSERT(node != NULL);
// Any start must be runnable for promotion to be runnable
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *start = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pcmk_action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pcmk_action_t *promote = promote_action(rsc, node, optional);
pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node));
if (is_expected_node(rsc, node)) {
/* This could be a problem if the promote becomes necessary for
* other reasons later.
*/
pcmk__rsc_trace(rsc,
"Promotion of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(promote, pcmk_action_pseudo);
}
} else {
pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable",
rsc->id, pcmk__node_name(node));
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE,
true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *promote = (pcmk_action_t *) iter->data;
pcmk__clear_action_flags(promote, pcmk_action_runnable);
}
g_list_free(action_list);
}
}
/*!
* \internal
* \brief Schedule actions needed to demote a resource wherever it is active
*
* \param[in,out] rsc Resource being demoted
* \param[in] node Node where resource should be demoted (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
/* Since this will only be called for a primitive (possibly as an instance
* of a collective resource), the resource is multiply active if it is
* running on more than one node, so we want to demote on all of them as
* part of recovery, regardless of which one is the desired node.
*/
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pcmk__rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
} else {
pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(current));
demote_action(rsc, current, optional);
}
}
}
static void
assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
CRM_ASSERT(false);
}
/*!
* \internal
* \brief Schedule cleanup of a resource
*
* \param[in,out] rsc Resource to clean up
* \param[in] node Node to clean up on
* \param[in] optional Whether clean-up should be optional
*/
void
pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node,
bool optional)
{
/* If the cleanup is required, its orderings are optional, because they're
* relevant only if both actions are required. Conversely, if the cleanup is
* optional, the orderings make the then action required if the first action
* becomes required.
*/
uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed",
rsc->id, pcmk__node_name(node));
return;
}
if (node->details->unclean || !node->details->online) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable",
rsc->id, pcmk__node_name(node));
return;
}
crm_notice("Scheduling clean-up of %s on %s",
rsc->id, pcmk__node_name(node));
delete_action(rsc, node, optional);
// stop -> clean-up -> start
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_DELETE, flag);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE,
rsc, PCMK_ACTION_START, flag);
}
/*!
* \internal
* \brief Add primitive meta-attributes relevant to graph actions to XML
*
* \param[in] rsc Primitive resource whose meta-attributes should be added
* \param[in,out] xml Transition graph action attributes XML to add to
*/
void
pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pcmk_resource_t *parent = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (xml != NULL));
/* Clone instance numbers get set internally as meta-attributes, and are
* needed in the transition graph (for example, to tell unique clone
* instances apart).
*/
value = g_hash_table_lookup(rsc->meta, PCMK__META_CLONE);
if (value != NULL) {
name = crm_meta_name(PCMK__META_CLONE);
crm_xml_add(xml, name, value);
free(name);
}
// Not sure if this one is really needed ...
value = g_hash_table_lookup(rsc->meta, PCMK_META_REMOTE_NODE);
if (value != NULL) {
name = crm_meta_name(PCMK_META_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 "_" PCMK__META_CONTAINER,
parent->container->id);
}
}
/* Bundle replica children will get their external-ip set internally as a
* meta-attribute. The graph action needs it, but under a different naming
* convention than other meta-attributes.
*/
value = g_hash_table_lookup(rsc->meta, "external-ip");
if (value != NULL) {
crm_xml_add(xml, "pcmk_external_ip", value);
}
}
// Primitive implementation of pcmk_assignment_methods_t:add_utilization()
void
pcmk__primitive_add_utilization(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc,
GList *all_rscs, GHashTable *utilization)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (orig_rsc != NULL) && (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return;
}
pcmk__rsc_trace(orig_rsc,
"%s: Adding primitive %s as colocated utilization",
orig_rsc->id, rsc->id);
pcmk__release_node_capacity(utilization, rsc);
}
/*!
* \internal
* \brief Get epoch time of node's shutdown attribute (or now if none)
*
* \param[in,out] node Node to check
*
* \return Epoch time corresponding to shutdown attribute if set or now if not
*/
static time_t
shutdown_time(pcmk_node_t *node)
{
const char *shutdown = pe_node_attribute_raw(node,
PCMK__NODE_ATTR_SHUTDOWN);
time_t result = 0;
if (shutdown != NULL) {
long long result_ll;
if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) {
result = (time_t) result_ll;
}
}
return (result == 0)? get_effective_time(node->details->data_set) : result;
}
/*!
* \internal
* \brief Ban a resource from a node if it's not locked to the node
*
* \param[in] data Node to check
* \param[in,out] user_data Resource to check
*/
static void
ban_if_not_locked(gpointer data, gpointer user_data)
{
const pcmk_node_t *node = (const pcmk_node_t *) data;
pcmk_resource_t *rsc = (pcmk_resource_t *) user_data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
PCMK_OPT_SHUTDOWN_LOCK, rsc->cluster);
}
}
// Primitive implementation of pcmk_assignment_methods_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc)
{
const char *class = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
// Fence devices and remote connections can't be locked
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches)
|| pe__resource_is_remote_conn(rsc)) {
return;
}
if (rsc->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->running_on != NULL) {
/* The resource was started elsewhere even though it is now
* considered locked. This shouldn't be possible, but as a
* failsafe, we don't want to disturb the resource now.
*/
pcmk__rsc_info(rsc,
"Cancelling shutdown lock "
"because %s is already active", rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pcmk_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pcmk__rsc_debug(rsc,
"Not locking %s to unclean %s for shutdown",
rsc->id, pcmk__node_name(node));
} else {
rsc->lock_node = node;
rsc->lock_time = shutdown_time(node);
}
}
}
if (rsc->lock_node == NULL) {
// No lock needed
return;
}
if (rsc->cluster->shutdown_lock > 0) {
time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock;
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pcmk__node_name(rsc->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, rsc->cluster,
"shutdown lock expiration");
} else {
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pcmk__node_name(rsc->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c
index 877cfdbbcb..ad4ec20225 100644
--- a/lib/pengine/complex.c
+++ b/lib/pengine/complex.c
@@ -1,1220 +1,1220 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include <crm/msg_xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/scheduler_internal.h>
#include "pe_status_private.h"
void populate_hash(xmlNode * nvpair_list, GHashTable * hash, const char **attrs, int attrs_length);
static pcmk_node_t *active_node(const pcmk_resource_t *rsc,
unsigned int *count_all,
unsigned int *count_clean);
pcmk_rsc_methods_t resource_class_functions[] = {
{
native_unpack,
native_find_rsc,
native_parameter,
native_print,
native_active,
native_resource_state,
native_location,
native_free,
pe__count_common,
pe__native_is_filtered,
active_node,
pe__primitive_max_per_node,
},
{
group_unpack,
native_find_rsc,
native_parameter,
group_print,
group_active,
group_resource_state,
native_location,
group_free,
pe__count_common,
pe__group_is_filtered,
active_node,
pe__group_max_per_node,
},
{
clone_unpack,
native_find_rsc,
native_parameter,
clone_print,
clone_active,
clone_resource_state,
native_location,
clone_free,
pe__count_common,
pe__clone_is_filtered,
active_node,
pe__clone_max_per_node,
},
{
pe__unpack_bundle,
native_find_rsc,
native_parameter,
pe__print_bundle,
pe__bundle_active,
pe__bundle_resource_state,
native_location,
pe__free_bundle,
pe__count_bundle,
pe__bundle_is_filtered,
pe__bundle_active_node,
pe__bundle_max_per_node,
}
};
static enum pe_obj_types
get_resource_type(const char *name)
{
if (pcmk__str_eq(name, PCMK_XE_PRIMITIVE, pcmk__str_casei)) {
return pcmk_rsc_variant_primitive;
} else if (pcmk__str_eq(name, PCMK_XE_GROUP, pcmk__str_casei)) {
return pcmk_rsc_variant_group;
} else if (pcmk__str_eq(name, PCMK_XE_CLONE, pcmk__str_casei)) {
return pcmk_rsc_variant_clone;
} else if (pcmk__str_eq(name, PCMK__XE_PROMOTABLE_LEGACY,
pcmk__str_casei)) {
// @COMPAT deprecated since 2.0.0
return pcmk_rsc_variant_clone;
} else if (pcmk__str_eq(name, PCMK_XE_BUNDLE, pcmk__str_casei)) {
return pcmk_rsc_variant_bundle;
}
return pcmk_rsc_variant_unknown;
}
static void
dup_attr(gpointer key, gpointer value, gpointer user_data)
{
add_hash_param(user_data, key, value);
}
static void
expand_parents_fixed_nvpairs(pcmk_resource_t *rsc,
pe_rule_eval_data_t *rule_data,
GHashTable *meta_hash, pcmk_scheduler_t *scheduler)
{
GHashTable *parent_orig_meta = pcmk__strkey_table(free, free);
pcmk_resource_t *p = rsc->parent;
if (p == NULL) {
return ;
}
/* Search all parent resources, get the fixed value of
* PCMK_XE_META_ATTRIBUTES set only in the original xml, and stack it in the
* hash table. The fixed value of the lower parent resource takes precedence
* and is not overwritten.
*/
while(p != NULL) {
/* A hash table for comparison is generated, including the id-ref. */
pe__unpack_dataset_nvpairs(p->xml, PCMK_XE_META_ATTRIBUTES, rule_data,
parent_orig_meta, NULL, FALSE, scheduler);
p = p->parent;
}
/* If there is a fixed value of PCMK_XE_META_ATTRIBUTES of the parent
* resource, it will be processed.
*/
if (parent_orig_meta != NULL) {
GHashTableIter iter;
char *key = NULL;
char *value = NULL;
g_hash_table_iter_init(&iter, parent_orig_meta);
while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
/* Parameters set in the original xml of the parent resource will also try to overwrite the child resource. */
/* Attributes that already exist in the child lease are not updated. */
dup_attr(key, value, meta_hash);
}
}
if (parent_orig_meta != NULL) {
g_hash_table_destroy(parent_orig_meta);
}
return ;
}
void
get_meta_attributes(GHashTable * meta_hash, pcmk_resource_t * rsc,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pe_rsc_eval_data_t rsc_rule_data = {
.standard = crm_element_value(rsc->xml, PCMK_XA_CLASS),
.provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER),
.agent = crm_element_value(rsc->xml, PCMK_XA_TYPE)
};
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = &rsc_rule_data,
.op_data = NULL
};
if (node) {
rule_data.node_hash = node->details->attrs;
}
for (xmlAttrPtr a = pcmk__xe_first_attr(rsc->xml); a != NULL; a = a->next) {
const char *prop_name = (const char *) a->name;
const char *prop_value = pcmk__xml_attr_value(a);
add_hash_param(meta_hash, prop_name, prop_value);
}
pe__unpack_dataset_nvpairs(rsc->xml, PCMK_XE_META_ATTRIBUTES, &rule_data,
meta_hash, NULL, FALSE, scheduler);
/* Set the PCMK_XE_META_ATTRIBUTES explicitly set in the parent resource to
* the hash table of the child resource. If it is already explicitly set as
* a child, it will not be overwritten.
*/
if (rsc->parent != NULL) {
expand_parents_fixed_nvpairs(rsc, &rule_data, meta_hash, scheduler);
}
/* check the defaults */
pe__unpack_dataset_nvpairs(scheduler->rsc_defaults, PCMK_XE_META_ATTRIBUTES,
&rule_data, meta_hash, NULL, FALSE, scheduler);
/* If there is PCMK_XE_META_ATTRIBUTES that the parent resource has not
* explicitly set, set a value that is not set from PCMK_XE_RSC_DEFAULTS
* either. The values already set up to this point will not be overwritten.
*/
if (rsc->parent) {
g_hash_table_foreach(rsc->parent->meta, dup_attr, meta_hash);
}
}
void
get_rsc_attributes(GHashTable *meta_hash, const pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
if (node) {
rule_data.node_hash = node->details->attrs;
}
pe__unpack_dataset_nvpairs(rsc->xml, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_data, meta_hash, NULL, FALSE, scheduler);
/* set anything else based on the parent */
if (rsc->parent != NULL) {
get_rsc_attributes(meta_hash, rsc->parent, node, scheduler);
} else {
/* and finally check the defaults */
pe__unpack_dataset_nvpairs(scheduler->rsc_defaults,
PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data,
meta_hash, NULL, FALSE, scheduler);
}
}
static char *
template_op_key(xmlNode * op)
{
const char *name = crm_element_value(op, PCMK_XA_NAME);
const char *role = crm_element_value(op, PCMK_XA_ROLE);
char *key = NULL;
if ((role == NULL)
|| pcmk__strcase_any_of(role, PCMK__ROLE_STARTED, PCMK__ROLE_UNPROMOTED,
PCMK__ROLE_UNPROMOTED_LEGACY, NULL)) {
role = PCMK__ROLE_UNKNOWN;
}
key = crm_strdup_printf("%s-%s", name, role);
return key;
}
static gboolean
unpack_template(xmlNode *xml_obj, xmlNode **expanded_xml,
pcmk_scheduler_t *scheduler)
{
xmlNode *cib_resources = NULL;
xmlNode *template = NULL;
xmlNode *new_xml = NULL;
xmlNode *child_xml = NULL;
xmlNode *rsc_ops = NULL;
xmlNode *template_ops = NULL;
const char *template_ref = NULL;
const char *id = NULL;
if (xml_obj == NULL) {
pcmk__config_err("No resource object for template unpacking");
return FALSE;
}
template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE);
if (template_ref == NULL) {
return TRUE;
}
id = ID(xml_obj);
if (id == NULL) {
pcmk__config_err("'%s' object must have a id", xml_obj->name);
return FALSE;
}
if (pcmk__str_eq(template_ref, id, pcmk__str_none)) {
pcmk__config_err("The resource object '%s' should not reference itself",
id);
return FALSE;
}
cib_resources = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input,
LOG_TRACE);
if (cib_resources == NULL) {
pcmk__config_err("No resources configured");
return FALSE;
}
template = pcmk__xe_match(cib_resources, PCMK_XE_TEMPLATE,
PCMK_XA_ID, template_ref);
if (template == NULL) {
pcmk__config_err("No template named '%s'", template_ref);
return FALSE;
}
new_xml = copy_xml(template);
xmlNodeSetName(new_xml, xml_obj->name);
crm_xml_add(new_xml, PCMK_XA_ID, id);
crm_xml_add(new_xml, PCMK__META_CLONE,
crm_element_value(xml_obj, PCMK__META_CLONE));
template_ops = find_xml_node(new_xml, PCMK_XE_OPERATIONS, FALSE);
for (child_xml = pcmk__xe_first_child(xml_obj); child_xml != NULL;
child_xml = pcmk__xe_next(child_xml)) {
xmlNode *new_child = NULL;
new_child = add_node_copy(new_xml, child_xml);
if (pcmk__str_eq((const char *) new_child->name, PCMK_XE_OPERATIONS,
pcmk__str_none)) {
rsc_ops = new_child;
}
}
if (template_ops && rsc_ops) {
xmlNode *op = NULL;
GHashTable *rsc_ops_hash = pcmk__strkey_table(free, NULL);
for (op = pcmk__xe_first_child(rsc_ops); op != NULL;
op = pcmk__xe_next(op)) {
char *key = template_op_key(op);
g_hash_table_insert(rsc_ops_hash, key, op);
}
for (op = pcmk__xe_first_child(template_ops); op != NULL;
op = pcmk__xe_next(op)) {
char *key = template_op_key(op);
if (g_hash_table_lookup(rsc_ops_hash, key) == NULL) {
add_node_copy(rsc_ops, op);
}
free(key);
}
if (rsc_ops_hash) {
g_hash_table_destroy(rsc_ops_hash);
}
free_xml(template_ops);
}
/*free_xml(*expanded_xml); */
*expanded_xml = new_xml;
#if 0 /* Disable multi-level templates for now */
if (!unpack_template(new_xml, expanded_xml, scheduler)) {
free_xml(*expanded_xml);
*expanded_xml = NULL;
return FALSE;
}
#endif
return TRUE;
}
static gboolean
add_template_rsc(xmlNode *xml_obj, pcmk_scheduler_t *scheduler)
{
const char *template_ref = NULL;
const char *id = NULL;
if (xml_obj == NULL) {
pcmk__config_err("No resource object for processing resource list "
"of template");
return FALSE;
}
template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE);
if (template_ref == NULL) {
return TRUE;
}
id = ID(xml_obj);
if (id == NULL) {
pcmk__config_err("'%s' object must have a id", xml_obj->name);
return FALSE;
}
if (pcmk__str_eq(template_ref, id, pcmk__str_none)) {
pcmk__config_err("The resource object '%s' should not reference itself",
id);
return FALSE;
}
if (add_tag_ref(scheduler->template_rsc_sets, template_ref, id) == FALSE) {
return FALSE;
}
return TRUE;
}
static bool
detect_promotable(pcmk_resource_t *rsc)
{
const char *promotable = g_hash_table_lookup(rsc->meta,
PCMK_META_PROMOTABLE);
if (crm_is_true(promotable)) {
return TRUE;
}
// @COMPAT deprecated since 2.0.0
if (pcmk__xe_is(rsc->xml, PCMK__XE_PROMOTABLE_LEGACY)) {
/* @TODO in some future version, pcmk__warn_once() here,
* then drop support in even later version
*/
g_hash_table_insert(rsc->meta, strdup(PCMK_META_PROMOTABLE),
strdup(PCMK_VALUE_TRUE));
return TRUE;
}
return FALSE;
}
static void
free_params_table(gpointer data)
{
g_hash_table_destroy((GHashTable *) data);
}
/*!
* \brief Get a table of resource parameters
*
* \param[in,out] rsc Resource to query
* \param[in] node Node for evaluating rules (NULL for defaults)
* \param[in,out] scheduler Scheduler data
*
* \return Hash table containing resource parameter names and values
* (or NULL if \p rsc or \p scheduler is NULL)
* \note The returned table will be destroyed when the resource is freed, so
* callers should not destroy it.
*/
GHashTable *
pe_rsc_params(pcmk_resource_t *rsc, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
GHashTable *params_on_node = NULL;
/* A NULL node is used to request the resource's default parameters
* (not evaluated for node), but we always want something non-NULL
* as a hash table key.
*/
const char *node_name = "";
// Sanity check
if ((rsc == NULL) || (scheduler == NULL)) {
return NULL;
}
if ((node != NULL) && (node->details->uname != NULL)) {
node_name = node->details->uname;
}
// Find the parameter table for given node
if (rsc->parameter_cache == NULL) {
rsc->parameter_cache = pcmk__strikey_table(free, free_params_table);
} else {
params_on_node = g_hash_table_lookup(rsc->parameter_cache, node_name);
}
// If none exists yet, create one with parameters evaluated for node
if (params_on_node == NULL) {
params_on_node = pcmk__strkey_table(free, free);
get_rsc_attributes(params_on_node, rsc, node, scheduler);
g_hash_table_insert(rsc->parameter_cache, strdup(node_name),
params_on_node);
}
return params_on_node;
}
/*!
* \internal
- * \brief Unpack a resource's "requires" meta-attribute
+ * \brief Unpack a resource's \c PCMK_META_REQUIRES meta-attribute
*
* \param[in,out] rsc Resource being unpacked
- * \param[in] value Value of "requires" meta-attribute
+ * \param[in] value Value of \c PCMK_META_REQUIRES meta-attribute
* \param[in] is_default Whether \p value was selected by default
*/
static void
unpack_requires(pcmk_resource_t *rsc, const char *value, bool is_default)
{
if (pcmk__str_eq(value, PCMK_VALUE_NOTHING, pcmk__str_casei)) {
} else if (pcmk__str_eq(value, PCMK_VALUE_QUORUM, pcmk__str_casei)) {
pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_quorum);
} else if (pcmk__str_eq(value, PCMK_VALUE_FENCING, pcmk__str_casei)) {
pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_fencing);
if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) {
pcmk__config_warn("%s requires fencing but fencing is disabled",
rsc->id);
}
} else if (pcmk__str_eq(value, PCMK_VALUE_UNFENCING, pcmk__str_casei)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s "
"to \"" PCMK_VALUE_QUORUM "\" because fencing "
"devices cannot require unfencing", rsc->id);
unpack_requires(rsc, PCMK_VALUE_QUORUM, true);
return;
} else if (!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s "
"to \"" PCMK_VALUE_QUORUM "\" because fencing is "
"disabled", rsc->id);
unpack_requires(rsc, PCMK_VALUE_QUORUM, true);
return;
} else {
pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_fencing
|pcmk_rsc_needs_unfencing);
}
} else {
const char *orig_value = value;
if (pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
value = PCMK_VALUE_QUORUM;
} else if ((rsc->variant == pcmk_rsc_variant_primitive)
&& xml_contains_remote_node(rsc->xml)) {
value = PCMK_VALUE_QUORUM;
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_enable_unfencing)) {
value = PCMK_VALUE_UNFENCING;
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
value = PCMK_VALUE_FENCING;
} else if (rsc->cluster->no_quorum_policy == pcmk_no_quorum_ignore) {
value = PCMK_VALUE_NOTHING;
} else {
value = PCMK_VALUE_QUORUM;
}
if (orig_value != NULL) {
pcmk__config_err("Resetting '" PCMK_META_REQUIRES "' for %s "
"to '%s' because '%s' is not valid",
rsc->id, value, orig_value);
}
unpack_requires(rsc, value, true);
return;
}
pcmk__rsc_trace(rsc, "\tRequired to start: %s%s", value,
(is_default? " (default)" : ""));
}
#ifndef PCMK__COMPAT_2_0
static void
warn_about_deprecated_classes(pcmk_resource_t *rsc)
{
const char *std = crm_element_value(rsc->xml, PCMK_XA_CLASS);
if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_none)) {
pcmk__warn_once(pcmk__wo_upstart,
"Support for Upstart resources (such as %s) is "
"deprecated and will be removed in a future release",
rsc->id);
} else if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_none)) {
pcmk__warn_once(pcmk__wo_nagios,
"Support for Nagios resources (such as %s) is "
"deprecated and will be removed in a future release",
rsc->id);
}
}
#endif
/*!
* \internal
* \brief Unpack configuration XML for a given resource
*
* Unpack the XML object containing a resource's configuration into a new
* \c pcmk_resource_t object.
*
* \param[in] xml_obj XML node containing the resource's configuration
* \param[out] rsc Where to store the unpacked resource information
* \param[in] parent Resource's parent, if any
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code
* \note If pcmk_rc_ok is returned, \p *rsc is guaranteed to be non-NULL, and
* the caller is responsible for freeing it using its variant-specific
* free() method. Otherwise, \p *rsc is guaranteed to be NULL.
*/
int
pe__unpack_resource(xmlNode *xml_obj, pcmk_resource_t **rsc,
pcmk_resource_t *parent, pcmk_scheduler_t *scheduler)
{
xmlNode *expanded_xml = NULL;
xmlNode *ops = NULL;
const char *value = NULL;
const char *id = NULL;
bool guest_node = false;
bool remote_node = false;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = NULL,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
CRM_CHECK(rsc != NULL, return EINVAL);
CRM_CHECK((xml_obj != NULL) && (scheduler != NULL),
*rsc = NULL;
return EINVAL);
rule_data.now = scheduler->now;
crm_log_xml_trace(xml_obj, "[raw XML]");
id = crm_element_value(xml_obj, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring <%s> configuration without " PCMK_XA_ID,
xml_obj->name);
return pcmk_rc_unpack_error;
}
if (unpack_template(xml_obj, &expanded_xml, scheduler) == FALSE) {
return pcmk_rc_unpack_error;
}
*rsc = calloc(1, sizeof(pcmk_resource_t));
if (*rsc == NULL) {
pcmk__sched_err("Unable to allocate memory for resource '%s'", id);
return ENOMEM;
}
(*rsc)->cluster = scheduler;
if (expanded_xml) {
crm_log_xml_trace(expanded_xml, "[expanded XML]");
(*rsc)->xml = expanded_xml;
(*rsc)->orig_xml = xml_obj;
} else {
(*rsc)->xml = xml_obj;
(*rsc)->orig_xml = NULL;
}
/* Do not use xml_obj from here on, use (*rsc)->xml in case templates are involved */
(*rsc)->parent = parent;
ops = find_xml_node((*rsc)->xml, PCMK_XE_OPERATIONS, FALSE);
(*rsc)->ops_xml = expand_idref(ops, scheduler->input);
(*rsc)->variant = get_resource_type((const char *) (*rsc)->xml->name);
if ((*rsc)->variant == pcmk_rsc_variant_unknown) {
pcmk__config_err("Ignoring resource '%s' of unknown type '%s'",
id, (*rsc)->xml->name);
common_free(*rsc);
*rsc = NULL;
return pcmk_rc_unpack_error;
}
#ifndef PCMK__COMPAT_2_0
warn_about_deprecated_classes(*rsc);
#endif
(*rsc)->meta = pcmk__strkey_table(free, free);
(*rsc)->allowed_nodes = pcmk__strkey_table(NULL, free);
(*rsc)->known_on = pcmk__strkey_table(NULL, free);
value = crm_element_value((*rsc)->xml, PCMK__META_CLONE);
if (value) {
(*rsc)->id = crm_strdup_printf("%s:%s", id, value);
add_hash_param((*rsc)->meta, PCMK__META_CLONE, value);
} else {
(*rsc)->id = strdup(id);
}
(*rsc)->fns = &resource_class_functions[(*rsc)->variant];
get_meta_attributes((*rsc)->meta, *rsc, NULL, scheduler);
(*rsc)->parameters = pe_rsc_params(*rsc, NULL, scheduler); // \deprecated
(*rsc)->flags = 0;
pcmk__set_rsc_flags(*rsc, pcmk_rsc_runnable|pcmk_rsc_unassigned);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_managed);
}
(*rsc)->rsc_cons = NULL;
(*rsc)->rsc_tickets = NULL;
(*rsc)->actions = NULL;
(*rsc)->role = pcmk_role_stopped;
(*rsc)->next_role = pcmk_role_unknown;
(*rsc)->recovery_type = pcmk_multiply_active_restart;
(*rsc)->stickiness = 0;
(*rsc)->migration_threshold = INFINITY;
(*rsc)->failure_timeout = 0;
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_PRIORITY);
(*rsc)->priority = char2score(value);
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_CRITICAL);
if ((value == NULL) || crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_critical);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_NOTIFY);
if (crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_notify);
}
if (xml_contains_remote_node((*rsc)->xml)) {
(*rsc)->is_remote_node = TRUE;
if (g_hash_table_lookup((*rsc)->meta, PCMK__META_CONTAINER)) {
guest_node = true;
} else {
remote_node = true;
}
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_ALLOW_MIGRATE);
if (crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_migratable);
} else if ((value == NULL) && remote_node) {
/* By default, we want remote nodes to be able
* to float around the cluster without having to stop all the
* resources within the remote-node before moving. Allowing
* migration support enables this feature. If this ever causes
* problems, migration support can be explicitly turned off with
* PCMK_META_ALLOW_MIGRATE=false.
*/
pcmk__set_rsc_flags(*rsc, pcmk_rsc_migratable);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_IS_MANAGED);
if ((value != NULL)
&& !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
if (crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_managed);
} else {
pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed);
}
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MAINTENANCE);
if (crm_is_true(value)) {
pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed);
pcmk__set_rsc_flags(*rsc, pcmk_rsc_maintenance);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed);
pcmk__set_rsc_flags(*rsc, pcmk_rsc_maintenance);
}
if (pe_rsc_is_clone(pe__const_top_resource(*rsc, false))) {
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_GLOBALLY_UNIQUE);
if (crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_unique);
}
if (detect_promotable(*rsc)) {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_promotable);
}
} else {
pcmk__set_rsc_flags(*rsc, pcmk_rsc_unique);
}
// @COMPAT Deprecated meta-attribute
value = g_hash_table_lookup((*rsc)->meta, PCMK__META_RESTART_TYPE);
if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) {
(*rsc)->restart_type = pe_restart_restart;
pcmk__rsc_trace(*rsc, "%s dependency restart handling: restart",
(*rsc)->id);
pcmk__warn_once(pcmk__wo_restart_type,
"Support for " PCMK__META_RESTART_TYPE " is deprecated "
"and will be removed in a future release");
} else {
(*rsc)->restart_type = pe_restart_ignore;
pcmk__rsc_trace(*rsc, "%s dependency restart handling: ignore",
(*rsc)->id);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MULTIPLE_ACTIVE);
if (pcmk__str_eq(value, "stop_only", pcmk__str_casei)) {
(*rsc)->recovery_type = pcmk_multiply_active_stop;
pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop only",
(*rsc)->id);
} else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) {
(*rsc)->recovery_type = pcmk_multiply_active_block;
pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: block",
(*rsc)->id);
} else if (pcmk__str_eq(value, "stop_unexpected", pcmk__str_casei)) {
(*rsc)->recovery_type = pcmk_multiply_active_unexpected;
pcmk__rsc_trace(*rsc,
"%s multiple running resource recovery: "
"stop unexpected instances",
(*rsc)->id);
} else { // "stop_start"
if (!pcmk__str_eq(value, "stop_start",
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_warn("%s is not a valid value for "
PCMK_META_MULTIPLE_ACTIVE
", using default of \"stop_start\"", value);
}
(*rsc)->recovery_type = pcmk_multiply_active_restart;
pcmk__rsc_trace(*rsc,
"%s multiple running resource recovery: stop/start",
(*rsc)->id);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_RESOURCE_STICKINESS);
if ((value != NULL)
&& !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
(*rsc)->stickiness = char2score(value);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MIGRATION_THRESHOLD);
if ((value != NULL)
&& !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
(*rsc)->migration_threshold = char2score(value);
if ((*rsc)->migration_threshold < 0) {
/* @TODO We use 1 here to preserve previous behavior, but this
* should probably use the default (INFINITY) or 0 (to disable)
* instead.
*/
pcmk__warn_once(pcmk__wo_neg_threshold,
PCMK_META_MIGRATION_THRESHOLD
" must be non-negative, using 1 instead");
(*rsc)->migration_threshold = 1;
}
}
if (pcmk__str_eq(crm_element_value((*rsc)->xml, PCMK_XA_CLASS),
PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing);
pcmk__set_rsc_flags(*rsc, pcmk_rsc_fence_device);
}
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_REQUIRES);
unpack_requires(*rsc, value, false);
value = g_hash_table_lookup((*rsc)->meta, PCMK_META_FAILURE_TIMEOUT);
if (value != NULL) {
guint interval_ms = 0U;
// Stored as seconds
pcmk_parse_interval_spec(value, &interval_ms);
(*rsc)->failure_timeout = (int) (interval_ms / 1000);
}
if (remote_node) {
GHashTable *params = pe_rsc_params(*rsc, NULL, scheduler);
/* Grabbing the value now means that any rules based on node attributes
* will evaluate to false, so such rules should not be used with
* PCMK_REMOTE_RA_RECONNECT_INTERVAL.
*
* @TODO Evaluate per node before using
*/
value = g_hash_table_lookup(params, PCMK_REMOTE_RA_RECONNECT_INTERVAL);
if (value) {
/* reconnect delay works by setting failure_timeout and preventing the
* connection from starting until the failure is cleared. */
pcmk_parse_interval_spec(value, &((*rsc)->remote_reconnect_ms));
/* We want to override any default failure_timeout in use when remote
* PCMK_REMOTE_RA_RECONNECT_INTERVAL is in use.
*/
(*rsc)->failure_timeout = (*rsc)->remote_reconnect_ms / 1000;
}
}
get_target_role(*rsc, &((*rsc)->next_role));
pcmk__rsc_trace(*rsc, "%s desired next state: %s", (*rsc)->id,
((*rsc)->next_role == pcmk_role_unknown)?
"default" : pcmk_role_text((*rsc)->next_role));
if ((*rsc)->fns->unpack(*rsc, scheduler) == FALSE) {
(*rsc)->fns->free(*rsc);
*rsc = NULL;
return pcmk_rc_unpack_error;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
// This tag must stay exactly the same because it is tested elsewhere
resource_location(*rsc, NULL, 0, "symmetric_default", scheduler);
} else if (guest_node) {
/* remote resources tied to a container resource must always be allowed
* to opt-in to the cluster. Whether the connection resource is actually
* allowed to be placed on a node is dependent on the container resource */
resource_location(*rsc, NULL, 0, "remote_connection_default",
scheduler);
}
pcmk__rsc_trace(*rsc, "%s action notification: %s", (*rsc)->id,
pcmk_is_set((*rsc)->flags, pcmk_rsc_notify)? "required" : "not required");
(*rsc)->utilization = pcmk__strkey_table(free, free);
pe__unpack_dataset_nvpairs((*rsc)->xml, PCMK_XE_UTILIZATION, &rule_data,
(*rsc)->utilization, NULL, FALSE, scheduler);
if (expanded_xml) {
if (add_template_rsc(xml_obj, scheduler) == FALSE) {
(*rsc)->fns->free(*rsc);
*rsc = NULL;
return pcmk_rc_unpack_error;
}
}
return pcmk_rc_ok;
}
gboolean
is_parent(pcmk_resource_t *child, pcmk_resource_t *rsc)
{
pcmk_resource_t *parent = child;
if (parent == NULL || rsc == NULL) {
return FALSE;
}
while (parent->parent != NULL) {
if (parent->parent == rsc) {
return TRUE;
}
parent = parent->parent;
}
return FALSE;
}
pcmk_resource_t *
uber_parent(pcmk_resource_t *rsc)
{
pcmk_resource_t *parent = rsc;
if (parent == NULL) {
return NULL;
}
while ((parent->parent != NULL)
&& (parent->parent->variant != pcmk_rsc_variant_bundle)) {
parent = parent->parent;
}
return parent;
}
/*!
* \internal
* \brief Get the topmost parent of a resource as a const pointer
*
* \param[in] rsc Resource to check
* \param[in] include_bundle If true, go all the way to bundle
*
* \return \p NULL if \p rsc is NULL, \p rsc if \p rsc has no parent,
* the bundle if \p rsc is bundled and \p include_bundle is true,
* otherwise the topmost parent of \p rsc up to a clone
*/
const pcmk_resource_t *
pe__const_top_resource(const pcmk_resource_t *rsc, bool include_bundle)
{
const pcmk_resource_t *parent = rsc;
if (parent == NULL) {
return NULL;
}
while (parent->parent != NULL) {
if (!include_bundle
&& (parent->parent->variant == pcmk_rsc_variant_bundle)) {
break;
}
parent = parent->parent;
}
return parent;
}
void
common_free(pcmk_resource_t * rsc)
{
if (rsc == NULL) {
return;
}
pcmk__rsc_trace(rsc, "Freeing %s %d", rsc->id, rsc->variant);
g_list_free(rsc->rsc_cons);
g_list_free(rsc->rsc_cons_lhs);
g_list_free(rsc->rsc_tickets);
g_list_free(rsc->dangling_migrations);
if (rsc->parameter_cache != NULL) {
g_hash_table_destroy(rsc->parameter_cache);
}
if (rsc->meta != NULL) {
g_hash_table_destroy(rsc->meta);
}
if (rsc->utilization != NULL) {
g_hash_table_destroy(rsc->utilization);
}
if ((rsc->parent == NULL)
&& pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
free_xml(rsc->xml);
rsc->xml = NULL;
free_xml(rsc->orig_xml);
rsc->orig_xml = NULL;
/* if rsc->orig_xml, then rsc->xml is an expanded xml from a template */
} else if (rsc->orig_xml) {
free_xml(rsc->xml);
rsc->xml = NULL;
}
if (rsc->running_on) {
g_list_free(rsc->running_on);
rsc->running_on = NULL;
}
if (rsc->known_on) {
g_hash_table_destroy(rsc->known_on);
rsc->known_on = NULL;
}
if (rsc->actions) {
g_list_free(rsc->actions);
rsc->actions = NULL;
}
if (rsc->allowed_nodes) {
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = NULL;
}
g_list_free(rsc->fillers);
g_list_free(rsc->rsc_location);
pcmk__rsc_trace(rsc, "Resource freed");
free(rsc->id);
free(rsc->clone_name);
free(rsc->allocated_to);
free(rsc->variant_opaque);
free(rsc->pending_task);
free(rsc);
}
/*!
* \internal
* \brief Count a node and update most preferred to it as appropriate
*
* \param[in] rsc An active resource
* \param[in] node A node that \p rsc is active on
* \param[in,out] active This will be set to \p node if \p node is more
* preferred than the current value
* \param[in,out] count_all If not NULL, this will be incremented
* \param[in,out] count_clean If not NULL, this will be incremented if \p node
* is online and clean
*
* \return true if the count should continue, or false if sufficiently known
*/
bool
pe__count_active_node(const pcmk_resource_t *rsc, pcmk_node_t *node,
pcmk_node_t **active, unsigned int *count_all,
unsigned int *count_clean)
{
bool keep_looking = false;
bool is_happy = false;
CRM_CHECK((rsc != NULL) && (node != NULL) && (active != NULL),
return false);
is_happy = node->details->online && !node->details->unclean;
if (count_all != NULL) {
++*count_all;
}
if ((count_clean != NULL) && is_happy) {
++*count_clean;
}
if ((count_all != NULL) || (count_clean != NULL)) {
keep_looking = true; // We're counting, so go through entire list
}
if (rsc->partial_migration_source != NULL) {
if (pcmk__same_node(node, rsc->partial_migration_source)) {
*active = node; // This is the migration source
} else {
keep_looking = true;
}
} else if (!pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
if (is_happy && ((*active == NULL) || !(*active)->details->online
|| (*active)->details->unclean)) {
*active = node; // This is the first clean node
} else {
keep_looking = true;
}
}
if (*active == NULL) {
*active = node; // This is the first node checked
}
return keep_looking;
}
// Shared implementation of pcmk_rsc_methods_t:active_node()
static pcmk_node_t *
active_node(const pcmk_resource_t *rsc, unsigned int *count_all,
unsigned int *count_clean)
{
pcmk_node_t *active = NULL;
if (count_all != NULL) {
*count_all = 0;
}
if (count_clean != NULL) {
*count_clean = 0;
}
if (rsc == NULL) {
return NULL;
}
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
if (!pe__count_active_node(rsc, (pcmk_node_t *) iter->data, &active,
count_all, count_clean)) {
break; // Don't waste time iterating if we don't have to
}
}
return active;
}
/*!
* \brief
- * \internal Find and count active nodes according to "requires"
+ * \internal Find and count active nodes according to \c PCMK_META_REQUIRES
*
* \param[in] rsc Resource to check
* \param[out] count If not NULL, will be set to count of active nodes
*
* \return An active node (or NULL if resource is not active anywhere)
*
* \note This is a convenience wrapper for active_node() where the count of all
* active nodes or only clean active nodes is desired according to the
- * "requires" meta-attribute.
+ * \c PCMK_META_REQUIRES meta-attribute.
*/
pcmk_node_t *
pe__find_active_requires(const pcmk_resource_t *rsc, unsigned int *count)
{
if (rsc == NULL) {
if (count != NULL) {
*count = 0;
}
return NULL;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
return rsc->fns->active_node(rsc, count, NULL);
} else {
return rsc->fns->active_node(rsc, NULL, count);
}
}
void
pe__count_common(pcmk_resource_t *rsc)
{
if (rsc->children != NULL) {
for (GList *item = rsc->children; item != NULL; item = item->next) {
((pcmk_resource_t *) item->data)->fns->count(item->data);
}
} else if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)
|| (rsc->role > pcmk_role_stopped)) {
rsc->cluster->ninstances++;
if (pe__resource_is_disabled(rsc)) {
rsc->cluster->disabled_resources++;
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
rsc->cluster->blocked_resources++;
}
}
}
/*!
* \internal
* \brief Update a resource's next role
*
* \param[in,out] rsc Resource to be updated
* \param[in] role Resource's new next role
* \param[in] why Human-friendly reason why role is changing (for logs)
*/
void
pe__set_next_role(pcmk_resource_t *rsc, enum rsc_role_e role, const char *why)
{
CRM_ASSERT((rsc != NULL) && (why != NULL));
if (rsc->next_role != role) {
pcmk__rsc_trace(rsc, "Resetting next role for %s from %s to %s (%s)",
rsc->id, pcmk_role_text(rsc->next_role),
pcmk_role_text(role), why);
rsc->next_role = role;
}
}