diff --git a/include/crm/common/scheduler_internal.h b/include/crm/common/scheduler_internal.h
index e3b9b302b7..295ca22c49 100644
--- a/include/crm/common/scheduler_internal.h
+++ b/include/crm/common/scheduler_internal.h
@@ -1,132 +1,134 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
#define PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
#include <crm/common/action_relation_internal.h>
#include <crm/common/actions_internal.h>
#include <crm/common/attrs_internal.h>
#include <crm/common/bundles_internal.h>
#include <crm/common/clone_internal.h>
#include <crm/common/digest_internal.h>
#include <crm/common/failcounts_internal.h>
#include <crm/common/group_internal.h>
#include <crm/common/history_internal.h>
#include <crm/common/location_internal.h>
#include <crm/common/nodes_internal.h>
#include <crm/common/primitive_internal.h>
#include <crm/common/remote_internal.h>
#include <crm/common/resources_internal.h>
#include <crm/common/roles_internal.h>
#include <crm/common/rules_internal.h>
#include <crm/common/tickets_internal.h>
#ifdef __cplusplus
extern "C" {
#endif
enum pcmk__check_parameters {
/* Clear fail count if parameters changed for un-expired start or monitor
* last_failure.
*/
pcmk__check_last_failure,
/* Clear fail count if parameters changed for start, monitor, promote, or
* migrate_from actions for active resources.
*/
pcmk__check_active,
};
// Group of enum pcmk__warnings flags for warnings we want to log once
extern uint32_t pcmk__warnings;
/*!
* \internal
* \brief Log a resource-tagged message at info severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_info(rsc, fmt, args...) \
crm_log_tag(LOG_INFO, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at debug severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_debug(rsc, fmt, args...) \
crm_log_tag(LOG_DEBUG, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at trace severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_trace(rsc, fmt, args...) \
crm_log_tag(LOG_TRACE, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log an error and remember that current scheduler input has errors
*
- * \param[in] fmt... printf(3)-style format and arguments
+ * \param[in,out] scheduler Scheduler data
+ * \param[in] fmt... printf(3)-style format and arguments
*/
-#define pcmk__sched_err(fmt...) do { \
- was_processing_error = TRUE; \
- crm_err(fmt); \
+#define pcmk__sched_err(scheduler, fmt...) do { \
+ was_processing_error = TRUE; \
+ crm_err(fmt); \
} while (0)
/*!
* \internal
* \brief Log a warning and remember that current scheduler input has warnings
*
- * \param[in] fmt... printf(3)-style format and arguments
+ * \param[in,out] scheduler Scheduler data
+ * \param[in] fmt... printf(3)-style format and arguments
*/
-#define pcmk__sched_warn(fmt...) do { \
- was_processing_warning = TRUE; \
- crm_warn(fmt); \
+#define pcmk__sched_warn(scheduler, fmt...) do { \
+ was_processing_warning = TRUE; \
+ crm_warn(fmt); \
} while (0)
/*!
* \internal
* \brief Set scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_set Group of enum pcmk_scheduler_flags to set
*/
#define pcmk__set_scheduler_flags(scheduler, flags_to_set) do { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_set), #flags_to_set); \
} while (0)
/*!
* \internal
* \brief Clear scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_clear Group of enum pcmk_scheduler_flags to clear
*/
#define pcmk__clear_scheduler_flags(scheduler, flags_to_clear) do { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c
index deb443c2be..d46b78df4d 100644
--- a/lib/pacemaker/pcmk_graph_producer.c
+++ b/lib/pacemaker/pcmk_graph_producer.c
@@ -1,1105 +1,1108 @@
/*
* 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 <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.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), pcmk__action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pcmk__action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pcmk__action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->private->name)
/*!
* \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 = pcmk__xe_create(xml, PCMK_XE_NODE);
crm_xml_add(node_xml, PCMK_XA_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 pcmk_node_t *node, void *xml)
{
add_node_to_xml_by_id(node->private->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] scheduler Scheduler data
*
* \return Count of nodes added
* \note Only Pacemaker Remote nodes are considered currently
*/
static int
add_maintenance_nodes(xmlNode *xml, const pcmk_scheduler_t *scheduler)
{
xmlNode *maintenance = NULL;
int count = 0;
if (xml != NULL) {
maintenance = pcmk__xe_create(xml, PCMK__XE_MAINTENANCE);
}
for (const GList *iter = scheduler->nodes;
iter != NULL; iter = iter->next) {
const pcmk_node_t *node = iter->data;
if (!pcmk__is_pacemaker_remote_node(node)) {
continue;
}
if ((node->details->maintenance
&& !pcmk_is_set(node->private->flags, pcmk__node_remote_maint))
|| (!node->details->maintenance
&& pcmk_is_set(node->private->flags, pcmk__node_remote_maint))) {
if (maintenance != NULL) {
crm_xml_add(add_node_to_xml_by_id(node->private->id,
maintenance),
PCMK__XA_NODE_IN_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] scheduler Scheduler data
*/
static void
add_maintenance_update(pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = NULL;
if (add_maintenance_nodes(NULL, scheduler) != 0) {
action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, scheduler);
pcmk__set_action_flags(action, pcmk__action_always_in_graph);
}
}
/*!
* \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 pcmk_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 = pcmk__xe_create(xml, PCMK__XE_DOWNED);
add_node_to_xml_by_id(action->node->private->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,
PCMK__META_STONITH_ACTION);
if (pcmk__is_fencing_action(fence)) {
xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED);
add_node_to_xml_by_id(action->node->private->id, downed);
pe_foreach_guest_node(action->node->private->scheduler,
action->node, add_node_to_xml, downed);
}
} else if ((action->rsc != NULL)
&& pcmk_is_set(action->rsc->flags,
pcmk__rsc_is_remote_connection)
&& 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;
pcmk_action_t *input;
bool migrating = false;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((pcmk__related_action_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 = pcmk__xe_create(xml, PCMK__XE_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 pcmk_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");
return pcmk__notify_key(action->rsc->private->history_id, n_type,
n_task);
}
return pcmk__op_key(action->rsc->private->history_id,
pcmk__s(action->cancel_task, 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 pcmk_action_t *action, xmlNode *xml)
{
pcmk_node_t *router_node = pcmk__connection_host_for_action(action);
crm_xml_add(xml, PCMK__META_ON_NODE, action->node->private->name);
crm_xml_add(xml, PCMK__META_ON_NODE_UUID, action->node->private->id);
if (router_node != NULL) {
crm_xml_add(xml, PCMK__XA_ROUTER_NODE, router_node->private->name);
}
}
/*!
* \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 pcmk_action_t *action, xmlNode *action_xml)
{
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
PCMK_XA_CLASS,
PCMK_XA_PROVIDER,
PCMK_XA_TYPE,
};
/* If a resource is locked to a node via PCMK_OPT_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, PCMK_OPT_SHUTDOWN_LOCK,
(long long) action->rsc->private->lock_time);
}
// List affected resource
rsc_xml = pcmk__xe_create(action_xml,
(const char *) action->rsc->private->xml->name);
if (pcmk_is_set(action->rsc->flags, pcmk__rsc_removed)
&& (action->rsc->private->history_id != 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 history ID %s",
action->rsc->id, action->rsc->private->history_id);
crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->private->history_id);
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id);
} else if (!pcmk_is_set(action->rsc->flags, pcmk__rsc_unique)) {
const char *xml_id = pcmk__xe_id(action->rsc->private->xml);
crm_debug("Using anonymous clone name %s for %s (aka %s)",
xml_id, action->rsc->id, action->rsc->private->history_id);
/* ID is what we'd like client to use
* LONG_ID is what they might know it as instead
*
* LONG_ID 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, PCMK_XA_ID, xml_id);
if ((action->rsc->private->history_id != NULL)
&& !pcmk__str_eq(xml_id, action->rsc->private->history_id,
pcmk__str_none)) {
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID,
action->rsc->private->history_id);
} else {
crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id);
}
} else {
CRM_ASSERT(action->rsc->private->history_id == NULL);
crm_xml_add(rsc_xml, PCMK_XA_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->private->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(pcmk_action_t *action, xmlNode *action_xml)
{
xmlNode *args_xml = NULL;
pcmk_resource_t *rsc = action->rsc;
/* 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 = pcmk__xe_create(action_xml, PCMK__XE_ATTRIBUTES);
crm_xml_add(args_xml, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET);
g_hash_table_foreach(action->extra, hash2field, args_xml);
if ((rsc != NULL) && (action->node != NULL)) {
// Get the resource instance attributes, evaluated properly for node
GHashTable *params = pe_rsc_params(rsc, action->node,
rsc->private->scheduler);
pcmk__substitute_remote_addr(rsc, params);
g_hash_table_foreach(params, hash2smartfield, args_xml);
} else if ((rsc != NULL)
&& (rsc->private->variant <= pcmk__rsc_variant_primitive)) {
GHashTable *params = pe_rsc_params(rsc, NULL, rsc->private->scheduler);
g_hash_table_foreach(params, hash2smartfield, args_xml);
}
g_hash_table_foreach(action->meta, hash2metafield, args_xml);
if (rsc != NULL) {
pcmk_resource_t *parent = rsc;
while (parent != NULL) {
parent->private->cmds->add_graph_meta(parent, args_xml);
parent = parent->private->parent;
}
pcmk__add_guest_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->private->attrs, hash2metafield,
args_xml);
}
pcmk__xe_sort_attrs(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] scheduler Scheduler data
*/
static void
create_graph_action(xmlNode *parent, pcmk_action_t *action, bool skip_details,
const pcmk_scheduler_t *scheduler)
{
bool needs_node_info = true;
bool needs_maintenance_info = false;
xmlNode *action_xml = NULL;
if ((action == NULL) || (scheduler == 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, pcmk__action_pseudo)) {
action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT);
} else {
action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT);
}
} else if (pcmk__str_any_of(action->task,
PCMK_ACTION_DO_SHUTDOWN,
PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) {
action_xml = pcmk__xe_create(parent, PCMK__XE_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 = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT);
crm_xml_add(action_xml, PCMK__XA_MODE, PCMK__VALUE_CIB);
} else if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES,
pcmk__str_none)) {
needs_maintenance_info = true;
}
action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT);
needs_node_info = false;
} else {
action_xml = pcmk__xe_create(parent, PCMK__XE_RSC_OP);
}
crm_xml_add_int(action_xml, PCMK_XA_ID, action->id);
crm_xml_add(action_xml, PCMK_XA_OPERATION, action->task);
if ((action->rsc != NULL) && (action->rsc->private->history_id != NULL)) {
char *clone_key = NULL;
guint interval_ms;
if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
clone_key = clone_op_key(action, interval_ms);
crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, clone_key);
crm_xml_add(action_xml, "internal_" PCMK__XA_OPERATION_KEY,
action->uuid);
free(clone_key);
} else {
crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, action->uuid);
}
if (needs_node_info && (action->node != NULL)) {
add_node_details(action, action_xml);
pcmk__insert_dup(action->meta, PCMK__META_ON_NODE,
action->node->private->name);
pcmk__insert_dup(action->meta, PCMK__META_ON_NODE_UUID,
action->node->private->id);
}
if (skip_details) {
return;
}
if ((action->rsc != NULL)
&& !pcmk_is_set(action->flags, pcmk__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, scheduler);
}
}
/*!
* \internal
* \brief Check whether an action should be added to the transition graph
*
- * \param[in] action Action to check
+ * \param[in,out] action Action to check
*
* \return true if action should be added to graph, otherwise false
*/
static bool
-should_add_action_to_graph(const pcmk_action_t *action)
+should_add_action_to_graph(pcmk_action_t *action)
{
if (!pcmk_is_set(action->flags, pcmk__action_runnable)) {
crm_trace("Ignoring action %s (%d): unrunnable",
action->uuid, action->id);
return false;
}
if (pcmk_is_set(action->flags, pcmk__action_optional)
&& !pcmk_is_set(action->flags, pcmk__action_always_in_graph)) {
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, pcmk__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, PCMK_META_INTERVAL);
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, pcmk__action_pseudo)
|| pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH,
PCMK_ACTION_DO_SHUTDOWN, NULL)) {
return true;
}
if (action->node == NULL) {
- pcmk__sched_err("Skipping action %s (%d) "
+ pcmk__sched_err(action->scheduler,
+ "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, pcmk__action_on_dc)) {
crm_trace("Action %s (%d) should be dumped: "
"can run on DC instead of %s",
action->uuid, action->id, pcmk__node_name(action->node));
} else if (pcmk__is_guest_or_bundle_node(action->node)
&& !pcmk_is_set(action->node->private->flags,
pcmk__node_remote_reset)) {
crm_trace("Action %s (%d) should be dumped: "
"assuming will be runnable on guest %s",
action->uuid, action->id, pcmk__node_name(action->node));
} else if (!action->node->details->online) {
- pcmk__sched_err("Skipping action %s (%d) "
+ pcmk__sched_err(action->scheduler,
+ "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) {
- pcmk__sched_err("Skipping action %s (%d) "
+ pcmk__sched_err(action->scheduler,
+ "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 pcmk__related_action_t *ordering)
{
return pcmk_any_flags_set(ordering->flags,
~(pcmk__ar_then_implies_first_graphed
|pcmk__ar_first_implies_then_graphed
|pcmk__ar_ordered));
}
/*!
* \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 pcmk_action_t *action,
pcmk__related_action_t *input)
{
if (input->graphed) {
return true;
}
if (input->flags == pcmk__ar_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, pcmk__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, pcmk__action_runnable)
&& pcmk_is_set(input->flags, pcmk__ar_min_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"minimum number of instances required but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->flags, pcmk__ar_unmigratable_then_blocks)
&& !pcmk_is_set(input->action->flags, pcmk__action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input blocked if 'then' unmigratable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->flags, pcmk__ar_if_first_unmigratable)
&& pcmk_is_set(input->action->flags, pcmk__action_migratable)) {
crm_trace("Ignoring %s (%d) input %s (%d): ordering applies "
"only if input is unmigratable, but it is migratable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->flags == pcmk__ar_ordered)
&& pcmk_is_set(input->action->flags, pcmk__action_migratable)
&& 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->flags == pcmk__ar_if_on_same_node_or_target) {
pcmk_node_t *input_node = input->action->node;
if ((action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
pcmk_node_t *assigned = action->rsc->private->assigned_node;
/* For load_stopped -> migrate_to orderings, we care about where
* the resource has been assigned, not where migrate_to will be
* executed.
*/
if (!pcmk__same_node(input_node, assigned)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"migration target %s is not same as input node %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(assigned? assigned->private->name : "<none>"),
(input_node? input_node->private->name : "<none>"));
input->flags = pcmk__ar_none;
return false;
}
} else if (!pcmk__same_node(input_node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"not on same node (%s vs %s)",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->private->name : "<none>"),
(input_node? input_node->private->name : "<none>"));
input->flags = pcmk__ar_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->flags = pcmk__ar_none;
return false;
}
} else if (input->flags == pcmk__ar_if_required_on_same_node) {
if (input->action->node && action->node
&& !pcmk__same_node(input->action->node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"not on same node (%s vs %s)",
action->uuid, action->id,
input->action->uuid, input->action->id,
pcmk__node_name(action->node),
pcmk__node_name(input->action->node));
input->flags = pcmk__ar_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->flags = pcmk__ar_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& pcmk_is_set(input->action->rsc->flags, pcmk__rsc_failed)
&& !pcmk_is_set(input->action->rsc->flags, pcmk__rsc_managed)
&& pcmk__ends_with(input->action->uuid, "_stop_0")
&& pcmk__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, pcmk__action_optional)
&& !pcmk_any_flags_set(input->action->flags,
pcmk__action_always_in_graph
|pcmk__action_added_to_graph)
&& !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->flags);
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 pcmk_action_t *init_action,
const pcmk_action_t *action, pcmk__related_action_t *input)
{
bool has_loop = false;
if (pcmk_is_set(input->action->flags, pcmk__action_detect_loop)) {
crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->private->name : "",
action->uuid,
action->node? action->node->private->name : "",
input->flags);
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->private->name : "",
init_action->uuid,
init_action->node? init_action->node->private->name : "");
return true;
}
pcmk__set_action_flags(input->action, pcmk__action_detect_loop);
crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)"
"for graph loop with %s@%s ",
action->uuid,
action->node? action->node->private->name : "",
input->action->uuid,
input->action->node? input->action->node->private->name : "",
input->flags,
init_action->uuid,
init_action->node? init_action->node->private->name : "");
// 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,
(pcmk__related_action_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
break;
}
}
pcmk__clear_action_flags(input->action, pcmk__action_detect_loop);
if (!has_loop) {
crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->private->name : "",
action->uuid,
action->node? action->node->private->name : "",
input->flags);
}
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] scheduler Scheduler data containing graph
*
* \return Newly added XML element for new graph synapse
*/
static xmlNode *
create_graph_synapse(const pcmk_action_t *action, pcmk_scheduler_t *scheduler)
{
int synapse_priority = 0;
xmlNode *syn = pcmk__xe_create(scheduler->graph, "synapse");
crm_xml_add_int(syn, PCMK_XA_ID, scheduler->num_synapse);
scheduler->num_synapse++;
if (action->rsc != NULL) {
synapse_priority = action->rsc->private->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, PCMK__XA_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 Scheduler data
*
* \note This will de-duplicate the action inputs, meaning that the
* pcmk__related_action_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)
{
pcmk_action_t *action = (pcmk_action_t *) data;
pcmk_scheduler_t *scheduler = (pcmk_scheduler_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, pcmk__action_inputs_deduplicated)) {
pcmk__deduplicate_action_inputs(action);
pcmk__set_action_flags(action, pcmk__action_inputs_deduplicated);
}
if (pcmk_is_set(action->flags, pcmk__action_added_to_graph)
|| !should_add_action_to_graph(action)) {
return; // Already added, or shouldn't be
}
pcmk__set_action_flags(action, pcmk__action_added_to_graph);
crm_trace("Adding action %d (%s%s%s) to graph",
action->id, action->uuid,
((action->node == NULL)? "" : " on "),
((action->node == NULL)? "" : action->node->private->name));
syn = create_graph_synapse(action, scheduler);
set = pcmk__xe_create(syn, "action_set");
in = pcmk__xe_create(syn, "inputs");
create_graph_action(set, action, false, scheduler);
for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *input = lpc->data;
if (should_add_input_to_graph(action, input)) {
xmlNode *input_xml = pcmk__xe_create(in, "trigger");
input->graphed = true;
create_graph_action(input_xml, input->action, true, scheduler);
}
}
}
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_config_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_config_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(pcmk_resource_t *rsc)
{
GList *iter = NULL;
CRM_ASSERT(rsc != NULL);
pcmk__rsc_trace(rsc, "Adding actions for %s to graph", rsc->id);
// First add the resource's own actions
g_list_foreach(rsc->private->actions, add_action_to_graph,
rsc->private->scheduler);
// Then recursively add its children's actions (appropriate to variant)
for (iter = rsc->private->children; iter != NULL; iter = iter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data;
child_rsc->private->cmds->add_actions_to_graph(child_rsc);
}
}
/*!
* \internal
* \brief Create a transition graph with all cluster actions needed
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__create_graph(pcmk_scheduler_t *scheduler)
{
GList *iter = NULL;
const char *value = NULL;
long long limit = 0LL;
GHashTable *config_hash = scheduler->config_hash;
transition_id++;
crm_trace("Creating transition graph %d", transition_id);
scheduler->graph = pcmk__xe_create(NULL, PCMK__XE_TRANSITION_GRAPH);
value = pcmk__cluster_option(config_hash, PCMK_OPT_CLUSTER_DELAY);
crm_xml_add(scheduler->graph, PCMK_OPT_CLUSTER_DELAY, value);
value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT);
crm_xml_add(scheduler->graph, PCMK_OPT_STONITH_TIMEOUT, value);
crm_xml_add(scheduler->graph, "failed-stop-offset", "INFINITY");
if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) {
crm_xml_add(scheduler->graph, "failed-start-offset", "INFINITY");
} else {
crm_xml_add(scheduler->graph, "failed-start-offset", "1");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_BATCH_LIMIT);
crm_xml_add(scheduler->graph, PCMK_OPT_BATCH_LIMIT, value);
crm_xml_add_int(scheduler->graph, "transition_id", transition_id);
value = pcmk__cluster_option(config_hash, PCMK_OPT_MIGRATION_LIMIT);
if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) {
crm_xml_add(scheduler->graph, PCMK_OPT_MIGRATION_LIMIT, value);
}
if (scheduler->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu",
(long long) scheduler->recheck_by);
crm_xml_add(scheduler->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 = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
pcmk__rsc_trace(rsc, "Processing actions for %s", rsc->id);
rsc->private->cmds->add_actions_to_graph(rsc);
}
// Add pseudo-action for list of nodes with maintenance state update
add_maintenance_update(scheduler);
// Add non-resource (node) actions
for (iter = scheduler->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
if ((action->rsc != NULL)
&& (action->node != NULL)
&& action->node->details->shutdown
&& !pcmk_is_set(action->rsc->flags, pcmk__rsc_maintenance)
&& !pcmk_any_flags_set(action->flags,
pcmk__action_optional|pcmk__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(scheduler->flags, pcmk_sched_quorate)
|| (scheduler->no_quorum_policy == pcmk_no_quorum_ignore)) {
const bool managed = pcmk_is_set(action->rsc->flags,
pcmk__rsc_managed);
const bool failed = pcmk_is_set(action->rsc->flags,
pcmk__rsc_failed);
crm_crit("Cannot %s %s because of %s:%s%s (%s)",
action->node->details->unclean? "fence" : "shut down",
pcmk__node_name(action->node), action->rsc->id,
(managed? " blocked" : " unmanaged"),
(failed? " failed" : ""), action->uuid);
}
}
add_action_to_graph((gpointer) action, (gpointer) scheduler);
}
crm_log_xml_trace(scheduler->graph, "graph");
}
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index 6713c7df51..1b8ae71cd0 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1716 +1,1718 @@
/*
* 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/common/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->private->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->private->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->private->assigned_node != 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->private->allowed_nodes,
prefer->private->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->assign->score < best->assign->score) {
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 (!pcmk__is_unique_clone(rsc->private->parent)
&& (chosen->assign->score > 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->assign->score != chosen->assign->score) {
// 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->assign->score >= PCMK_SCORE_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->assign->score));
}
}
}
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->private->assigned_node != 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 > -PCMK_SCORE_INFINITY))) {
archive = pcmk__copy_node_table(rsc->private->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->private->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->private->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->private->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->private->allowed_nodes);
rsc->private->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 = pcmk_find_node(connection->private->scheduler,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->private->assigned_node != NULL)
&& (connection->private->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->private->id);
remote_node->details->online = TRUE;
if (!pcmk_is_set(remote_node->private->flags, pcmk__node_seen)) {
// 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->private->id,
((connection->private->assigned_node == NULL)? "un" : ""),
pcmk_role_text(connection->private->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_resource_t *parent = NULL;
pcmk__colocation_t *colocation = NULL;
pcmk_scheduler_t *scheduler = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc));
scheduler = rsc->private->scheduler;
parent = rsc->private->parent;
// Never assign a child without parent being assigned first
if ((parent != NULL) && !pcmk_is_set(parent->flags, pcmk__rsc_assigning)) {
pcmk__rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, parent->id);
parent->private->cmds->assign(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->private->assigned_node != NULL) {
node_name = pcmk__node_name(rsc->private->assigned_node);
}
pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->private->assigned_node;
}
// 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->private->allowed_nodes, scheduler);
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 <= -PCMK_SCORE_INFINITY)
|| (colocation->score >= PCMK_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -PCMK_SCORE_INFINITY)
|| (colocation->score >= PCMK_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->private->allowed_nodes, scheduler);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -PCMK_SCORE_INFINITY)
&& (colocation->score < PCMK_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -PCMK_SCORE_INFINITY)
&& (colocation->score < PCMK_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->private->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, -PCMK_SCORE_INFINITY,
PCMK_META_TARGET_ROLE, scheduler);
} else if ((rsc->private->next_role > rsc->private->orig_role)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_quorate)
&& (scheduler->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->private->orig_role),
pcmk_role_text(rsc->private->next_role));
pe__set_next_role(rsc, rsc->private->orig_role,
PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE);
}
pe__show_node_scores(!pcmk_is_set(scheduler->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->private->allowed_nodes, scheduler);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(scheduler->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->private->orig_role, "unmanaged");
assign_to = pcmk__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->private->orig_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->private->name : "no node"),
reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(scheduler->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->private->active_nodes != NULL) && stop_if_fail) {
pcmk__rsc_info(rsc, "Stopping removed resource %s", rsc->id);
}
}
pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning);
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
remote_connection_assigned(rsc);
}
return rsc->private->assigned_node;
}
/*!
* \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->private->orig_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->private->orig_role <= rsc->private->next_role)
&& (role != rsc->private->orig_role)
&& !pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->private->orig_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->private->assigned_node, !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->private->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->private->assigned_node == 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->private->assigned_node, 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->private->orig_role;
while (role != rsc->private->next_role) {
enum rsc_role_e next_role =
rsc_state_matrix[role][rsc->private->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->private->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->private->assigned_node, 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;
pcmk_node_t *migration_target = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc));
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->private->orig_role),
pcmk_role_text(rsc->private->next_role), next_role_source,
pcmk__node_name(rsc->private->assigned_node));
current = rsc->private->fns->active_node(rsc, &num_all_active,
&num_clean_active);
g_list_foreach(rsc->private->dangling_migration_sources,
pcmk__abort_dangling_migration, rsc);
if ((current != NULL) && (rsc->private->assigned_node != NULL)
&& !pcmk__same_node(current, rsc->private->assigned_node)
&& (rsc->private->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->private->assigned_node));
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
migration_target = rsc->private->partial_migration_target;
if ((rsc->private->partial_migration_source != NULL)
&& (migration_target != NULL) && allow_migrate && (num_all_active == 2)
&& pcmk__same_node(current, rsc->private->partial_migration_source)
&& pcmk__same_node(rsc->private->assigned_node, 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->private->partial_migration_source),
pcmk__node_name(migration_target));
} else if ((rsc->private->partial_migration_source != NULL)
|| (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->private->partial_migration_source),
pcmk__node_name(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->private->partial_migration_source),
pcmk__node_name(migration_target));
}
need_stop = true;
rsc->private->partial_migration_source = NULL;
rsc->private->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 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->private->xml, PCMK_XA_CLASS);
// Resource was (possibly) incorrectly multiply active
- pcmk__sched_err("%s resource %s might be active on %u nodes (%s)",
+ pcmk__sched_err(rsc->private->scheduler,
+ "%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), rsc->id, num_all_active,
pcmk__multiply_active_text(rsc));
crm_notice("For more information, see \"What are multiply active "
"resources?\" at "
"https://projects.clusterlabs.org/w/clusterlabs/faq/");
switch (rsc->private->multiply_active_policy) {
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->private->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->private->orig_role > pcmk_role_started)
&& (current != NULL)
&& (rsc->private->assigned_node != NULL)) {
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->private->assigned_node, 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->private->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (node->private->remote != NULL) {
node->assign->score = -PCMK_SCORE_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->private->allowed_nodes != NULL) {
allowed_nodes = g_hash_table_get_values(rsc->private->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;
pcmk_scheduler_t *scheduler = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc));
scheduler = rsc->private->scheduler;
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(scheduler->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->private->utilization) > 0)
&& !pcmk__str_eq(scheduler->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, scheduler);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_promotable)
|| (rsc->private->orig_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, scheduler);
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, scheduler);
}
// 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,
scheduler);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization
|| (rsc->private->launcher != 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->private->launcher != NULL) {
pcmk_resource_t *remote_rsc = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
// 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->private->launcher);
}
/* If someone cleans up a guest or bundle node's launcher, we will
* likely schedule a (re-)probe of the launcher and recovery of the
* connection. Order the connection stop after the launcher probe,
* so that if we detect the launcher running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->private->launcher,
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 PCMK__META_CONTAINER
* 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 PCMK__META_CONTAINER set to a remote node or guest node resource.
*/
} else if (pcmk_is_set(rsc->private->launcher->flags,
pcmk__rsc_is_remote_connection)) {
remote_rsc = rsc->private->launcher;
} else {
remote_rsc =
pe__resource_contains_guest_node(scheduler,
rsc->private->launcher);
}
if (remote_rsc != NULL) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the launcher.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pcmk_node_t *node = item->data;
if (node->private->remote != remote_rsc) {
node->assign->score = -PCMK_SCORE_INFINITY;
}
}
} else {
/* This resource is either launched by a resource 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 launcher %s",
rsc->id, rsc->private->launcher->id);
pcmk__new_ordering(rsc->private->launcher,
pcmk__op_key(rsc->private->launcher->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, scheduler);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->private->launcher,
pcmk__op_key(rsc->private->launcher->id,
PCMK_ACTION_STOP, 0),
NULL, pcmk__ar_then_implies_first, scheduler);
if (pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = PCMK_SCORE_INFINITY; // Force to run on same host
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->private->launcher, NULL, NULL,
pcmk__coloc_influence);
}
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)
|| 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->private->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)
{
const pcmk_resource_t *parent = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc) && (list != NULL));
parent = rsc->private->parent;
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->private->with_this_colocations,
orig_rsc);
if (parent != NULL) {
parent->private->cmds->with_this_colocations(parent, orig_rsc,
list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->private->with_this_colocations;
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)
{
const pcmk_resource_t *parent = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc) && (list != NULL));
parent = rsc->private->parent;
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->private->this_with_colocations,
orig_rsc);
if (parent != NULL) {
parent->private->cmds->this_with_colocations(parent, orig_rsc,
list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->private->this_with_colocations;
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 PCMK_VALUE_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->private->next_role > pcmk_role_stopped)
&& pcmk__same_node(rsc->private->assigned_node, 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->private->active_nodes;
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=PCMK_VALUE_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->private->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pcmk__same_node(current, rsc->private->partial_migration_target)
&& pcmk__same_node(current, rsc->private->assigned_node)) {
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->private->assigned_node == 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->private->scheduler->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->private->scheduler);
order_actions(stop, unfence, pcmk__ar_then_implies_first);
if (!pcmk__node_unfenced(current)) {
- pcmk__sched_err("Stopping %s until %s can be unfenced",
+ pcmk__sched_err(rsc->private->scheduler,
+ "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->assign->score);
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->private->active_nodes;
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(pcmk__is_primitive(rsc) && (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->private->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->private->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 PCMK__META_CONTAINER meta-attribute can be set on the primitive
* itself or one of its ancestors, so check them all and keep the highest.
*/
for (parent = rsc; parent != NULL; parent = parent->private->parent) {
if (parent->private->launcher != NULL) {
crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER,
parent->private->launcher->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->private->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(pcmk__is_primitive(rsc)
&& (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 = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL,
pcmk__rsc_node_current);
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->private->scheduler) : 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 (!pcmk__same_node(node, rsc->private->lock_node)) {
resource_location(rsc, node, -PCMK_SCORE_INFINITY,
PCMK_OPT_SHUTDOWN_LOCK, rsc->private->scheduler);
}
}
// Primitive implementation of pcmk__assignment_methods_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc)
{
const char *class = NULL;
pcmk_scheduler_t *scheduler = NULL;
CRM_ASSERT(pcmk__is_primitive(rsc));
scheduler = rsc->private->scheduler;
class = crm_element_value(rsc->private->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)
|| pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
return;
}
if (rsc->private->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->private->active_nodes != 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->private->lock_node);
rsc->private->lock_node = NULL;
rsc->private->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->private->active_nodes)) {
pcmk_node_t *node = rsc->private->active_nodes->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->private->lock_node = node;
rsc->private->lock_time = shutdown_time(node);
}
}
}
if (rsc->private->lock_node == NULL) {
// No lock needed
return;
}
if (scheduler->shutdown_lock > 0) {
time_t lock_expiration;
lock_expiration = rsc->private->lock_time + scheduler->shutdown_lock;
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pcmk__node_name(rsc->private->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, scheduler,
"shutdown lock expiration");
} else {
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pcmk__node_name(rsc->private->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(scheduler->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pacemaker/pcmk_sched_promotable.c b/lib/pacemaker/pcmk_sched_promotable.c
index 13144b9a61..1e4dc13691 100644
--- a/lib/pacemaker/pcmk_sched_promotable.c
+++ b/lib/pacemaker/pcmk_sched_promotable.c
@@ -1,1378 +1,1379 @@
/*
* 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 <crm/common/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(pcmk_resource_t *clone, pcmk_resource_t *child,
pcmk_resource_t *last)
{
// "Promote clone" -> promote instance -> "clone promoted"
pcmk__order_resource_actions(clone, PCMK_ACTION_PROMOTE,
child, PCMK_ACTION_PROMOTE,
pcmk__ar_ordered);
pcmk__order_resource_actions(child, PCMK_ACTION_PROMOTE,
clone, PCMK_ACTION_PROMOTED,
pcmk__ar_ordered);
// 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,
pcmk__ar_ordered);
}
}
/*!
* \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(pcmk_resource_t *clone, pcmk_resource_t *child,
pcmk_resource_t *last)
{
// "Demote clone" -> demote instance -> "clone demoted"
pcmk__order_resource_actions(clone, PCMK_ACTION_DEMOTE, child,
PCMK_ACTION_DEMOTE,
pcmk__ar_then_implies_first_graphed);
pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE,
clone, PCMK_ACTION_DEMOTED,
pcmk__ar_first_implies_then_graphed);
// 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, pcmk__ar_ordered);
}
}
/*!
* \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 pcmk_resource_t *rsc, bool *demoting,
bool *promoting)
{
const GList *iter = NULL;
// If this is a cloned group, check group members recursively
if (rsc->private->children != NULL) {
for (iter = rsc->private->children; iter != NULL; iter = iter->next) {
check_for_role_change((const pcmk_resource_t *) iter->data,
demoting, promoting);
}
return;
}
for (iter = rsc->private->actions; iter != NULL; iter = iter->next) {
const pcmk_action_t *action = (const pcmk_action_t *) iter->data;
if (*promoting && *demoting) {
return;
} else if (pcmk_is_set(action->flags, pcmk__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(pcmk_resource_t *child,
const GList *location_constraints,
const pcmk_node_t *chosen)
{
for (const GList *iter = location_constraints; iter; iter = iter->next) {
const pcmk__location_t *location = iter->data;
const pcmk_node_t *constraint_node = NULL;
if (location->role_filter == pcmk_role_promoted) {
constraint_node = pe_find_node_id(location->nodes,
chosen->private->id);
}
if (constraint_node != NULL) {
int new_priority = pcmk__add_scores(child->private->priority,
constraint_node->assign->score);
pcmk__rsc_trace(child,
"Applying location %s to %s promotion priority on "
"%s: %s + %s = %s",
location->id, child->id,
pcmk__node_name(constraint_node),
pcmk_readable_score(child->private->priority),
pcmk_readable_score(constraint_node->assign->score),
pcmk_readable_score(new_priority));
child->private->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 pcmk_node_t *
node_to_be_promoted_on(const pcmk_resource_t *rsc)
{
pcmk_node_t *node = NULL;
pcmk_node_t *local_node = NULL;
const pcmk_resource_t *parent = NULL;
// If this is a cloned group, bail if any group member can't be promoted
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) iter->data;
if (node_to_be_promoted_on(child) == NULL) {
pcmk__rsc_trace(rsc,
"%s can't be promoted because member %s can't",
rsc->id, child->id);
return NULL;
}
}
node = rsc->private->fns->location(rsc, NULL, FALSE);
if (node == NULL) {
pcmk__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, pcmk__rsc_managed)) {
if (rsc->private->fns->state(rsc, TRUE) == pcmk_role_promoted) {
crm_notice("Unmanaged instance %s will be left promoted on %s",
rsc->id, pcmk__node_name(node));
} else {
pcmk__rsc_trace(rsc, "%s can't be promoted because it is unmanaged",
rsc->id);
return NULL;
}
} else if (rsc->private->priority < 0) {
pcmk__rsc_trace(rsc,
"%s can't be promoted because its promotion priority "
"%d is negative",
rsc->id, rsc->private->priority);
return NULL;
} else if (!pcmk__node_available(node, false, true)) {
pcmk__rsc_trace(rsc,
"%s can't be promoted because %s can't run resources",
rsc->id, pcmk__node_name(node));
return NULL;
}
parent = pe__const_top_resource(rsc, false);
local_node = g_hash_table_lookup(parent->private->allowed_nodes,
node->private->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, pcmk__rsc_managed)) {
- pcmk__sched_err("%s can't be promoted because %s is not allowed "
+ pcmk__sched_err(node->private->scheduler,
+ "%s can't be promoted because %s is not allowed "
"on %s (scheduler bug?)",
rsc->id, parent->id, pcmk__node_name(node));
} // else the instance is unmanaged and already promoted
return NULL;
} else if ((local_node->assign->count >= pe__clone_promoted_node_max(parent))
&& pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
pcmk__rsc_trace(rsc,
"%s can't be promoted because %s has "
"maximum promoted instances already",
rsc->id, pcmk__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 pcmk_resource_t *rsc1 = (const pcmk_resource_t *) a;
const pcmk_resource_t *rsc2 = (const pcmk_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 promotion priority set by pcmk__set_instance_roles()
if (rsc1->private->promotion_priority > rsc2->private->promotion_priority) {
pcmk__rsc_trace(rsc1,
"%s has higher promotion priority (%s) than %s (%d)",
rsc1->id,
pcmk_readable_score(rsc1->private->promotion_priority),
rsc2->id, rsc2->private->promotion_priority);
return -1;
}
if (rsc1->private->promotion_priority < rsc2->private->promotion_priority) {
pcmk__rsc_trace(rsc1,
"%s has lower promotion priority (%s) than %s (%d)",
rsc1->id,
pcmk_readable_score(rsc1->private->promotion_priority),
rsc2->id, rsc2->private->promotion_priority);
return 1;
}
// If those are the same, prefer instance whose current role is higher
role1 = rsc1->private->fns->state(rsc1, TRUE);
role2 = rsc2->private->fns->state(rsc2, TRUE);
if (role1 > role2) {
pcmk__rsc_trace(rsc1,
"%s has higher promotion priority than %s "
"(higher current role)",
rsc1->id, rsc2->id);
return -1;
} else if (role1 < role2) {
pcmk__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 promotable clone instance's promotion priority to its node's score
*
* Add a promotable clone instance's promotion priority (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_promotion_priority_to_node_score(gpointer data, gpointer user_data)
{
const pcmk_resource_t *child = (const pcmk_resource_t *) data;
pcmk_resource_t *clone = (pcmk_resource_t *) user_data;
pcmk_node_t *node = NULL;
const pcmk_node_t *chosen = NULL;
const int promotion_priority = child->private->promotion_priority;
if (promotion_priority < 0) {
pcmk__rsc_trace(clone,
"Not adding promotion priority of %s: negative (%s)",
child->id, pcmk_readable_score(promotion_priority));
return;
}
chosen = child->private->fns->location(child, NULL, FALSE);
if (chosen == NULL) {
pcmk__rsc_trace(clone, "Not adding promotion priority of %s: inactive",
child->id);
return;
}
node = g_hash_table_lookup(clone->private->allowed_nodes,
chosen->private->id);
CRM_ASSERT(node != NULL);
node->assign->score = pcmk__add_scores(promotion_priority,
node->assign->score);
pcmk__rsc_trace(clone,
"Added cumulative priority of %s (%s) to score on %s "
"(now %d)",
child->id, pcmk_readable_score(promotion_priority),
pcmk__node_name(node), node->assign->score);
}
/*!
* \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;
pcmk_resource_t *clone = user_data;
pcmk_resource_t *primary = colocation->primary;
uint32_t flags = pcmk__coloc_select_default;
float factor = colocation->score / (float) PCMK_SCORE_INFINITY;
if (colocation->dependent_role != pcmk_role_promoted) {
return;
}
if (colocation->score < PCMK_SCORE_INFINITY) {
flags = pcmk__coloc_select_active;
}
pcmk__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->private->cmds->add_colocated_node_scores(primary, clone, clone->id,
&(clone->private->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;
pcmk_resource_t *clone = user_data;
pcmk_resource_t *dependent = colocation->dependent;
const float factor = colocation->score / (float) PCMK_SCORE_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;
}
pcmk__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->private->cmds->add_colocated_node_scores(dependent, clone,
clone->id,
&(clone->private->allowed_nodes),
colocation, factor,
flags);
}
/*!
* \internal
* \brief Set clone instance's promotion priority to its node's score
*
* \param[in,out] data Promotable clone instance
* \param[in] user_data Parent clone of \p data
*/
static void
set_promotion_priority_to_node_score(gpointer data, gpointer user_data)
{
pcmk_resource_t *child = (pcmk_resource_t *) data;
const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data;
pcmk_node_t *chosen = child->private->fns->location(child, NULL, FALSE);
if (!pcmk_is_set(child->flags, pcmk__rsc_managed)
&& (child->private->next_role == pcmk_role_promoted)) {
child->private->promotion_priority = PCMK_SCORE_INFINITY;
pcmk__rsc_trace(clone,
"Final promotion priority for %s is %s "
"(unmanaged promoted)",
child->id, pcmk_readable_score(PCMK_SCORE_INFINITY));
} else if (chosen == NULL) {
child->private->promotion_priority = -PCMK_SCORE_INFINITY;
pcmk__rsc_trace(clone,
"Final promotion priority for %s is %s "
"(will not be active)",
child->id, pcmk_readable_score(-PCMK_SCORE_INFINITY));
} else if (child->private->promotion_priority < 0) {
pcmk__rsc_trace(clone,
"Final promotion priority for %s is %s "
"(ignoring node score)",
child->id,
pcmk_readable_score(child->private->promotion_priority));
} else {
const pcmk_node_t *node = NULL;
node = g_hash_table_lookup(clone->private->allowed_nodes,
chosen->private->id);
CRM_ASSERT(node != NULL);
child->private->promotion_priority = node->assign->score;
pcmk__rsc_trace(clone,
"Adding scores for %s: "
"final promotion priority for %s is %s",
clone->id, child->id,
pcmk_readable_score(child->private->promotion_priority));
}
}
/*!
* \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(pcmk_resource_t *clone)
{
GList *colocations = NULL;
if (pe__set_clone_flag(clone, pcmk__clone_promotion_constrained)
== pcmk_rc_already) {
return;
}
pcmk__set_rsc_flags(clone, pcmk__rsc_updating_nodes);
for (GList *iter = clone->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) iter->data;
pcmk__rsc_trace(clone,
"Adding scores for %s: "
"initial promotion priority for %s is %s",
clone->id, child->id,
pcmk_readable_score(child->private->promotion_priority));
}
pe__show_node_scores(true, clone, "Before", clone->private->allowed_nodes,
clone->private->scheduler);
g_list_foreach(clone->private->children,
add_promotion_priority_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->private->allowed_nodes,
clone->private->scheduler);
// Reset promotion priorities to final node scores
g_list_foreach(clone->private->children,
set_promotion_priority_to_node_score, clone);
// Finally, sort instances in descending order of promotion priority
clone->private->children = g_list_sort(clone->private->children,
cmp_promotable_instance);
pcmk__clear_rsc_flags(clone, pcmk__rsc_updating_nodes);
}
/*!
* \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 pcmk_resource_t *
find_active_anon_instance(const pcmk_resource_t *clone, const char *id,
const pcmk_node_t *node)
{
for (GList *iter = clone->private->children; iter; iter = iter->next) {
pcmk_resource_t *child = iter->data;
pcmk_resource_t *active = NULL;
// Use ->find_rsc() in case this is a cloned group
active = clone->private->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 pcmk_resource_t *clone, const char *id,
const pcmk_node_t *node)
{
for (GList *iter = clone->private->children; iter; iter = iter->next) {
pcmk_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->private->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->private->probed_nodes,
node->private->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 pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk_node_t *allowed = g_hash_table_lookup(rsc->private->allowed_nodes,
node->private->id);
return (allowed != NULL) && (allowed->assign->score >= 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 pcmk_resource_t *rsc, const pcmk_node_t *node)
{
char *id = clone_strip(rsc->id);
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
pcmk_resource_t *active = NULL;
const char *reason = "allowed";
// Some checks apply only to anonymous clone instances
if (!pcmk_is_set(rsc->flags, pcmk__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->private->active_nodes == NULL)
&& (g_hash_table_size(rsc->private->probed_nodes) == 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->private->probed_nodes,
node->private->id) != NULL)
|| (pe_find_node_id(rsc->private->active_nodes,
node->private->id) != NULL)) {
reason = "known";
} else {
pcmk__rsc_trace(rsc,
"Ignoring %s promotion score (for %s) on %s: "
"not probed",
rsc->id, id, pcmk__node_name(node));
free(id);
return false;
}
check_allowed:
if (is_allowed(rsc, node)) {
pcmk__rsc_trace(rsc, "Counting %s promotion score (for %s) on %s: %s",
rsc->id, id, pcmk__node_name(node), reason);
free(id);
return true;
}
pcmk__rsc_trace(rsc,
"Ignoring %s promotion score (for %s) on %s: not allowed",
rsc->id, id, pcmk__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 pcmk_resource_t *rsc, const pcmk_node_t *node,
const char *name)
{
char *attr_name = NULL;
const char *attr_value = NULL;
const char *target = NULL;
enum pcmk__rsc_node node_type = pcmk__rsc_node_assigned;
if (pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) {
// Not assigned yet
node_type = pcmk__rsc_node_current;
}
target = g_hash_table_lookup(rsc->private->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
attr_name = pcmk_promotion_score_name(name);
attr_value = pcmk__node_attr(node, attr_name, target, node_type);
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 pcmk_resource_t *rsc, const pcmk_node_t *node,
bool *is_default)
{
const 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->private->children != NULL) {
int score = 0;
for (const GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
const pcmk_resource_t *child = (const pcmk_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 = pcmk__s(rsc->private->history_id, rsc->id);
attr_value = promotion_attr_value(rsc, node, name);
if (attr_value != NULL) {
pcmk__rsc_trace(rsc, "Promotion score for %s on %s = %s",
name, pcmk__node_name(node),
pcmk__s(attr_value, "(unset)"));
} else if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
/* If we don't have any resource history yet, we won't have history_id.
* In that case, for anonymous clones, try the resource name without
* any instance number.
*/
char *rsc_name = clone_strip(rsc->id);
if (strcmp(rsc->id, rsc_name) != 0) {
attr_value = promotion_attr_value(rsc, node, rsc_name);
pcmk__rsc_trace(rsc, "Promotion score for %s on %s (for %s) = %s",
rsc_name, pcmk__node_name(node), rsc->id,
pcmk__s(attr_value, "(unset)"));
}
free(rsc_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(pcmk_resource_t *rsc)
{
if (pe__set_clone_flag(rsc,
pcmk__clone_promotion_added) == pcmk_rc_already) {
return;
}
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data;
GHashTableIter iter;
pcmk_node_t *node = NULL;
int score, new_score;
g_hash_table_iter_init(&iter, child_rsc->private->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->assign->score, score);
if (new_score != node->assign->score) { // Could remain INFINITY
node->assign->score = new_score;
pcmk__rsc_trace(rsc,
"Added %s promotion priority (%s) to score "
"on %s (now %s)",
child_rsc->id, pcmk_readable_score(score),
pcmk__node_name(node),
pcmk_readable_score(new_score));
}
}
if (score > child_rsc->private->priority) {
pcmk__rsc_trace(rsc,
"Updating %s priority to promotion score "
"(%d->%d)",
child_rsc->id, child_rsc->private->priority,
score);
child_rsc->private->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)
{
pcmk_resource_t *rsc = (pcmk_resource_t *) data;
if (rsc->private->orig_role == pcmk_role_started) {
// Promotable clones should use unpromoted role instead of started
rsc->private->orig_role = pcmk_role_unpromoted;
}
g_list_foreach(rsc->private->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)
{
pcmk_resource_t *rsc = (pcmk_resource_t *) data;
GList *assigned = NULL;
rsc->private->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->private->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)
{
pcmk_resource_t *rsc = (pcmk_resource_t *) data;
if (rsc->private->next_role == pcmk_role_unknown) {
pe__set_next_role(rsc, pcmk_role_promoted, "promoted instance");
}
g_list_foreach(rsc->private->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(pcmk_resource_t *instance)
{
pcmk_node_t *chosen = instance->private->fns->location(instance, NULL,
FALSE);
const char *score_s = NULL;
score_s = pcmk_readable_score(instance->private->promotion_priority);
if (pcmk_is_set(instance->private->scheduler->flags,
pcmk_sched_output_scores)
&& !pcmk__is_daemon && (instance->private->scheduler->priv != NULL)) {
pcmk__output_t *out = instance->private->scheduler->priv;
out->message(out, "promotion-score", instance, chosen, score_s);
} else if (chosen == NULL) {
pcmk__rsc_debug(pe__const_top_resource(instance, false),
"%s promotion score (inactive): %s (priority=%d)",
instance->id, score_s, instance->private->priority);
} else {
pcmk__rsc_debug(pe__const_top_resource(instance, false),
"%s promotion score on %s: %s (priority=%d)",
instance->id, pcmk__node_name(chosen),
score_s, instance->private->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)
{
pcmk_resource_t *instance = (pcmk_resource_t *) data;
const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data;
const pcmk_node_t *chosen = NULL;
enum rsc_role_e next_role = pcmk_role_unknown;
GList *list = NULL;
pcmk__rsc_trace(clone, "Assigning priority for %s: %s", instance->id,
pcmk_role_text(instance->private->next_role));
if (instance->private->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->private->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->private->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->private->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
* PCMK_XE_RSC_LOCATION constraints, but prevents any
* instance from being promoted if neither a constraint nor
* a promotion score is present.
*/
instance->private->priority = -1;
}
}
break;
case pcmk_role_unpromoted:
case pcmk_role_stopped:
// Instance can't be promoted
instance->private->priority = -PCMK_SCORE_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->private->location_constraints,
chosen);
apply_promoted_locations(instance, clone->private->location_constraints,
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->private->cmds->apply_coloc_score(instance, cons->primary,
cons, true);
}
g_list_free(list);
instance->private->promotion_priority = instance->private->priority;
if (next_role == pcmk_role_promoted) {
instance->private->promotion_priority = PCMK_SCORE_INFINITY;
}
pcmk__rsc_trace(clone, "Assigning %s priority = %d",
instance->id, instance->private->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)
{
pcmk_resource_t *instance = (pcmk_resource_t *) data;
int *count = (int *) user_data;
const pcmk_resource_t *clone = pe__const_top_resource(instance, false);
const pcmk_scheduler_t *scheduler = instance->private->scheduler;
pcmk_node_t *chosen = NULL;
show_promotion_score(instance);
if (instance->private->promotion_priority < 0) {
pcmk__rsc_trace(clone, "Not supposed to promote instance %s",
instance->id);
} else if ((*count < pe__clone_promoted_max(instance))
|| !pcmk_is_set(clone->flags, pcmk__rsc_managed)) {
chosen = node_to_be_promoted_on(instance);
}
if (chosen == NULL) {
set_next_role_unpromoted(instance, NULL);
return;
}
if ((instance->private->orig_role < pcmk_role_promoted)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_quorate)
&& (scheduler->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->assign->count++;
pcmk__rsc_info(clone, "Choosing %s (%s) on %s for promotion",
instance->id, pcmk_role_text(instance->private->orig_role),
pcmk__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(pcmk_resource_t *rsc)
{
int promoted = 0;
GHashTableIter iter;
pcmk_node_t *node = NULL;
// Repurpose count to track the number of promoted instances assigned
g_hash_table_iter_init(&iter, rsc->private->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
node->assign->count = 0;
}
// Set instances' promotion priorities and sort by highest priority first
g_list_foreach(rsc->private->children, set_instance_priority, rsc);
sort_promotable_instances(rsc);
// Choose the first N eligible instances to be promoted
g_list_foreach(rsc->private->children, set_instance_role, &promoted);
pcmk__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(pcmk_resource_t *clone,
bool *any_promoting, bool *any_demoting)
{
for (GList *iter = clone->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *instance = (pcmk_resource_t *) iter->data;
instance->private->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(pcmk_resource_t *clone)
{
for (GList *iter = clone->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *instance = (pcmk_resource_t *) iter->data;
instance->private->priority = clone->private->priority;
}
}
/*!
* \internal
* \brief Create actions specific to promotable clones
*
* \param[in,out] clone Promotable clone to create actions for
*/
void
pcmk__create_promotable_actions(pcmk_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(pcmk_resource_t *clone)
{
pcmk_resource_t *previous = NULL; // Needed for ordered clones
pcmk__promotable_restart_ordering(clone);
for (GList *iter = clone->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *instance = (pcmk_resource_t *) iter->data;
// Demote before promote
pcmk__order_resource_actions(instance, PCMK_ACTION_DEMOTE,
instance, PCMK_ACTION_PROMOTE,
pcmk__ar_ordered);
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(pcmk_resource_t *dependent,
const pcmk_resource_t *primary,
const pcmk_node_t *primary_node,
const pcmk__colocation_t *colocation)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
const char *primary_value = NULL;
const char *attr = colocation->node_attribute;
if (colocation->score >= PCMK_SCORE_INFINITY) {
return; // Colocation is mandatory, so allowed node scores don't matter
}
primary_value = pcmk__colocation_node_attr(primary_node, attr, primary);
pcmk__rsc_trace(colocation->primary,
"Applying %s (%s with %s on %s by %s @%d) to %s",
colocation->id, colocation->dependent->id,
colocation->primary->id, pcmk__node_name(primary_node),
attr, colocation->score, dependent->id);
g_hash_table_iter_init(&iter, dependent->private->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->assign->score = pcmk__add_scores(node->assign->score,
colocation->score);
pcmk__rsc_trace(colocation->primary,
"Added %s score (%s) to %s (now %s)",
colocation->id,
pcmk_readable_score(colocation->score),
pcmk__node_name(node),
pcmk_readable_score(node->assign->score));
}
}
}
/*!
* \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 pcmk_resource_t *primary,
pcmk_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->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *instance = (pcmk_resource_t *) iter->data;
pcmk_node_t *node = instance->private->fns->location(instance, NULL,
FALSE);
if (node == NULL) {
continue;
}
if (instance->private->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 >= PCMK_SCORE_INFINITY)
&& ((colocation->dependent_role != pcmk_role_promoted)
|| (colocation->primary_role != pcmk_role_promoted))) {
pcmk__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 pcmk_resource_t *primary,
pcmk_resource_t *dependent,
const pcmk__colocation_t *colocation)
{
pcmk_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->private->priority,
colocation->score);
pcmk__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->private->priority),
pcmk_readable_score(colocation->score),
pcmk_readable_score(new_priority));
dependent->private->priority = new_priority;
} else if (colocation->score >= PCMK_SCORE_INFINITY) {
// Mandatory colocation, but primary won't be here
pcmk__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->private->priority = -PCMK_SCORE_INFINITY;
}
}
diff --git a/lib/pacemaker/pcmk_sched_resource.c b/lib/pacemaker/pcmk_sched_resource.c
index e555a30ef3..7a7051cc95 100644
--- a/lib/pacemaker/pcmk_sched_resource.c
+++ b/lib/pacemaker/pcmk_sched_resource.c
@@ -1,795 +1,796 @@
/*
* Copyright 2014-2024 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 <stdlib.h>
#include <string.h>
#include <crm/common/xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
// Resource assignment methods by resource variant
static pcmk__assignment_methods_t assignment_methods[] = {
{
pcmk__primitive_assign,
pcmk__primitive_create_actions,
pcmk__probe_rsc_on_node,
pcmk__primitive_internal_constraints,
pcmk__primitive_apply_coloc_score,
pcmk__colocated_resources,
pcmk__with_primitive_colocations,
pcmk__primitive_with_colocations,
pcmk__add_colocated_node_scores,
pcmk__apply_location,
pcmk__primitive_action_flags,
pcmk__update_ordered_actions,
pcmk__output_resource_actions,
pcmk__add_rsc_actions_to_graph,
pcmk__primitive_add_graph_meta,
pcmk__primitive_add_utilization,
pcmk__primitive_shutdown_lock,
},
{
pcmk__group_assign,
pcmk__group_create_actions,
pcmk__probe_rsc_on_node,
pcmk__group_internal_constraints,
pcmk__group_apply_coloc_score,
pcmk__group_colocated_resources,
pcmk__with_group_colocations,
pcmk__group_with_colocations,
pcmk__group_add_colocated_node_scores,
pcmk__group_apply_location,
pcmk__group_action_flags,
pcmk__group_update_ordered_actions,
pcmk__output_resource_actions,
pcmk__add_rsc_actions_to_graph,
pcmk__noop_add_graph_meta,
pcmk__group_add_utilization,
pcmk__group_shutdown_lock,
},
{
pcmk__clone_assign,
pcmk__clone_create_actions,
pcmk__clone_create_probe,
pcmk__clone_internal_constraints,
pcmk__clone_apply_coloc_score,
pcmk__colocated_resources,
pcmk__with_clone_colocations,
pcmk__clone_with_colocations,
pcmk__add_colocated_node_scores,
pcmk__clone_apply_location,
pcmk__clone_action_flags,
pcmk__instance_update_ordered_actions,
pcmk__output_resource_actions,
pcmk__clone_add_actions_to_graph,
pcmk__clone_add_graph_meta,
pcmk__clone_add_utilization,
pcmk__clone_shutdown_lock,
},
{
pcmk__bundle_assign,
pcmk__bundle_create_actions,
pcmk__bundle_create_probe,
pcmk__bundle_internal_constraints,
pcmk__bundle_apply_coloc_score,
pcmk__colocated_resources,
pcmk__with_bundle_colocations,
pcmk__bundle_with_colocations,
pcmk__add_colocated_node_scores,
pcmk__bundle_apply_location,
pcmk__bundle_action_flags,
pcmk__instance_update_ordered_actions,
pcmk__output_bundle_actions,
pcmk__bundle_add_actions_to_graph,
pcmk__noop_add_graph_meta,
pcmk__bundle_add_utilization,
pcmk__bundle_shutdown_lock,
}
};
/*!
* \internal
* \brief Check whether a resource's agent standard, provider, or type changed
*
* \param[in,out] rsc Resource to check
* \param[in,out] node Node needing unfencing if agent changed
* \param[in] rsc_entry XML with previously known agent information
* \param[in] active_on_node Whether \p rsc is active on \p node
*
* \return true if agent for \p rsc changed, otherwise false
*/
bool
pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node,
const xmlNode *rsc_entry, bool active_on_node)
{
bool changed = false;
const char *attr_list[] = {
PCMK_XA_TYPE,
PCMK_XA_CLASS,
PCMK_XA_PROVIDER,
};
for (int i = 0; i < PCMK__NELEM(attr_list); i++) {
const char *value = crm_element_value(rsc->private->xml, attr_list[i]);
const char *old_value = crm_element_value(rsc_entry, attr_list[i]);
if (!pcmk__str_eq(value, old_value, pcmk__str_none)) {
changed = true;
trigger_unfencing(rsc, node, "Device definition changed", NULL,
rsc->private->scheduler);
if (active_on_node) {
crm_notice("Forcing restart of %s on %s "
"because %s changed from '%s' to '%s'",
rsc->id, pcmk__node_name(node), attr_list[i],
pcmk__s(old_value, ""), pcmk__s(value, ""));
}
}
}
if (changed && active_on_node) {
// Make sure the resource is restarted
custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE,
rsc->private->scheduler);
pcmk__set_rsc_flags(rsc, pcmk__rsc_start_pending);
}
return changed;
}
/*!
* \internal
* \brief Add resource (and any matching children) to list if it matches ID
*
* \param[in] result List to add resource to
* \param[in] rsc Resource to check
* \param[in] id ID to match
*
* \return (Possibly new) head of list
*/
static GList *
add_rsc_if_matching(GList *result, pcmk_resource_t *rsc, const char *id)
{
if (pcmk__str_eq(id, rsc->id, pcmk__str_none)
|| pcmk__str_eq(id, rsc->private->history_id, pcmk__str_none)) {
result = g_list_prepend(result, rsc);
}
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) iter->data;
result = add_rsc_if_matching(result, child, id);
}
return result;
}
/*!
* \internal
* \brief Find all resources matching a given ID by either ID or clone name
*
* \param[in] id Resource ID to check
* \param[in] scheduler Scheduler data
*
* \return List of all resources that match \p id
* \note The caller is responsible for freeing the return value with
* g_list_free().
*/
GList *
pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler)
{
GList *result = NULL;
CRM_CHECK((id != NULL) && (scheduler != NULL), return NULL);
for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) {
result = add_rsc_if_matching(result, (pcmk_resource_t *) iter->data,
id);
}
return result;
}
/*!
* \internal
* \brief Set the variant-appropriate assignment methods for a resource
*
* \param[in,out] data Resource to set assignment methods for
* \param[in] user_data Ignored
*/
static void
set_assignment_methods_for_rsc(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
rsc->private->cmds = &assignment_methods[rsc->private->variant];
g_list_foreach(rsc->private->children, set_assignment_methods_for_rsc,
NULL);
}
/*!
* \internal
* \brief Set the variant-appropriate assignment methods for all resources
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler)
{
g_list_foreach(scheduler->resources, set_assignment_methods_for_rsc, NULL);
}
/*!
* \internal
* \brief Wrapper for colocated_resources() method for readability
*
* \param[in] rsc Resource to add to colocated list
* \param[in] orig_rsc Resource originally requested
* \param[in,out] list Pointer to list to add to
*
* \return (Possibly new) head of list
*/
static inline void
add_colocated_resources(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
*list = rsc->private->cmds->colocated_resources(rsc, orig_rsc, *list);
}
// Shared implementation of pcmk__assignment_methods_t:colocated_resources()
GList *
pcmk__colocated_resources(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc,
GList *colocated_rscs)
{
const GList *iter = NULL;
GList *colocations = NULL;
if (orig_rsc == NULL) {
orig_rsc = rsc;
}
if ((rsc == NULL) || (g_list_find(colocated_rscs, rsc) != NULL)) {
return colocated_rscs;
}
pcmk__rsc_trace(orig_rsc, "%s is in colocation chain with %s",
rsc->id, orig_rsc->id);
colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc);
// Follow colocations where this resource is the dependent resource
colocations = pcmk__this_with_colocations(rsc);
for (iter = colocations; iter != NULL; iter = iter->next) {
const pcmk__colocation_t *constraint = iter->data;
const pcmk_resource_t *primary = constraint->primary;
if (primary == orig_rsc) {
continue; // Break colocation loop
}
if ((constraint->score == PCMK_SCORE_INFINITY) &&
(pcmk__colocation_affects(rsc, primary, constraint,
true) == pcmk__coloc_affects_location)) {
add_colocated_resources(primary, orig_rsc, &colocated_rscs);
}
}
g_list_free(colocations);
// Follow colocations where this resource is the primary resource
colocations = pcmk__with_this_colocations(rsc);
for (iter = colocations; iter != NULL; iter = iter->next) {
const pcmk__colocation_t *constraint = iter->data;
const pcmk_resource_t *dependent = constraint->dependent;
if (dependent == orig_rsc) {
continue; // Break colocation loop
}
if (pcmk__is_clone(rsc) && !pcmk__is_clone(dependent)) {
continue; // We can't be sure whether dependent will be colocated
}
if ((constraint->score == PCMK_SCORE_INFINITY) &&
(pcmk__colocation_affects(dependent, rsc, constraint,
true) == pcmk__coloc_affects_location)) {
add_colocated_resources(dependent, orig_rsc, &colocated_rscs);
}
}
g_list_free(colocations);
return colocated_rscs;
}
// No-op function for variants that don't need to implement add_graph_meta()
void
pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml)
{
}
/*!
* \internal
* \brief Output a summary of scheduled actions for a resource
*
* \param[in,out] rsc Resource to output actions for
*/
void
pcmk__output_resource_actions(pcmk_resource_t *rsc)
{
pcmk_node_t *next = NULL;
pcmk_node_t *current = NULL;
pcmk__output_t *out = NULL;
CRM_ASSERT(rsc != NULL);
out = rsc->private->scheduler->priv;
if (rsc->private->children != NULL) {
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) iter->data;
child->private->cmds->output_actions(child);
}
return;
}
next = rsc->private->assigned_node;
if (rsc->private->active_nodes != NULL) {
current = pcmk__current_node(rsc);
if (rsc->private->orig_role == pcmk_role_stopped) {
/* This can occur when resources are being recovered because
* the current role can change in pcmk__primitive_create_actions()
*/
rsc->private->orig_role = pcmk_role_started;
}
}
if ((current == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
/* Don't log stopped orphans */
return;
}
out->message(out, "rsc-action", rsc, current, next);
}
/*!
* \internal
* \brief Add a resource to a node's list of assigned resources
*
* \param[in,out] node Node to add resource to
* \param[in] rsc Resource to add
*/
static inline void
add_assigned_resource(pcmk_node_t *node, pcmk_resource_t *rsc)
{
node->private->assigned_resources =
g_list_prepend(node->private->assigned_resources, rsc);
}
/*!
* \internal
* \brief Assign a specified resource (of any variant) to a node
*
* Assign a specified resource and its children (if any) to a specified node, if
* the node can run the resource (or unconditionally, if \p force is true). Mark
* the resources as no longer provisional.
*
* If a resource can't be assigned (or \p node is \c NULL), unassign any
* previous assignment. If \p stop_if_fail is \c true, set next role to stopped
* and update any existing actions scheduled for the resource.
*
* \param[in,out] rsc Resource to assign
* \param[in,out] node Node to assign \p rsc to
* \param[in] force If true, assign to \p node even if unavailable
* \param[in] stop_if_fail If \c true and either \p rsc can't be assigned
* or \p chosen is \c NULL, set next role to
* stopped and update existing actions (if \p rsc
* is not a primitive, this applies to its
* primitive descendants instead)
*
* \return \c true if the assignment of \p rsc changed, or \c false otherwise
*
* \note Assigning a resource to the NULL node using this function is different
* from calling pcmk__unassign_resource(), in that it may also update any
* actions created for the resource.
* \note The \c pcmk__assignment_methods_t:assign() method is preferred, unless
* a resource should be assigned to the \c NULL node or every resource in
* a tree should be assigned to the same 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.
*/
bool
pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force,
bool stop_if_fail)
{
bool changed = false;
pcmk_scheduler_t *scheduler = NULL;
CRM_ASSERT(rsc != NULL);
scheduler = rsc->private->scheduler;
if (rsc->private->children != NULL) {
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk_resource_t *child_rsc = iter->data;
changed |= pcmk__assign_resource(child_rsc, node, force,
stop_if_fail);
}
return changed;
}
// Assigning a primitive
if (!force && (node != NULL)
&& ((node->assign->score < 0)
// Allow graph to assume that guest node connections will come up
|| (!pcmk__node_available(node, true, false)
&& !pcmk__is_guest_or_bundle_node(node)))) {
pcmk__rsc_debug(rsc,
"All nodes for resource %s are unavailable, unclean or "
"shutting down (%s can%s run resources, with score %s)",
rsc->id, pcmk__node_name(node),
(pcmk__node_available(node, true, false)? "" : "not"),
pcmk_readable_score(node->assign->score));
if (stop_if_fail) {
pe__set_next_role(rsc, pcmk_role_stopped, "node availability");
}
node = NULL;
}
if (rsc->private->assigned_node != NULL) {
changed = !pcmk__same_node(rsc->private->assigned_node, node);
} else {
changed = (node != NULL);
}
pcmk__unassign_resource(rsc);
pcmk__clear_rsc_flags(rsc, pcmk__rsc_unassigned);
if (node == NULL) {
char *rc_stopped = NULL;
pcmk__rsc_debug(rsc, "Could not assign %s to a node", rsc->id);
if (!stop_if_fail) {
return changed;
}
pe__set_next_role(rsc, pcmk_role_stopped, "unable to assign");
for (GList *iter = rsc->private->actions;
iter != NULL; iter = iter->next) {
pcmk_action_t *op = (pcmk_action_t *) iter->data;
pcmk__rsc_debug(rsc, "Updating %s for %s assignment failure",
op->uuid, rsc->id);
if (pcmk__str_eq(op->task, PCMK_ACTION_STOP, pcmk__str_none)) {
pcmk__clear_action_flags(op, pcmk__action_optional);
} else if (pcmk__str_eq(op->task, PCMK_ACTION_START,
pcmk__str_none)) {
pcmk__clear_action_flags(op, pcmk__action_runnable);
} else {
// Cancel recurring actions, unless for stopped state
const char *interval_ms_s = NULL;
const char *target_rc_s = NULL;
interval_ms_s = g_hash_table_lookup(op->meta,
PCMK_META_INTERVAL);
target_rc_s = g_hash_table_lookup(op->meta,
PCMK__META_OP_TARGET_RC);
if (rc_stopped == NULL) {
rc_stopped = pcmk__itoa(PCMK_OCF_NOT_RUNNING);
}
if (!pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)
&& !pcmk__str_eq(rc_stopped, target_rc_s, pcmk__str_none)) {
pcmk__clear_action_flags(op, pcmk__action_runnable);
}
}
}
free(rc_stopped);
return changed;
}
pcmk__rsc_debug(rsc, "Assigning %s to %s", rsc->id, pcmk__node_name(node));
rsc->private->assigned_node = pe__copy_node(node);
add_assigned_resource(node, rsc);
node->private->num_resources++;
node->assign->count++;
pcmk__consume_node_capacity(node->private->utilization, rsc);
if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) {
pcmk__output_t *out = scheduler->priv;
out->message(out, "resource-util", rsc, node, __func__);
}
return changed;
}
/*!
* \internal
* \brief Remove any node assignment from a specified resource and its children
*
* If a specified resource has been assigned to a node, remove that assignment
* and mark the resource as provisional again.
*
* \param[in,out] rsc Resource to unassign
*
* \note This function is called recursively on \p rsc and its children.
*/
void
pcmk__unassign_resource(pcmk_resource_t *rsc)
{
pcmk_node_t *old = rsc->private->assigned_node;
if (old == NULL) {
crm_info("Unassigning %s", rsc->id);
} else {
crm_info("Unassigning %s from %s", rsc->id, pcmk__node_name(old));
}
pcmk__set_rsc_flags(rsc, pcmk__rsc_unassigned);
if (rsc->private->children == NULL) {
if (old == NULL) {
return;
}
rsc->private->assigned_node = NULL;
/* We're going to free the pcmk_node_t, but its details member is shared
* and will remain, so update that appropriately first.
*/
old->private->assigned_resources =
g_list_remove(old->private->assigned_resources, rsc);
old->private->num_resources--;
pcmk__release_node_capacity(old->private->utilization, rsc);
free(old);
return;
}
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
pcmk__unassign_resource((pcmk_resource_t *) iter->data);
}
}
/*!
* \internal
* \brief Check whether a resource has reached its migration threshold on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node Node to check
* \param[out] failed If threshold has been reached, this will be set to
* resource that failed (possibly a parent of \p rsc)
*
* \return true if the migration threshold has been reached, false otherwise
*/
bool
pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node,
pcmk_resource_t **failed)
{
int fail_count, remaining_tries;
pcmk_resource_t *rsc_to_ban = rsc;
// Migration threshold of 0 means never force away
if (rsc->private->ban_after_failures == 0) {
return false;
}
// If we're ignoring failures, also ignore the migration threshold
if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) {
return false;
}
// If there are no failures, there's no need to force away
fail_count = pe_get_failcount(node, rsc, NULL,
pcmk__fc_effective|pcmk__fc_launched, NULL);
if (fail_count <= 0) {
return false;
}
// If failed resource is anonymous clone instance, we'll force clone away
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
rsc_to_ban = uber_parent(rsc);
}
// How many more times recovery will be tried on this node
remaining_tries = rsc->private->ban_after_failures - fail_count;
if (remaining_tries <= 0) {
- pcmk__sched_warn("%s cannot run on %s due to reaching migration "
+ pcmk__sched_warn(rsc->private->scheduler,
+ "%s cannot run on %s due to reaching migration "
"threshold (clean up resource to allow again)"
QB_XS " failures=%d "
PCMK_META_MIGRATION_THRESHOLD "=%d",
rsc_to_ban->id, pcmk__node_name(node), fail_count,
rsc->private->ban_after_failures);
if (failed != NULL) {
*failed = rsc_to_ban;
}
return true;
}
crm_info("%s can fail %d more time%s on "
"%s before reaching migration threshold (%d)",
rsc_to_ban->id, remaining_tries, pcmk__plural_s(remaining_tries),
pcmk__node_name(node), rsc->private->ban_after_failures);
return false;
}
/*!
* \internal
* \brief Get a node's score
*
* \param[in] node Node with ID to check
* \param[in] nodes List of nodes to look for \p node score in
*
* \return Node's score, or -INFINITY if not found
*/
static int
get_node_score(const pcmk_node_t *node, GHashTable *nodes)
{
pcmk_node_t *found_node = NULL;
if ((node != NULL) && (nodes != NULL)) {
found_node = g_hash_table_lookup(nodes, node->private->id);
}
if (found_node == NULL) {
return -PCMK_SCORE_INFINITY;
}
return found_node->assign->score;
}
/*!
* \internal
* \brief Compare two resources according to which should be assigned first
*
* \param[in] a First resource to compare
* \param[in] b Second resource to compare
* \param[in] data Sorted list of all nodes in cluster
*
* \return -1 if \p a should be assigned before \b, 0 if they are equal,
* or +1 if \p a should be assigned after \b
*/
static gint
cmp_resources(gconstpointer a, gconstpointer b, gpointer data)
{
/* GLib insists that this function require gconstpointer arguments, but we
* make a small, temporary change to each argument (setting the
* pe_rsc_merging flag) during comparison
*/
pcmk_resource_t *resource1 = (pcmk_resource_t *) a;
pcmk_resource_t *resource2 = (pcmk_resource_t *) b;
const GList *nodes = data;
int rc = 0;
int r1_score = -PCMK_SCORE_INFINITY;
int r2_score = -PCMK_SCORE_INFINITY;
pcmk_node_t *r1_node = NULL;
pcmk_node_t *r2_node = NULL;
GHashTable *r1_nodes = NULL;
GHashTable *r2_nodes = NULL;
const char *reason = NULL;
// Resources with highest priority should be assigned first
reason = "priority";
r1_score = resource1->private->priority;
r2_score = resource2->private->priority;
if (r1_score > r2_score) {
rc = -1;
goto done;
}
if (r1_score < r2_score) {
rc = 1;
goto done;
}
// We need nodes to make any other useful comparisons
reason = "no node list";
if (nodes == NULL) {
goto done;
}
// Calculate and log node scores
resource1->private->cmds->add_colocated_node_scores(resource1, NULL,
resource1->id,
&r1_nodes, NULL, 1,
pcmk__coloc_select_this_with);
resource2->private->cmds->add_colocated_node_scores(resource2, NULL,
resource2->id,
&r2_nodes, NULL, 1,
pcmk__coloc_select_this_with);
pe__show_node_scores(true, NULL, resource1->id, r1_nodes,
resource1->private->scheduler);
pe__show_node_scores(true, NULL, resource2->id, r2_nodes,
resource2->private->scheduler);
// The resource with highest score on its current node goes first
reason = "current location";
if (resource1->private->active_nodes != NULL) {
r1_node = pcmk__current_node(resource1);
}
if (resource2->private->active_nodes != NULL) {
r2_node = pcmk__current_node(resource2);
}
r1_score = get_node_score(r1_node, r1_nodes);
r2_score = get_node_score(r2_node, r2_nodes);
if (r1_score > r2_score) {
rc = -1;
goto done;
}
if (r1_score < r2_score) {
rc = 1;
goto done;
}
// Otherwise a higher score on any node will do
reason = "score";
for (const GList *iter = nodes; iter != NULL; iter = iter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) iter->data;
r1_score = get_node_score(node, r1_nodes);
r2_score = get_node_score(node, r2_nodes);
if (r1_score > r2_score) {
rc = -1;
goto done;
}
if (r1_score < r2_score) {
rc = 1;
goto done;
}
}
done:
crm_trace("%s (%d)%s%s %c %s (%d)%s%s: %s",
resource1->id, r1_score,
((r1_node == NULL)? "" : " on "),
((r1_node == NULL)? "" : r1_node->private->id),
((rc < 0)? '>' : ((rc > 0)? '<' : '=')),
resource2->id, r2_score,
((r2_node == NULL)? "" : " on "),
((r2_node == NULL)? "" : r2_node->private->id),
reason);
if (r1_nodes != NULL) {
g_hash_table_destroy(r1_nodes);
}
if (r2_nodes != NULL) {
g_hash_table_destroy(r2_nodes);
}
return rc;
}
/*!
* \internal
* \brief Sort resources in the order they should be assigned to nodes
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__sort_resources(pcmk_scheduler_t *scheduler)
{
GList *nodes = g_list_copy(scheduler->nodes);
nodes = pcmk__sort_nodes(nodes, NULL);
scheduler->resources = g_list_sort_with_data(scheduler->resources,
cmp_resources, nodes);
g_list_free(nodes);
}
diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c
index 74d91ebbc4..39c95ecfc9 100644
--- a/lib/pacemaker/pcmk_scheduler.c
+++ b/lib/pacemaker/pcmk_scheduler.c
@@ -1,907 +1,908 @@
/*
* 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 <crm/crm.h>
#include <crm/cib.h>
#include <crm/cib/internal.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(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_op,
enum pcmk__check_parameters check)
{
const char *reason = NULL;
pcmk__op_digest_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, pcmk__fc_effective,
NULL)) {
reason = "action definition changed";
}
break;
case pcmk__check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node,
rsc->private->scheduler);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s has "
"no digest to compare",
rsc->id, pcmk__xe_id(rsc_op), node->private->id);
break;
case pcmk__digest_match:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason != NULL) {
pe__clear_failcount(rsc, node, reason, rsc->private->scheduler);
}
}
/*!
* \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 pcmk_node_t *node,
const pcmk_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)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
// If this is a collective resource, apply recursively to children instead
if (rsc->private->children != NULL) {
g_list_foreach(rsc->private->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.
*/
pcmk_resource_t *failed = NULL;
if (pcmk__threshold_reached(rsc, node, &failed)) {
resource_location(failed, node, -PCMK_SCORE_INFINITY,
"__fail_limit__", rsc->private->scheduler);
}
}
}
/*!
* \internal
* \brief If resource has exclusive discovery, ban node if not allowed
*
* Location constraints have a PCMK_XA_RESOURCE_DISCOVERY option that allows
* users to specify where probes are done for the affected resource. If this is
* set to \c 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)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
if (pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes)
|| pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_exclusive_probes)) {
pcmk_node_t *match = NULL;
// If this is a collective resource, apply recursively to children
g_list_foreach(rsc->private->children, apply_exclusive_discovery,
user_data);
match = g_hash_table_lookup(rsc->private->allowed_nodes,
node->private->id);
if ((match != NULL)
&& (match->assign->probe_mode != pcmk__probe_exclusive)) {
match->assign->score = -PCMK_SCORE_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)
{
pcmk_resource_t *rsc = data;
pcmk_node_t *node = NULL;
// If this is a collective resource, apply recursively to children instead
if (rsc->private->children != NULL) {
g_list_foreach(rsc->private->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, pcmk__rsc_managed)
|| (rsc->private->stickiness < 1)
|| !pcmk__list_of_1(rsc->private->active_nodes)) {
return;
}
node = rsc->private->active_nodes->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->private->scheduler->flags,
pcmk_sched_symmetric_cluster)
&& (g_hash_table_lookup(rsc->private->allowed_nodes,
node->private->id) == NULL)) {
pcmk__rsc_debug(rsc,
"Ignoring %s stickiness because the cluster is "
"asymmetric and %s is not explicitly allowed",
rsc->id, pcmk__node_name(node));
return;
}
pcmk__rsc_debug(rsc, "Resource %s has %d stickiness on %s",
rsc->id, rsc->private->stickiness, pcmk__node_name(node));
resource_location(rsc, node, rsc->private->stickiness, "stickiness",
rsc->private->scheduler);
}
/*!
* \internal
* \brief Apply shutdown locks for all resources as appropriate
*
* \param[in,out] scheduler Scheduler data
*/
static void
apply_shutdown_locks(pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
return;
}
for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->private->cmds->shutdown_lock(rsc);
}
}
/*!
* \internal
* \brief Calculate the number of available nodes in the cluster
*
* \param[in,out] scheduler Scheduler data
*/
static void
count_available_nodes(pcmk_scheduler_t *scheduler)
{
if (pcmk_is_set(scheduler->flags, pcmk_sched_no_compat)) {
return;
}
// @COMPAT for API backward compatibility only (cluster does not use value)
for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
if ((node != NULL) && (node->assign->score >= 0)
&& node->details->online
&& (node->private->variant != pcmk__node_variant_ping)) {
scheduler->max_valid_nodes++;
}
}
crm_trace("Online node count: %d", scheduler->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(pcmk_scheduler_t *scheduler)
{
crm_trace("Applying node-specific scheduling criteria");
apply_shutdown_locks(scheduler);
count_available_nodes(scheduler);
pcmk__apply_locations(scheduler);
g_list_foreach(scheduler->resources, apply_stickiness, NULL);
for (GList *node_iter = scheduler->nodes; node_iter != NULL;
node_iter = node_iter->next) {
for (GList *rsc_iter = scheduler->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] scheduler Scheduler data
*/
static void
assign_resources(pcmk_scheduler_t *scheduler)
{
GList *iter = NULL;
crm_trace("Assigning resources to nodes");
if (!pcmk__str_eq(scheduler->placement_strategy, PCMK_VALUE_DEFAULT,
pcmk__str_casei)) {
pcmk__sort_resources(scheduler);
}
pcmk__show_node_capacities("Original", scheduler);
if (pcmk_is_set(scheduler->flags, pcmk_sched_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 = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
const pcmk_node_t *target = rsc->private->partial_migration_target;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
pcmk__rsc_trace(rsc, "Assigning remote connection resource '%s'",
rsc->id);
rsc->private->cmds->assign(rsc, target, true);
}
}
}
/* now do the rest of the resources */
for (iter = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
pcmk__rsc_trace(rsc, "Assigning %s resource '%s'",
rsc->private->xml->name, rsc->id);
rsc->private->cmds->assign(rsc, NULL, true);
}
}
pcmk__show_node_capacities("Remaining", scheduler);
}
/*!
* \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)
{
pcmk_resource_t *rsc = data;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
return;
}
crm_trace("Clear fail counts for orphaned resource %s", rsc->id);
/* There's no need to recurse into rsc->private->children because those
* should just be unassigned clone instances.
*/
for (GList *iter = rsc->private->scheduler->nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
pcmk_action_t *clear_op = NULL;
if (!node->details->online) {
continue;
}
if (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) == 0) {
continue;
}
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
rsc->private->scheduler);
/* We can't use order_action_then_stop() here because its
* pcmk__ar_guest_allowed breaks things
*/
pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc),
NULL, pcmk__ar_ordered, rsc->private->scheduler);
}
}
/*!
* \internal
* \brief Schedule any resource actions needed
*
* \param[in,out] scheduler Scheduler data
*/
static void
schedule_resource_actions(pcmk_scheduler_t *scheduler)
{
// Process deferred action checks
pe__foreach_param_check(scheduler, check_params);
pe__free_param_checks(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) {
crm_trace("Scheduling probes");
pcmk__schedule_probes(scheduler);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
g_list_foreach(scheduler->resources, clear_failcounts_if_orphaned,
NULL);
}
crm_trace("Scheduling resource actions");
for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->private->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 pcmk_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
return true;
}
for (GList *iter = rsc->private->children;
iter != NULL; iter = iter->next) {
if (is_managed((pcmk_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether any resources in the cluster are managed
*
* \param[in] scheduler Scheduler data
*
* \return true if any resource is managed, otherwise false
*/
static bool
any_managed_resources(const pcmk_scheduler_t *scheduler)
{
for (const GList *iter = scheduler->resources;
iter != NULL; iter = iter->next) {
if (is_managed((const pcmk_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 pcmk_node_t *node, bool have_managed)
{
return have_managed && node->details->unclean
&& pe_can_fence(node->private->scheduler, 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 pcmk_node_t *node)
{
if (pcmk__is_pacemaker_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] scheduler Scheduler data
*
* \return (Possibly new) head of \p list
*/
static GList *
add_nondc_fencing(GList *list, pcmk_action_t *action,
const pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk_sched_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((pcmk_action_t *) list->data, action, pcmk__ar_ordered);
}
return g_list_prepend(list, action);
}
/*!
* \internal
* \brief Schedule a node for fencing
*
* \param[in,out] node Node that requires fencing
*/
static pcmk_action_t *
schedule_fencing(pcmk_node_t *node)
{
pcmk_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean",
FALSE, node->private->scheduler);
- pcmk__sched_warn("Scheduling node %s for fencing", pcmk__node_name(node));
+ pcmk__sched_warn(node->private->scheduler, "Scheduling node %s for fencing",
+ pcmk__node_name(node));
pcmk__order_vs_fence(fencing, node->private->scheduler);
return fencing;
}
/*!
* \internal
* \brief Create and order node fencing and shutdown actions
*
* \param[in,out] scheduler Scheduler data
*/
static void
schedule_fencing_and_shutdowns(pcmk_scheduler_t *scheduler)
{
pcmk_action_t *dc_down = NULL;
bool integrity_lost = false;
bool have_managed = any_managed_resources(scheduler);
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 = scheduler->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
pcmk_action_t *fencing = NULL;
const bool is_dc = pcmk__same_node(node, scheduler->dc_node);
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pcmk__is_guest_or_bundle_node(node)) {
if (pcmk_is_set(node->private->flags, pcmk__node_remote_reset)
&& have_managed && pe_can_fence(scheduler, 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 (is_dc) {
dc_down = fencing;
} else {
fencing_ops = add_nondc_fencing(fencing_ops, fencing,
scheduler);
}
} else if (needs_shutdown(node)) {
pcmk_action_t *down_op = pcmk__new_shutdown_action(node);
// Track DC and non-DC shutdown actions separately
if (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;
pcmk__config_warn("Node %s is unclean but cannot be fenced",
pcmk__node_name(node));
}
}
if (integrity_lost) {
if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
pcmk__config_warn("Resource functionality and data integrity "
"cannot be guaranteed (configure, enable, "
"and test fencing to correct this)");
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
crm_notice("Unclean nodes will not be fenced until quorum is "
"attained or " PCMK_OPT_NO_QUORUM_POLICY " is set to "
PCMK_VALUE_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(scheduler->flags, pcmk_sched_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((pcmk_action_t *) fencing_ops->data, dc_down,
pcmk__ar_ordered);
}
}
g_list_free(fencing_ops);
g_list_free(shutdown_ops);
}
static void
log_resource_details(pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->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 = scheduler->resources; item != NULL; item = item->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) item->data;
// Log all resources except inactive orphans
if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)
|| (rsc->private->orig_role != pcmk_role_stopped)) {
out->message(out, pcmk__map_element_name(rsc->private->xml), 0UL,
rsc, all, all);
}
}
g_list_free(all);
}
static void
log_all_actions(pcmk_scheduler_t *scheduler)
{
/* This only ever outputs to the log, so ignore whatever output object was
* previously set and just log instead.
*/
pcmk__output_t *prev_out = scheduler->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);
scheduler->priv = out;
out->begin_list(out, NULL, NULL, "Actions");
pcmk__output_actions(scheduler);
out->end_list(out);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
scheduler->priv = prev_out;
}
/*!
* \internal
* \brief Log all required but unrunnable actions at trace level
*
* \param[in] scheduler Scheduler data
*/
static void
log_unrunnable_actions(const pcmk_scheduler_t *scheduler)
{
const uint64_t flags = pcmk__action_optional
|pcmk__action_runnable
|pcmk__action_pseudo;
crm_trace("Required but unrunnable actions:");
for (const GList *iter = scheduler->actions;
iter != NULL; iter = iter->next) {
const pcmk_action_t *action = (const pcmk_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 Scheduler flags to set in addition to defaults
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_cib(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler)
{
const char* localhost_save = NULL;
if (pcmk_is_set(scheduler->flags, pcmk_sched_have_status)) {
crm_trace("Reusing previously calculated cluster status");
pcmk__set_scheduler_flags(scheduler, flags);
return;
}
if (scheduler->localhost) {
localhost_save = scheduler->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 pcmk_sched_have_status is set (i.e. cluster_status() was
* previously called, whether directly or via pcmk__schedule_actions()).
*/
set_working_set_defaults(scheduler);
if (localhost_save) {
scheduler->localhost = localhost_save;
}
pcmk__set_scheduler_flags(scheduler, flags);
scheduler->input = cib;
cluster_status(scheduler); // Sets pcmk_sched_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 Scheduler flags to set in addition to defaults
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__schedule_actions(xmlNode *cib, unsigned long long flags,
pcmk_scheduler_t *scheduler)
{
unpack_cib(cib, flags, scheduler);
pcmk__set_assignment_methods(scheduler);
pcmk__apply_node_health(scheduler);
pcmk__unpack_constraints(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk_sched_validate_only)) {
return;
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only)
&& pcmk__is_daemon) {
log_resource_details(scheduler);
}
apply_node_criteria(scheduler);
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
return;
}
pcmk__create_internal_constraints(scheduler);
pcmk__handle_rsc_config_changes(scheduler);
assign_resources(scheduler);
schedule_resource_actions(scheduler);
/* 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(scheduler);
schedule_fencing_and_shutdowns(scheduler);
pcmk__apply_orderings(scheduler);
log_all_actions(scheduler);
pcmk__create_graph(scheduler);
if (get_crm_log_level() == LOG_TRACE) {
log_unrunnable_actions(scheduler);
}
}
/*!
* \internal
* \brief Initialize scheduler data
*
* Make our own copies of the CIB XML and date/time object, if they're not
* \c NULL. This way we don't have to take ownership of the objects passed via
* the API.
*
* This function is most useful for public API functions that want the caller
* to retain ownership of the CIB object
*
* \param[in,out] out Output object
* \param[in] input The CIB XML to check (if \c NULL, use current CIB)
* \param[in] date Date and time to use in the scheduler (if \c NULL,
* use current date and time). This can be used for
* checking whether a rule is in effect at a certa
* date and time.
* \param[out] scheduler Where to store initialized scheduler data
*
* \return Standard Pacemaker return code
*/
int
pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date,
pcmk_scheduler_t **scheduler)
{
// Allows for cleaner syntax than dereferencing the scheduler argument
pcmk_scheduler_t *new_scheduler = NULL;
new_scheduler = pe_new_working_set();
if (new_scheduler == NULL) {
return ENOMEM;
}
pcmk__set_scheduler_flags(new_scheduler,
pcmk_sched_no_counts|pcmk_sched_no_compat);
// Populate the scheduler data
// Make our own copy of the given input or fetch the CIB and use that
if (input != NULL) {
new_scheduler->input = pcmk__xml_copy(NULL, input);
if (new_scheduler->input == NULL) {
out->err(out, "Failed to copy input XML");
pe_free_working_set(new_scheduler);
return ENOMEM;
}
} else {
int rc = cib__signon_query(out, NULL, &(new_scheduler->input));
if (rc != pcmk_rc_ok) {
pe_free_working_set(new_scheduler);
return rc;
}
}
// Make our own copy of the given crm_time_t object; otherwise
// cluster_status() populates with the current time
if (date != NULL) {
// pcmk_copy_time() guarantees non-NULL
new_scheduler->now = pcmk_copy_time(date);
}
// Unpack everything
cluster_status(new_scheduler);
*scheduler = new_scheduler;
return pcmk_rc_ok;
}
diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c
index e21d184fd6..a1f162ec95 100644
--- a/lib/pengine/complex.c
+++ b/lib/pengine/complex.c
@@ -1,1269 +1,1271 @@
/*
* 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/common/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);
static pcmk__rsc_methods_t resource_class_functions[] = {
{
native_unpack,
native_find_rsc,
native_parameter,
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_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_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__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 pcmk__rsc_variant
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;
}
/*!
* \internal
* \brief Insert a meta-attribute if not already present
*
* \param[in] key Meta-attribute name
* \param[in] value Meta-attribute value to add if not already present
* \param[in,out] table Meta-attribute hash table to insert into
*
* \note This is like pcmk__insert_meta() except it won't overwrite existing
* values.
*/
static void
dup_attr(gpointer key, gpointer value, gpointer user_data)
{
GHashTable *table = user_data;
CRM_CHECK((key != NULL) && (table != NULL), return);
if (pcmk__str_eq((const char *) value, "#default", pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting meta-attributes (such as %s) to "
"the explicit value '#default' is deprecated and "
"will be removed in a future release",
(const char *) key);
} else if ((value != NULL) && (g_hash_table_lookup(table, key) == NULL)) {
pcmk__insert_dup(table, (const char *) key, (const char *) 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->private->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->private->xml, PCMK_XE_META_ATTRIBUTES,
rule_data, parent_orig_meta, NULL, FALSE,
scheduler);
p = p->private->parent;
}
if (parent_orig_meta != NULL) {
// This will not overwrite any values already existing for child
g_hash_table_foreach(parent_orig_meta, dup_attr, 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->private->xml, PCMK_XA_CLASS),
.provider = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER),
.agent = crm_element_value(rsc->private->xml, PCMK_XA_TYPE)
};
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = &rsc_rule_data,
.op_data = NULL
};
if (node) {
/* @COMPAT Support for node attribute expressions in rules for
* meta-attributes is deprecated. When we can break behavioral backward
* compatibility, drop this block.
*/
rule_data.node_hash = node->private->attrs;
}
for (xmlAttrPtr a = pcmk__xe_first_attr(rsc->private->xml);
a != NULL; a = a->next) {
if (a->children != NULL) {
dup_attr((gpointer) a->name, (gpointer) a->children->content,
meta_hash);
}
}
pe__unpack_dataset_nvpairs(rsc->private->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->private->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->private->parent != NULL) {
g_hash_table_foreach(rsc->private->parent->private->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,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
if (node) {
rule_data.node_hash = node->private->attrs;
}
pe__unpack_dataset_nvpairs(rsc->private->xml, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_data, meta_hash, NULL, FALSE, scheduler);
/* set anything else based on the parent */
if (rsc->private->parent != NULL) {
get_rsc_attributes(meta_hash, rsc->private->parent, node, scheduler);
} else {
if (pcmk__xe_first_child(scheduler->rsc_defaults,
PCMK_XE_INSTANCE_ATTRIBUTES, NULL,
NULL) != NULL) {
/* Not possible with schema validation enabled
*
* @COMPAT Drop support when we can break behavioral
* backward compatibility
*/
pcmk__warn_once(pcmk__wo_instance_defaults,
"Support for " PCMK_XE_INSTANCE_ATTRIBUTES " in "
PCMK_XE_RSC_DEFAULTS " is deprecated and will be "
"removed in a future release");
}
/* 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 = pcmk__xe_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_first_child(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 = pcmk__xml_copy(NULL, 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 = pcmk__xe_first_child(new_xml, PCMK_XE_OPERATIONS, NULL,
NULL);
for (child_xml = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL);
child_xml != NULL; child_xml = pcmk__xe_next(child_xml)) {
xmlNode *new_child = pcmk__xml_copy(new_xml, child_xml);
if (pcmk__xe_is(new_child, PCMK_XE_OPERATIONS)) {
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, NULL, NULL, NULL); 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, NULL, NULL, NULL);
op != NULL; op = pcmk__xe_next(op)) {
char *key = template_op_key(op);
if (g_hash_table_lookup(rsc_ops_hash, key) == NULL) {
pcmk__xml_copy(rsc_ops, op);
}
free(key);
}
if (rsc_ops_hash) {
g_hash_table_destroy(rsc_ops_hash);
}
pcmk__xml_free(template_ops);
}
/*pcmk__xml_free(*expanded_xml); */
*expanded_xml = new_xml;
#if 0 /* Disable multi-level templates for now */
if (!unpack_template(new_xml, expanded_xml, scheduler)) {
pcmk__xml_free(*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 = pcmk__xe_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;
}
pcmk__add_idref(scheduler->template_rsc_sets, template_ref, id);
return TRUE;
}
static bool
detect_promotable(pcmk_resource_t *rsc)
{
const char *promotable = g_hash_table_lookup(rsc->private->meta,
PCMK_META_PROMOTABLE);
if (crm_is_true(promotable)) {
return TRUE;
}
// @COMPAT deprecated since 2.0.0
if (pcmk__xe_is(rsc->private->xml, PCMK__XE_PROMOTABLE_LEGACY)) {
pcmk__warn_once(pcmk__wo_master_element,
"Support for <" PCMK__XE_PROMOTABLE_LEGACY "> (such "
"as in %s) is deprecated and will be removed in a "
"future release. Use <" PCMK_XE_CLONE "> with a "
PCMK_META_PROMOTABLE " meta-attribute instead.",
rsc->id);
pcmk__insert_dup(rsc->private->meta, PCMK_META_PROMOTABLE,
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->private->name != NULL)) {
node_name = node->private->name;
}
// Find the parameter table for given node
if (rsc->private->parameter_cache == NULL) {
rsc->private->parameter_cache = pcmk__strikey_table(free,
free_params_table);
} else {
params_on_node = g_hash_table_lookup(rsc->private->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->private->parameter_cache, strdup(node_name),
params_on_node);
}
return params_on_node;
}
/*!
* \internal
* \brief Unpack a resource's \c PCMK_META_REQUIRES meta-attribute
*
* \param[in,out] rsc Resource being unpacked
* \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)
{
const pcmk_scheduler_t *scheduler = rsc->private->scheduler;
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(scheduler->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(scheduler->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 (pcmk__is_primitive(rsc)
&& xml_contains_remote_node(rsc->private->xml)) {
value = PCMK_VALUE_QUORUM;
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) {
value = PCMK_VALUE_UNFENCING;
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
value = PCMK_VALUE_FENCING;
} else if (scheduler->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)" : ""));
}
static void
warn_about_deprecated_classes(pcmk_resource_t *rsc)
{
const char *std = crm_element_value(rsc->private->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);
}
}
/*!
* \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;
pcmk__resource_private_t *rsc_private = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.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);
+ pcmk__sched_err(scheduler,
+ "Unable to allocate memory for resource '%s'", id);
return ENOMEM;
}
(*rsc)->private = calloc(1, sizeof(pcmk__resource_private_t));
if ((*rsc)->private == NULL) {
- pcmk__sched_err("Unable to allocate memory for resource '%s'", id);
+ pcmk__sched_err(scheduler,
+ "Unable to allocate memory for resource '%s'", id);
free(*rsc);
return ENOMEM;
}
rsc_private = (*rsc)->private;
rsc_private->scheduler = scheduler;
if (expanded_xml) {
crm_log_xml_trace(expanded_xml, "[expanded XML]");
rsc_private->xml = expanded_xml;
rsc_private->orig_xml = xml_obj;
} else {
rsc_private->xml = xml_obj;
rsc_private->orig_xml = NULL;
}
/* Do not use xml_obj from here on, use (*rsc)->xml in case templates are involved */
rsc_private->parent = parent;
ops = pcmk__xe_first_child(rsc_private->xml, PCMK_XE_OPERATIONS, NULL,
NULL);
rsc_private->ops_xml = pcmk__xe_resolve_idref(ops, scheduler->input);
rsc_private->variant = get_resource_type((const char *)
rsc_private->xml->name);
if (rsc_private->variant == pcmk__rsc_variant_unknown) {
pcmk__config_err("Ignoring resource '%s' of unknown type '%s'",
id, rsc_private->xml->name);
common_free(*rsc);
*rsc = NULL;
return pcmk_rc_unpack_error;
}
rsc_private->meta = pcmk__strkey_table(free, free);
rsc_private->utilization = pcmk__strkey_table(free, free);
rsc_private->probed_nodes = pcmk__strkey_table(NULL, free);
rsc_private->allowed_nodes = pcmk__strkey_table(NULL, free);
value = crm_element_value(rsc_private->xml, PCMK__META_CLONE);
if (value) {
(*rsc)->id = crm_strdup_printf("%s:%s", id, value);
pcmk__insert_meta(rsc_private, PCMK__META_CLONE, value);
} else {
(*rsc)->id = strdup(id);
}
warn_about_deprecated_classes(*rsc);
rsc_private->fns = &resource_class_functions[rsc_private->variant];
get_meta_attributes(rsc_private->meta, *rsc, NULL, scheduler);
(*rsc)->flags = 0;
pcmk__set_rsc_flags(*rsc, pcmk__rsc_unassigned);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
pcmk__set_rsc_flags(*rsc, pcmk__rsc_managed);
}
rsc_private->orig_role = pcmk_role_stopped;
rsc_private->next_role = pcmk_role_unknown;
rsc_private->ban_after_failures = PCMK_SCORE_INFINITY;
value = g_hash_table_lookup(rsc_private->meta, PCMK_META_PRIORITY);
rsc_private->priority = char2score(value);
value = g_hash_table_lookup(rsc_private->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_private->meta, PCMK_META_NOTIFY);
if (crm_is_true(value)) {
pcmk__set_rsc_flags(*rsc, pcmk__rsc_notify);
}
if (xml_contains_remote_node(rsc_private->xml)) {
pcmk__set_rsc_flags(*rsc, pcmk__rsc_is_remote_connection);
if (g_hash_table_lookup(rsc_private->meta, PCMK__META_CONTAINER)) {
guest_node = true;
} else {
remote_node = true;
}
}
value = g_hash_table_lookup(rsc_private->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_private->meta, PCMK_META_IS_MANAGED);
if (value != NULL) {
if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting " PCMK_META_IS_MANAGED
" to the explicit value '" PCMK_VALUE_DEFAULT
"' is deprecated and will be removed in a "
"future release (just leave it unset)");
} else 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_private->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 (pcmk__is_clone(pe__const_top_resource(*rsc, false))) {
value = g_hash_table_lookup(rsc_private->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_private->meta, PCMK__META_RESTART_TYPE);
if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) {
rsc_private->restart_type = pcmk__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_private->restart_type = pcmk__restart_ignore;
pcmk__rsc_trace(*rsc, "%s dependency restart handling: ignore",
(*rsc)->id);
}
value = g_hash_table_lookup(rsc_private->meta, PCMK_META_MULTIPLE_ACTIVE);
if (pcmk__str_eq(value, PCMK_VALUE_STOP_ONLY, pcmk__str_casei)) {
rsc_private->multiply_active_policy = 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_private->multiply_active_policy = pcmk__multiply_active_block;
pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: block",
(*rsc)->id);
} else if (pcmk__str_eq(value, PCMK_VALUE_STOP_UNEXPECTED,
pcmk__str_casei)) {
rsc_private->multiply_active_policy = pcmk__multiply_active_unexpected;
pcmk__rsc_trace(*rsc,
"%s multiple running resource recovery: "
"stop unexpected instances",
(*rsc)->id);
} else { // PCMK_VALUE_STOP_START
if (!pcmk__str_eq(value, PCMK_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 "
"\"" PCMK_VALUE_STOP_START "\"",
value);
}
rsc_private->multiply_active_policy = pcmk__multiply_active_restart;
pcmk__rsc_trace(*rsc,
"%s multiple running resource recovery: stop/start",
(*rsc)->id);
}
value = g_hash_table_lookup(rsc_private->meta,
PCMK_META_RESOURCE_STICKINESS);
if (value != NULL) {
if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting "
PCMK_META_RESOURCE_STICKINESS
" to the explicit value '" PCMK_VALUE_DEFAULT
"' is deprecated and will be removed in a "
"future release (just leave it unset)");
} else {
rsc_private->stickiness = char2score(value);
}
}
value = g_hash_table_lookup(rsc_private->meta,
PCMK_META_MIGRATION_THRESHOLD);
if (value != NULL) {
if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting "
PCMK_META_MIGRATION_THRESHOLD
" to the explicit value '" PCMK_VALUE_DEFAULT
"' is deprecated and will be removed in a "
"future release (just leave it unset)");
} else {
rsc_private->ban_after_failures = char2score(value);
if (rsc_private->ban_after_failures < 0) {
/* @COMPAT 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_private->ban_after_failures = 1;
}
}
}
if (pcmk__str_eq(crm_element_value(rsc_private->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_private->meta, PCMK_META_REQUIRES);
unpack_requires(*rsc, value, false);
value = g_hash_table_lookup(rsc_private->meta, PCMK_META_FAILURE_TIMEOUT);
if (value != NULL) {
pcmk_parse_interval_spec(value, &(rsc_private->failure_expiration_ms));
}
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_private->remote_reconnect_ms));
/* We want to override any default failure_timeout in use when remote
* PCMK_REMOTE_RA_RECONNECT_INTERVAL is in use.
*/
rsc_private->failure_expiration_ms =
rsc_private->remote_reconnect_ms;
}
}
get_target_role(*rsc, &(rsc_private->next_role));
pcmk__rsc_trace(*rsc, "%s desired next state: %s", (*rsc)->id,
(rsc_private->next_role == pcmk_role_unknown)?
"default" : pcmk_role_text(rsc_private->next_role));
if (rsc_private->fns->unpack(*rsc, scheduler) == FALSE) {
rsc_private->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");
pe__unpack_dataset_nvpairs(rsc_private->xml, PCMK_XE_UTILIZATION,
&rule_data, rsc_private->utilization, NULL,
FALSE, scheduler);
if (expanded_xml) {
if (add_template_rsc(xml_obj, scheduler) == FALSE) {
rsc_private->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->private->parent != NULL) {
if (parent->private->parent == rsc) {
return TRUE;
}
parent = parent->private->parent;
}
return FALSE;
}
pcmk_resource_t *
uber_parent(pcmk_resource_t *rsc)
{
pcmk_resource_t *parent = rsc;
if (parent == NULL) {
return NULL;
}
while ((parent->private->parent != NULL)
&& !pcmk__is_bundle(parent->private->parent)) {
parent = parent->private->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->private->parent != NULL) {
if (!include_bundle && pcmk__is_bundle(parent->private->parent)) {
break;
}
parent = parent->private->parent;
}
return parent;
}
void
common_free(pcmk_resource_t * rsc)
{
if (rsc == NULL) {
return;
}
pcmk__rsc_trace(rsc, "Freeing %s", rsc->id);
if (rsc->private->parameter_cache != NULL) {
g_hash_table_destroy(rsc->private->parameter_cache);
}
if ((rsc->private->parent == NULL)
&& pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
pcmk__xml_free(rsc->private->xml);
rsc->private->xml = NULL;
pcmk__xml_free(rsc->private->orig_xml);
rsc->private->orig_xml = NULL;
} else if (rsc->private->orig_xml != NULL) {
// rsc->private->xml was expanded from a template
pcmk__xml_free(rsc->private->xml);
rsc->private->xml = NULL;
}
free(rsc->id);
free(rsc->private->variant_opaque);
free(rsc->private->history_id);
free(rsc->private->pending_action);
free(rsc->private->assigned_node);
g_list_free(rsc->private->actions);
g_list_free(rsc->private->active_nodes);
g_list_free(rsc->private->launched);
g_list_free(rsc->private->dangling_migration_sources);
g_list_free(rsc->private->with_this_colocations);
g_list_free(rsc->private->this_with_colocations);
g_list_free(rsc->private->location_constraints);
g_list_free(rsc->private->ticket_constraints);
if (rsc->private->meta != NULL) {
g_hash_table_destroy(rsc->private->meta);
}
if (rsc->private->utilization != NULL) {
g_hash_table_destroy(rsc->private->utilization);
}
if (rsc->private->probed_nodes != NULL) {
g_hash_table_destroy(rsc->private->probed_nodes);
}
if (rsc->private->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->private->allowed_nodes);
}
free(rsc->private);
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->private->partial_migration_source != NULL) {
if (pcmk__same_node(node, rsc->private->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->private->active_nodes;
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 \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
* \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;
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) {
return rsc->private->fns->active_node(rsc, count, NULL);
} else {
return rsc->private->fns->active_node(rsc, NULL, count);
}
}
void
pe__count_common(pcmk_resource_t *rsc)
{
if (rsc->private->children != NULL) {
for (GList *item = rsc->private->children;
item != NULL; item = item->next) {
pcmk_resource_t *child = item->data;
child->private->fns->count(item->data);
}
} else if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)
|| (rsc->private->orig_role > pcmk_role_stopped)) {
rsc->private->scheduler->ninstances++;
if (pe__resource_is_disabled(rsc)) {
rsc->private->scheduler->disabled_resources++;
}
if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) {
rsc->private->scheduler->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->private->next_role != role) {
pcmk__rsc_trace(rsc, "Resetting next role for %s from %s to %s (%s)",
rsc->id, pcmk_role_text(rsc->private->next_role),
pcmk_role_text(role), why);
rsc->private->next_role = role;
}
}
diff --git a/lib/pengine/pe_digest.c b/lib/pengine/pe_digest.c
index 6d9842111f..b427b2d0de 100644
--- a/lib/pengine/pe_digest.c
+++ b/lib/pengine/pe_digest.c
@@ -1,611 +1,612 @@
/*
* 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 <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
extern bool pcmk__is_daemon;
/*!
* \internal
* \brief Free an operation digest cache entry
*
* \param[in,out] ptr Pointer to cache entry to free
*
* \note The argument is a gpointer so this can be used as a hash table
* free function.
*/
void
pe__free_digests(gpointer ptr)
{
pcmk__op_digest_t *data = ptr;
if (data != NULL) {
pcmk__xml_free(data->params_all);
pcmk__xml_free(data->params_secure);
pcmk__xml_free(data->params_restart);
free(data->digest_all_calc);
free(data->digest_restart_calc);
free(data->digest_secure_calc);
free(data);
}
}
// Return true if XML attribute name is not substring of a given string
static bool
attr_not_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) == NULL) {
crm_trace("Filtering %s (not found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
// Return true if XML attribute name is substring of a given string
static bool
attr_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) != NULL) {
crm_trace("Filtering %s (found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
/*!
* \internal
* \brief Add digest of all parameters to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in,out] rsc Resource that action was for
* \param[in] node Node action was performed on
* \param[in] params Resource parameters evaluated for node
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] xml_op Unused
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value table to override resource parameters
* \param[in,out] scheduler Scheduler data
*/
static void
calculate_main_digest(pcmk__op_digest_t *data, pcmk_resource_t *rsc,
const pcmk_node_t *node, GHashTable *params,
const char *task, guint *interval_ms,
const xmlNode *xml_op, const char *op_version,
GHashTable *overrides, pcmk_scheduler_t *scheduler)
{
xmlNode *action_config = NULL;
data->params_all = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
/* REMOTE_CONTAINER_HACK: Allow Pacemaker Remote nodes to run containers
* that themselves are Pacemaker Remote nodes
*/
(void) pe__add_bundle_remote_name(rsc, data->params_all,
PCMK_REMOTE_RA_ADDR);
if (overrides != NULL) {
// If interval was overridden, reset it
const char *meta_name = CRM_META "_" PCMK_META_INTERVAL;
const char *interval_s = g_hash_table_lookup(overrides, meta_name);
if (interval_s != NULL) {
long long value_ll;
if ((pcmk__scan_ll(interval_s, &value_ll, 0LL) == pcmk_rc_ok)
&& (value_ll >= 0) && (value_ll <= G_MAXUINT)) {
*interval_ms = (guint) value_ll;
}
}
// Add overrides to list of all parameters
g_hash_table_foreach(overrides, hash2field, data->params_all);
}
// Add provided instance parameters
g_hash_table_foreach(params, hash2field, data->params_all);
// Find action configuration XML in CIB
action_config = pcmk__find_action_config(rsc, task, *interval_ms, true);
/* Add action-specific resource instance attributes to the digest list.
*
* If this is a one-time action with action-specific instance attributes,
* enforce a restart instead of reload-agent in case the main digest doesn't
* match, even if the restart digest does. This ensures any changes of the
* action-specific parameters get applied for this specific action, and
* digests calculated for the resulting history will be correct. Default the
* result to RSC_DIGEST_RESTART for the case where the main digest doesn't
* match.
*/
params = pcmk__unpack_action_rsc_params(action_config, node->private->attrs,
scheduler);
if ((*interval_ms == 0) && (g_hash_table_size(params) > 0)) {
data->rc = pcmk__digest_restart;
}
g_hash_table_foreach(params, hash2field, data->params_all);
g_hash_table_destroy(params);
// Add action meta-attributes
params = pcmk__unpack_action_meta(rsc, node, task, *interval_ms,
action_config);
g_hash_table_foreach(params, hash2metafield, data->params_all);
g_hash_table_destroy(params);
pcmk__filter_op_for_digest(data->params_all);
data->digest_all_calc = pcmk__digest_operation(data->params_all);
}
// Return true if XML attribute name is a Pacemaker-defined fencing parameter
static bool
is_fence_param(xmlAttrPtr attr, void *user_data)
{
return pcmk_stonith_param((const char *) attr->name);
}
/*!
* \internal
* \brief Add secure digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] rsc Resource that action was for
* \param[in] params Resource parameters evaluated for node
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value hash table to override resource parameters
*/
static void
calculate_secure_digest(pcmk__op_digest_t *data, const pcmk_resource_t *rsc,
GHashTable *params, const xmlNode *xml_op,
const char *op_version, GHashTable *overrides)
{
const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS);
const char *secure_list = NULL;
bool old_version = (compare_version(op_version, "3.16.0") < 0);
if (xml_op == NULL) {
secure_list = " passwd password user ";
} else {
secure_list = crm_element_value(xml_op, PCMK__XA_OP_SECURE_PARAMS);
}
if (old_version) {
data->params_secure = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
if (overrides != NULL) {
g_hash_table_foreach(overrides, hash2field, data->params_secure);
}
g_hash_table_foreach(params, hash2field, data->params_secure);
} else {
// Start with a copy of all parameters
data->params_secure = pcmk__xml_copy(NULL, data->params_all);
}
if (secure_list != NULL) {
pcmk__xe_remove_matching_attrs(data->params_secure, attr_in_string,
(void *) secure_list);
}
if (old_version
&& pcmk_is_set(pcmk_get_ra_caps(class),
pcmk_ra_cap_fence_params)) {
/* For stonith resources, Pacemaker adds special parameters,
* but these are not listed in fence agent meta-data, so with older
* versions of DC, the controller will not hash them. That means we have
* to filter them out before calculating our hash for comparison.
*/
pcmk__xe_remove_matching_attrs(data->params_secure, is_fence_param,
NULL);
}
pcmk__filter_op_for_digest(data->params_secure);
/* CRM_meta_timeout *should* be part of a digest for recurring operations.
* However, with older versions of DC, the controller does not add timeout
* to secure digests, because it only includes parameters declared by the
* resource agent.
* Remove any timeout that made it this far, to match.
*/
if (old_version) {
pcmk__xe_remove_attr(data->params_secure,
CRM_META "_" PCMK_META_TIMEOUT);
}
data->digest_secure_calc = pcmk__digest_operation(data->params_secure);
}
/*!
* \internal
* \brief Add restart digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
*
* \note This function doesn't need to handle overrides because it starts with
* data->params_all, which already has overrides applied.
*/
static void
calculate_restart_digest(pcmk__op_digest_t *data, const xmlNode *xml_op,
const char *op_version)
{
const char *value = NULL;
// We must have XML of resource operation history
if (xml_op == NULL) {
return;
}
// And the history must have a restart digest to compare against
if (crm_element_value(xml_op, PCMK__XA_OP_RESTART_DIGEST) == NULL) {
return;
}
// Start with a copy of all parameters
data->params_restart = pcmk__xml_copy(NULL, data->params_all);
// Then filter out reloadable parameters, if any
value = crm_element_value(xml_op, PCMK__XA_OP_FORCE_RESTART);
if (value != NULL) {
pcmk__xe_remove_matching_attrs(data->params_restart, attr_not_in_string,
(void *) value);
}
value = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
data->digest_restart_calc = pcmk__digest_operation(data->params_restart);
}
/*!
* \internal
* \brief Create a new digest cache entry with calculated digests
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] overrides Key/value table to override resource parameters
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to new digest cache entry (or NULL on memory error)
* \note It is the caller's responsibility to free the result using
* pe__free_digests().
*/
pcmk__op_digest_t *
pe__calculate_digests(pcmk_resource_t *rsc, const char *task,
guint *interval_ms, const pcmk_node_t *node,
const xmlNode *xml_op, GHashTable *overrides,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
const char *op_version = NULL;
GHashTable *params = NULL;
CRM_CHECK(scheduler != NULL, return NULL);
data = calloc(1, sizeof(pcmk__op_digest_t));
if (data == NULL) {
- pcmk__sched_err("Could not allocate memory for operation digest");
+ pcmk__sched_err(scheduler,
+ "Could not allocate memory for operation digest");
return NULL;
}
data->rc = pcmk__digest_match;
if (xml_op != NULL) {
op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
}
if ((op_version == NULL) && (scheduler->input != NULL)) {
op_version = crm_element_value(scheduler->input,
PCMK_XA_CRM_FEATURE_SET);
}
if (op_version == NULL) {
op_version = CRM_FEATURE_SET;
}
params = pe_rsc_params(rsc, node, scheduler);
calculate_main_digest(data, rsc, node, params, task, interval_ms, xml_op,
op_version, overrides, scheduler);
if (calc_secure) {
calculate_secure_digest(data, rsc, params, xml_op, op_version,
overrides);
}
calculate_restart_digest(data, xml_op, op_version);
return data;
}
/*!
* \internal
* \brief Calculate action digests and store in node's digest cache
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in] interval_ms Action's interval
* \param[in,out] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry
*/
static pcmk__op_digest_t *
rsc_action_digest(pcmk_resource_t *rsc, const char *task, guint interval_ms,
pcmk_node_t *node, const xmlNode *xml_op,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
char *key = pcmk__op_key(rsc->id, task, interval_ms);
data = g_hash_table_lookup(node->private->digest_cache, key);
if (data == NULL) {
data = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op,
NULL, calc_secure, scheduler);
CRM_ASSERT(data != NULL);
g_hash_table_insert(node->private->digest_cache, strdup(key), data);
}
free(key);
return data;
}
/*!
* \internal
* \brief Calculate operation digests and compare against an XML history entry
*
* \param[in,out] rsc Resource to check
* \param[in] xml_op Resource history XML
* \param[in,out] node Node to use for digest calculation
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry, with comparison result set
*/
pcmk__op_digest_t *
rsc_action_digest_cmp(pcmk_resource_t *rsc, const xmlNode *xml_op,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
guint interval_ms = 0;
const char *op_version;
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
const char *digest_all;
const char *digest_restart;
CRM_ASSERT(node != NULL);
op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
digest_all = crm_element_value(xml_op, PCMK__XA_OP_DIGEST);
digest_restart = crm_element_value(xml_op, PCMK__XA_OP_RESTART_DIGEST);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
data = rsc_action_digest(rsc, task, interval_ms, node, xml_op,
pcmk_is_set(scheduler->flags,
pcmk_sched_sanitized),
scheduler);
if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) {
pcmk__rsc_info(rsc,
"Parameters to %ums-interval %s action for %s on %s "
"changed: hash was %s vs. now %s (restart:%s) %s",
interval_ms, task, rsc->id, pcmk__node_name(node),
pcmk__s(digest_restart, "missing"),
data->digest_restart_calc, op_version,
crm_element_value(xml_op, PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_restart;
} else if (digest_all == NULL) {
/* it is unknown what the previous op digest was */
data->rc = pcmk__digest_unknown;
} else if (strcmp(digest_all, data->digest_all_calc) != 0) {
/* Given a non-recurring operation with extra parameters configured,
* in case that the main digest doesn't match, even if the restart
* digest matches, enforce a restart rather than a reload-agent anyway.
* So that it ensures any changes of the extra parameters get applied
* for this specific operation, and the digests calculated for the
* resulting PCMK__XE_LRM_RSC_OP will be correct.
* Preserve the implied rc pcmk__digest_restart for the case that the
* main digest doesn't match.
*/
if ((interval_ms == 0) && (data->rc == pcmk__digest_restart)) {
pcmk__rsc_info(rsc,
"Parameters containing extra ones to %ums-interval"
" %s action for %s on %s "
"changed: hash was %s vs. now %s (restart:%s) %s",
interval_ms, task, rsc->id, pcmk__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc, op_version,
crm_element_value(xml_op,
PCMK__XA_TRANSITION_MAGIC));
} else {
pcmk__rsc_info(rsc,
"Parameters to %ums-interval %s action for %s on %s "
"changed: hash was %s vs. now %s (%s:%s) %s",
interval_ms, task, rsc->id, pcmk__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc,
(interval_ms > 0)? "reschedule" : "reload",
op_version,
crm_element_value(xml_op,
PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_mismatch;
}
} else {
data->rc = pcmk__digest_match;
}
return data;
}
/*!
* \internal
* \brief Create an unfencing summary for use in special node attribute
*
* Create a string combining a fence device's resource ID, agent type, and
* parameter digest (whether for all parameters or just non-private parameters).
* This can be stored in a special node attribute, allowing us to detect changes
* in either the agent type or parameters, to know whether unfencing must be
* redone or can be safely skipped when the device's history is cleaned.
*
* \param[in] rsc_id Fence device resource ID
* \param[in] agent_type Fence device agent
* \param[in] param_digest Fence device parameter digest
*
* \return Newly allocated string with unfencing digest
* \note The caller is responsible for freeing the result.
*/
static inline char *
create_unfencing_summary(const char *rsc_id, const char *agent_type,
const char *param_digest)
{
return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest);
}
/*!
* \internal
* \brief Check whether a node can skip unfencing
*
* Check whether a fence device's current definition matches a node's
* stored summary of when it was last unfenced by the device.
*
* \param[in] rsc_id Fence device's resource ID
* \param[in] agent Fence device's agent type
* \param[in] digest_calc Fence device's current parameter digest
* \param[in] node_summary Value of node's special unfencing node attribute
* (a comma-separated list of unfencing summaries for
* all devices that have unfenced this node)
*
* \return TRUE if digest matches, FALSE otherwise
*/
static bool
unfencing_digest_matches(const char *rsc_id, const char *agent,
const char *digest_calc, const char *node_summary)
{
bool matches = FALSE;
if (rsc_id && agent && digest_calc && node_summary) {
char *search_secure = create_unfencing_summary(rsc_id, agent,
digest_calc);
/* The digest was calculated including the device ID and agent,
* so there is no risk of collision using strstr().
*/
matches = (strstr(node_summary, search_secure) != NULL);
crm_trace("Calculated unfencing digest '%s' %sfound in '%s'",
search_secure, matches? "" : "not ", node_summary);
free(search_secure);
}
return matches;
}
/* Magic string to use as action name for digest cache entries used for
* unfencing checks. This is not a real action name (i.e. "on"), so
* pcmk__check_action_config() won't confuse these entries with real actions.
*/
#define STONITH_DIGEST_TASK "stonith-on"
/*!
* \internal
* \brief Calculate fence device digests and digest comparison result
*
* \param[in,out] rsc Fence device resource
* \param[in] agent Fence device's agent type
* \param[in,out] node Node with digest cache to use
* \param[in,out] scheduler Scheduler data
*
* \return Node's digest cache entry
*/
pcmk__op_digest_t *
pe__compare_fencing_digest(pcmk_resource_t *rsc, const char *agent,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
const char *node_summary = NULL;
// Calculate device's current parameter digests
pcmk__op_digest_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, 0U,
node, NULL, TRUE, scheduler);
// Check whether node has special unfencing summary node attribute
node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_ALL, NULL,
pcmk__rsc_node_current);
if (node_summary == NULL) {
data->rc = pcmk__digest_unknown;
return data;
}
// Check whether full parameter digest matches
if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc,
node_summary)) {
data->rc = pcmk__digest_match;
return data;
}
// Check whether secure parameter digest matches
node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_SECURE, NULL,
pcmk__rsc_node_current);
if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc,
node_summary)) {
data->rc = pcmk__digest_match;
if (!pcmk__is_daemon && scheduler->priv != NULL) {
pcmk__output_t *out = scheduler->priv;
out->info(out, "Only 'private' parameters to %s "
"for unfencing %s changed", rsc->id,
pcmk__node_name(node));
}
return data;
}
// Parameters don't match
data->rc = pcmk__digest_mismatch;
if (pcmk_is_set(scheduler->flags, pcmk_sched_sanitized)
&& (data->digest_secure_calc != NULL)) {
if (scheduler->priv != NULL) {
pcmk__output_t *out = scheduler->priv;
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
out->info(out, "Parameters to %s for unfencing "
"%s changed, try '%s'", rsc->id,
pcmk__node_name(node), digest);
free(digest);
} else if (!pcmk__is_daemon) {
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
printf("Parameters to %s for unfencing %s changed, try '%s'\n",
rsc->id, pcmk__node_name(node), digest);
free(digest);
}
}
return data;
}
diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c
index 3902024bde..bf2e806d0d 100644
--- a/lib/pengine/pe_notif.c
+++ b/lib/pengine/pe_notif.c
@@ -1,1014 +1,1016 @@
/*
* 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 <crm/common/xml.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include "pe_status_private.h"
typedef struct notify_entry_s {
const pcmk_resource_t *rsc;
const pcmk_node_t *node;
} notify_entry_t;
/*!
* \internal
* \brief Compare two notification entries
*
* Compare two notification entries, where the one with the alphabetically first
* resource name (or if equal, node ID) sorts as first, with NULL sorting as
* less than non-NULL.
*
* \param[in] a First notification entry to compare
* \param[in] b Second notification entry to compare
*
* \return -1 if \p a sorts before \p b, 0 if they are equal, otherwise 1
*/
static gint
compare_notify_entries(gconstpointer a, gconstpointer b)
{
int tmp;
const notify_entry_t *entry_a = a;
const notify_entry_t *entry_b = b;
// NULL a or b is not actually possible
if ((entry_a == NULL) && (entry_b == NULL)) {
return 0;
}
if (entry_a == NULL) {
return 1;
}
if (entry_b == NULL) {
return -1;
}
// NULL resources sort first
if ((entry_a->rsc == NULL) && (entry_b->rsc == NULL)) {
return 0;
}
if (entry_a->rsc == NULL) {
return 1;
}
if (entry_b->rsc == NULL) {
return -1;
}
// Compare resource names
tmp = strcmp(entry_a->rsc->id, entry_b->rsc->id);
if (tmp != 0) {
return tmp;
}
// Otherwise NULL nodes sort first
if ((entry_a->node == NULL) && (entry_b->node == NULL)) {
return 0;
}
if (entry_a->node == NULL) {
return 1;
}
if (entry_b->node == NULL) {
return -1;
}
// Finally, compare node IDs
return strcmp(entry_a->node->private->id, entry_b->node->private->id);
}
/*!
* \internal
* \brief Duplicate a notification entry
*
* \param[in] entry Entry to duplicate
*
* \return Newly allocated duplicate of \p entry
* \note It is the caller's responsibility to free the return value.
*/
static notify_entry_t *
dup_notify_entry(const notify_entry_t *entry)
{
notify_entry_t *dup = pcmk__assert_alloc(1, sizeof(notify_entry_t));
dup->rsc = entry->rsc;
dup->node = entry->node;
return dup;
}
/*!
* \internal
* \brief Given a list of nodes, create strings with node names
*
* \param[in] list List of nodes (as pcmk_node_t *)
* \param[out] all_node_names If not NULL, will be set to space-separated list
* of the names of all nodes in \p list
* \param[out] host_node_names Same as \p all_node_names, except active
* guest nodes will list the name of their host
*
* \note The caller is responsible for freeing the output argument values using
* \p g_string_free().
*/
static void
get_node_names(const GList *list, GString **all_node_names,
GString **host_node_names)
{
if (all_node_names != NULL) {
*all_node_names = NULL;
}
if (host_node_names != NULL) {
*host_node_names = NULL;
}
for (const GList *iter = list; iter != NULL; iter = iter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) iter->data;
if (node->private->name == NULL) {
continue;
}
// Always add to list of all node names
if (all_node_names != NULL) {
pcmk__add_word(all_node_names, 1024, node->private->name);
}
// Add to host node name list if appropriate
if (host_node_names != NULL) {
if (pcmk__is_guest_or_bundle_node(node)) {
const pcmk_resource_t *launcher = NULL;
launcher = node->private->remote->private->launcher;
if (launcher->private->active_nodes != NULL) {
node = pcmk__current_node(launcher);
if (node->private->name == NULL) {
continue;
}
}
}
pcmk__add_word(host_node_names, 1024, node->private->name);
}
}
if ((all_node_names != NULL) && (*all_node_names == NULL)) {
*all_node_names = g_string_new(" ");
}
if ((host_node_names != NULL) && (*host_node_names == NULL)) {
*host_node_names = g_string_new(" ");
}
}
/*!
* \internal
* \brief Create strings of instance and node names from notification entries
*
* \param[in,out] list List of notification entries (will be sorted here)
* \param[out] rsc_names If not NULL, will be set to space-separated list
* of clone instances from \p list
* \param[out] node_names If not NULL, will be set to space-separated list
* of node names from \p list
*
* \return (Possibly new) head of sorted \p list
* \note The caller is responsible for freeing the output argument values using
* \p g_list_free_full() and \p g_string_free().
*/
static GList *
notify_entries_to_strings(GList *list, GString **rsc_names,
GString **node_names)
{
const char *last_rsc_id = NULL;
// Initialize output lists to NULL
if (rsc_names != NULL) {
*rsc_names = NULL;
}
if (node_names != NULL) {
*node_names = NULL;
}
// Sort input list for user-friendliness (and ease of filtering duplicates)
list = g_list_sort(list, compare_notify_entries);
for (GList *gIter = list; gIter != NULL; gIter = gIter->next) {
notify_entry_t *entry = (notify_entry_t *) gIter->data;
// Entry must have a resource (with ID)
CRM_LOG_ASSERT((entry != NULL) && (entry->rsc != NULL)
&& (entry->rsc->id != NULL));
if ((entry == NULL) || (entry->rsc == NULL)
|| (entry->rsc->id == NULL)) {
continue;
}
// Entry must have a node unless listing inactive resources
CRM_LOG_ASSERT((node_names == NULL) || (entry->node != NULL));
if ((node_names != NULL) && (entry->node == NULL)) {
continue;
}
// Don't add duplicates of a particular clone instance
if (pcmk__str_eq(entry->rsc->id, last_rsc_id, pcmk__str_none)) {
continue;
}
last_rsc_id = entry->rsc->id;
if (rsc_names != NULL) {
pcmk__add_word(rsc_names, 1024, entry->rsc->id);
}
if ((node_names != NULL) && (entry->node->private->name != NULL)) {
pcmk__add_word(node_names, 1024, entry->node->private->name);
}
}
// If there are no entries, return "empty" lists
if ((rsc_names != NULL) && (*rsc_names == NULL)) {
*rsc_names = g_string_new(" ");
}
if ((node_names != NULL) && (*node_names == NULL)) {
*node_names = g_string_new(" ");
}
return list;
}
/*!
* \internal
* \brief Copy a meta-attribute into a notify action
*
* \param[in] key Name of meta-attribute to copy
* \param[in] value Value of meta-attribute to copy
* \param[in,out] user_data Notify action to copy into
*/
static void
copy_meta_to_notify(gpointer key, gpointer value, gpointer user_data)
{
pcmk_action_t *notify = (pcmk_action_t *) user_data;
/* Any existing meta-attributes (for example, the action timeout) are for
* the notify action itself, so don't override those.
*/
if (g_hash_table_lookup(notify->meta, (const char *) key) != NULL) {
return;
}
pcmk__insert_dup(notify->meta, (const char *) key, (const char *) value);
}
static void
add_notify_data_to_action_meta(const notify_data_t *n_data,
pcmk_action_t *action)
{
for (const GSList *item = n_data->keys; item; item = item->next) {
const pcmk_nvpair_t *nvpair = (const pcmk_nvpair_t *) item->data;
pcmk__insert_meta(action, nvpair->name, nvpair->value);
}
}
/*!
* \internal
* \brief Create a new notify pseudo-action for a clone resource
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in] action Action to use in notify action key
* \param[in] notif_action PCMK_ACTION_NOTIFY or PCMK_ACTION_NOTIFIED
* \param[in] notif_type "pre", "post", "confirmed-pre", "confirmed-post"
*
* \return Newly created notify pseudo-action
*/
static pcmk_action_t *
new_notify_pseudo_action(pcmk_resource_t *rsc, const pcmk_action_t *action,
const char *notif_action, const char *notif_type)
{
pcmk_action_t *notify = NULL;
notify = custom_action(rsc,
pcmk__notify_key(rsc->id, notif_type, action->task),
notif_action, NULL,
pcmk_is_set(action->flags, pcmk__action_optional),
rsc->private->scheduler);
pcmk__set_action_flags(notify, pcmk__action_pseudo);
pcmk__insert_meta(notify, "notify_key_type", notif_type);
pcmk__insert_meta(notify, "notify_key_operation", action->task);
return notify;
}
/*!
* \internal
* \brief Create a new notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] op Action that notification is for
* \param[in,out] notify_done Parent pseudo-action for notifications complete
* \param[in] n_data Notification values to add to action meta-data
*
* \return Newly created notify action
*/
static pcmk_action_t *
new_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
pcmk_action_t *op, pcmk_action_t *notify_done,
const notify_data_t *n_data)
{
char *key = NULL;
pcmk_action_t *notify_action = NULL;
const char *value = NULL;
const char *task = NULL;
const char *skip_reason = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL), return NULL);
// Ensure we have all the info we need
if (op == NULL) {
skip_reason = "no action";
} else if (notify_done == NULL) {
skip_reason = "no parent notification";
} else if (!node->details->online) {
skip_reason = "node offline";
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
skip_reason = "original action not runnable";
}
if (skip_reason != NULL) {
pcmk__rsc_trace(rsc, "Skipping notify action for %s on %s: %s",
rsc->id, pcmk__node_name(node), skip_reason);
return NULL;
}
value = g_hash_table_lookup(op->meta, "notify_type"); // "pre" or "post"
task = g_hash_table_lookup(op->meta, "notify_operation"); // original action
pcmk__rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)",
rsc->id, pcmk__node_name(node), value, task);
// Create the notify action
key = pcmk__notify_key(rsc->id, value, task);
notify_action = custom_action(rsc, key, op->task, node,
pcmk_is_set(op->flags, pcmk__action_optional),
rsc->private->scheduler);
// Add meta-data to notify action
g_hash_table_foreach(op->meta, copy_meta_to_notify, notify_action);
add_notify_data_to_action_meta(n_data, notify_action);
// Order notify after original action and before parent notification
order_actions(op, notify_action, pcmk__ar_ordered);
order_actions(notify_action, notify_done, pcmk__ar_ordered);
return notify_action;
}
/*!
* \internal
* \brief Create a new "post-" notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] n_data Notification values to add to action meta-data
*/
static void
new_post_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
notify_data_t *n_data)
{
pcmk_action_t *notify = NULL;
CRM_ASSERT(n_data != NULL);
// Create the "post-" notify action for specified instance
notify = new_notify_action(rsc, node, n_data->post, n_data->post_done,
n_data);
if (notify != NULL) {
notify->priority = PCMK_SCORE_INFINITY;
}
// Order recurring monitors after all "post-" notifications complete
if (n_data->post_done == NULL) {
return;
}
for (GList *iter = rsc->private->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *mon = (pcmk_action_t *) iter->data;
const char *interval_ms_s = NULL;
interval_ms_s = g_hash_table_lookup(mon->meta, PCMK_META_INTERVAL);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)
|| pcmk__str_eq(mon->task, PCMK_ACTION_CANCEL, pcmk__str_none)) {
continue; // Not a recurring monitor
}
order_actions(n_data->post_done, mon, pcmk__ar_ordered);
}
}
/*!
* \internal
* \brief Create and order notification pseudo-actions for a clone action
*
* In addition to the actual notify actions needed for each clone instance,
* clone notifications also require pseudo-actions to provide ordering points
* in the notification process. This creates the notification data, along with
* appropriate pseudo-actions and their orderings.
*
* For example, the ordering sequence for starting a clone is:
*
* "pre-" notify pseudo-action for clone
* -> "pre-" notify actions for each clone instance
* -> "pre-" notifications complete pseudo-action for clone
* -> start actions for each clone instance
* -> "started" pseudo-action for clone
* -> "post-" notify pseudo-action for clone
* -> "post-" notify actions for each clone instance
* -> "post-" notifications complete pseudo-action for clone
*
* \param[in,out] rsc Clone that notifications are for
* \param[in] task Name of action that notifications are for
* \param[in,out] action If not NULL, create a "pre-" pseudo-action ordered
* before a "pre-" complete pseudo-action, ordered
* before this action
* \param[in,out] complete If not NULL, create a "post-" pseudo-action ordered
* after this action, and a "post-" complete
* pseudo-action ordered after that
*
* \return Newly created notification data
*/
notify_data_t *
pe__action_notif_pseudo_ops(pcmk_resource_t *rsc, const char *task,
pcmk_action_t *action, pcmk_action_t *complete)
{
notify_data_t *n_data = NULL;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_notify)) {
return NULL;
}
n_data = pcmk__assert_alloc(1, sizeof(notify_data_t));
n_data->action = task;
if (action != NULL) { // Need "pre-" pseudo-actions
// Create "pre-" notify pseudo-action for clone
n_data->pre = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFY,
"pre");
pcmk__set_action_flags(n_data->pre, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre, "notify_type", "pre");
pcmk__insert_meta(n_data->pre, "notify_operation", n_data->action);
// Create "pre-" notifications complete pseudo-action for clone
n_data->pre_done = new_notify_pseudo_action(rsc, action,
PCMK_ACTION_NOTIFIED,
"confirmed-pre");
pcmk__set_action_flags(n_data->pre_done, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre_done, "notify_type", "pre");
pcmk__insert_meta(n_data->pre_done, "notify_operation", n_data->action);
// Order "pre-" -> "pre-" complete -> original action
order_actions(n_data->pre, n_data->pre_done, pcmk__ar_ordered);
order_actions(n_data->pre_done, action, pcmk__ar_ordered);
}
if (complete != NULL) { // Need "post-" pseudo-actions
// Create "post-" notify pseudo-action for clone
n_data->post = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFY, "post");
n_data->post->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post, "notify_type", "post");
pcmk__insert_meta(n_data->post, "notify_operation", n_data->action);
// Create "post-" notifications complete pseudo-action for clone
n_data->post_done = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFIED,
"confirmed-post");
n_data->post_done->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post_done, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post_done, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post_done, "notify_type", "post");
pcmk__insert_meta(n_data->post_done,
"notify_operation", n_data->action);
// Order original action complete -> "post-" -> "post-" complete
order_actions(complete, n_data->post, pcmk__ar_first_implies_then);
order_actions(n_data->post, n_data->post_done,
pcmk__ar_first_implies_then);
}
// If we created both, order "pre-" complete -> "post-"
if ((action != NULL) && (complete != NULL)) {
order_actions(n_data->pre_done, n_data->post, pcmk__ar_ordered);
}
return n_data;
}
/*!
* \internal
* \brief Create a new notification entry
*
* \param[in] rsc Resource for notification
* \param[in] node Node for notification
*
* \return Newly allocated notification entry
* \note The caller is responsible for freeing the return value.
*/
static notify_entry_t *
new_notify_entry(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
notify_entry_t *entry = pcmk__assert_alloc(1, sizeof(notify_entry_t));
entry->rsc = rsc;
entry->node = node;
return entry;
}
/*!
* \internal
* \brief Add notification data for resource state and optionally actions
*
- * \param[in] rsc Clone or clone instance being notified
+ * \param[in,out] rsc Clone or clone instance being notified
* \param[in] activity Whether to add notification entries for actions
* \param[in,out] n_data Notification data for clone
*/
static void
-collect_resource_data(const pcmk_resource_t *rsc, bool activity,
+collect_resource_data(pcmk_resource_t *rsc, bool activity,
notify_data_t *n_data)
{
const GList *iter = NULL;
notify_entry_t *entry = NULL;
const pcmk_node_t *node = NULL;
if (n_data == NULL) {
return;
}
if (n_data->allowed_nodes == NULL) {
n_data->allowed_nodes = rsc->private->allowed_nodes;
}
// If this is a clone, call recursively for each instance
if (rsc->private->children != NULL) {
for (iter = rsc->private->children; iter != NULL; iter = iter->next) {
- const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data;
+ pcmk_resource_t *child = iter->data;
collect_resource_data(child, activity, n_data);
}
return;
}
// This is a notification for a single clone instance
if (rsc->private->active_nodes != NULL) {
node = rsc->private->active_nodes->data; // First is sufficient
}
entry = new_notify_entry(rsc, node);
// Add notification indicating the resource state
switch (rsc->private->orig_role) {
case pcmk_role_stopped:
n_data->inactive = g_list_prepend(n_data->inactive, entry);
break;
case pcmk_role_started:
n_data->active = g_list_prepend(n_data->active, entry);
break;
case pcmk_role_unpromoted:
n_data->unpromoted = g_list_prepend(n_data->unpromoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
case pcmk_role_promoted:
n_data->promoted = g_list_prepend(n_data->promoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
default:
- pcmk__sched_err("Resource %s role on %s (%s) is not supported for "
+ pcmk__sched_err(rsc->private->scheduler,
+ "Resource %s role on %s (%s) is not supported for "
"notifications (bug?)",
rsc->id, pcmk__node_name(node),
pcmk_role_text(rsc->private->orig_role));
free(entry);
break;
}
if (!activity) {
return;
}
// Add notification entries for each of the resource's actions
for (iter = rsc->private->actions; iter != NULL; iter = iter->next) {
const pcmk_action_t *op = (const pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
enum pcmk__action_type task = pcmk__parse_action(op->task);
if ((task == pcmk__action_stop) && op->node->details->unclean) {
// Create anyway (additional noise if node can't be fenced)
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
continue;
}
entry = new_notify_entry(rsc, op->node);
switch (task) {
case pcmk__action_start:
n_data->start = g_list_prepend(n_data->start, entry);
break;
case pcmk__action_stop:
n_data->stop = g_list_prepend(n_data->stop, entry);
break;
case pcmk__action_promote:
n_data->promote = g_list_prepend(n_data->promote, entry);
break;
case pcmk__action_demote:
n_data->demote = g_list_prepend(n_data->demote, entry);
break;
default:
free(entry);
break;
}
}
}
}
// For (char *) value
#define add_notify_env(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, value); \
} while (0)
// For (GString *) value
#define add_notify_env_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
} while (0)
// For (GString *) value
#define add_notify_env_free_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
g_string_free(value, TRUE); value = NULL; \
} while (0)
/*!
* \internal
* \brief Create notification name/value pairs from structured data
*
* \param[in] rsc Resource that notification is for
* \param[in,out] n_data Notification data
*/
static void
add_notif_keys(const pcmk_resource_t *rsc, notify_data_t *n_data)
{
bool required = false; // Whether to make notify actions required
GString *rsc_list = NULL;
GString *node_list = NULL;
GString *metal_list = NULL;
const char *source = NULL;
GList *nodes = NULL;
n_data->stop = notify_entries_to_strings(n_data->stop,
&rsc_list, &node_list);
if ((strcmp(" ", (const char *) rsc_list->str) != 0)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_STOP, pcmk__str_none)) {
required = true;
}
add_notify_env_free_gs(n_data, "notify_stop_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_stop_uname", node_list);
if ((n_data->start != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_START, pcmk__str_none)) {
required = true;
}
n_data->start = notify_entries_to_strings(n_data->start,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_start_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_start_uname", node_list);
if ((n_data->demote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
required = true;
}
n_data->demote = notify_entries_to_strings(n_data->demote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_demote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_demote_uname", node_list);
if ((n_data->promote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
required = true;
}
n_data->promote = notify_entries_to_strings(n_data->promote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_promote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_promote_uname", node_list);
n_data->active = notify_entries_to_strings(n_data->active,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_active_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_active_uname", node_list);
n_data->unpromoted = notify_entries_to_strings(n_data->unpromoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_unpromoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_unpromoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_slave_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_slave_uname", node_list);
n_data->promoted = notify_entries_to_strings(n_data->promoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_promoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_promoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_master_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_master_uname", node_list);
n_data->inactive = notify_entries_to_strings(n_data->inactive,
&rsc_list, NULL);
add_notify_env_free_gs(n_data, "notify_inactive_resource", rsc_list);
nodes = g_hash_table_get_values(n_data->allowed_nodes);
if (!pcmk__is_daemon) {
/* For display purposes, sort the node list, for consistent
* regression test output (while avoiding the performance hit
* for the live cluster).
*/
nodes = g_list_sort(nodes, pe__cmp_node_name);
}
get_node_names(nodes, &node_list, NULL);
add_notify_env_free_gs(n_data, "notify_available_uname", node_list);
g_list_free(nodes);
source = g_hash_table_lookup(rsc->private->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
if (pcmk__str_eq(PCMK_VALUE_HOST, source, pcmk__str_none)) {
get_node_names(rsc->private->scheduler->nodes, &node_list, &metal_list);
add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list);
} else {
get_node_names(rsc->private->scheduler->nodes, &node_list, NULL);
}
add_notify_env_free_gs(n_data, "notify_all_uname", node_list);
if (required && (n_data->pre != NULL)) {
pcmk__clear_action_flags(n_data->pre, pcmk__action_optional);
pcmk__clear_action_flags(n_data->pre_done, pcmk__action_optional);
}
if (required && (n_data->post != NULL)) {
pcmk__clear_action_flags(n_data->post, pcmk__action_optional);
pcmk__clear_action_flags(n_data->post_done, pcmk__action_optional);
}
}
/*
* \internal
* \brief Find any remote connection start relevant to an action
*
* \param[in] action Action to check
*
* \return If action is behind a remote connection, connection's start
*/
static pcmk_action_t *
find_remote_start(pcmk_action_t *action)
{
if ((action != NULL) && (action->node != NULL)) {
pcmk_resource_t *remote_rsc = action->node->private->remote;
if (remote_rsc != NULL) {
return find_first_action(remote_rsc->private->actions, NULL,
PCMK_ACTION_START,
NULL);
}
}
return NULL;
}
/*!
* \internal
* \brief Create notify actions, and add notify data to original actions
*
* \param[in,out] rsc Clone or clone instance that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
static void
create_notify_actions(pcmk_resource_t *rsc, notify_data_t *n_data)
{
GList *iter = NULL;
pcmk_action_t *stop = NULL;
pcmk_action_t *start = NULL;
enum pcmk__action_type task = pcmk__parse_action(n_data->action);
// If this is a clone, call recursively for each instance
if (rsc->private->children != NULL) {
g_list_foreach(rsc->private->children, (GFunc) create_notify_actions,
n_data);
return;
}
// Add notification meta-attributes to original actions
for (iter = rsc->private->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *op = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
switch (pcmk__parse_action(op->task)) {
case pcmk__action_start:
case pcmk__action_stop:
case pcmk__action_promote:
case pcmk__action_demote:
add_notify_data_to_action_meta(n_data, op);
break;
default:
break;
}
}
}
// Skip notify action itself if original action was not needed
switch (task) {
case pcmk__action_start:
if (n_data->start == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_promote:
if (n_data->promote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_demote:
if (n_data->demote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
default:
// We cannot do same for stop because it might be implied by fencing
break;
}
pcmk__rsc_trace(rsc, "Creating notify actions for %s %s",
rsc->id, n_data->action);
// Create notify actions for stop or demote
if ((rsc->private->orig_role != pcmk_role_stopped)
&& ((task == pcmk__action_stop) || (task == pcmk__action_demote))) {
stop = find_first_action(rsc->private->actions, NULL, PCMK_ACTION_STOP,
NULL);
for (iter = rsc->private->active_nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *current_node = (pcmk_node_t *) iter->data;
/* If a stop is a pseudo-action implied by fencing, don't try to
* notify the node getting fenced.
*/
if ((stop != NULL)
&& pcmk_is_set(stop->flags, pcmk__action_pseudo)
&& (current_node->details->unclean
|| pcmk_is_set(current_node->private->flags,
pcmk__node_remote_reset))) {
continue;
}
new_notify_action(rsc, current_node, n_data->pre,
n_data->pre_done, n_data);
if ((task == pcmk__action_demote) || (stop == NULL)
|| pcmk_is_set(stop->flags, pcmk__action_optional)) {
new_post_notify_action(rsc, current_node, n_data);
}
}
}
// Create notify actions for start or promote
if ((rsc->private->next_role != pcmk_role_stopped)
&& ((task == pcmk__action_start) || (task == pcmk__action_promote))) {
start = find_first_action(rsc->private->actions, NULL,
PCMK_ACTION_START, NULL);
if (start != NULL) {
pcmk_action_t *remote_start = find_remote_start(start);
if ((remote_start != NULL)
&& !pcmk_is_set(remote_start->flags, pcmk__action_runnable)) {
/* Start and promote actions for a clone instance behind
* a Pacemaker Remote connection happen after the
* connection starts. If the connection start is blocked, do
* not schedule notifications for these actions.
*/
return;
}
}
if (rsc->private->assigned_node == NULL) {
- pcmk__sched_err("Next role '%s' but %s is not allocated",
+ pcmk__sched_err(rsc->private->scheduler,
+ "Next role '%s' but %s is not allocated",
pcmk_role_text(rsc->private->next_role), rsc->id);
return;
}
if ((task != pcmk__action_start) || (start == NULL)
|| pcmk_is_set(start->flags, pcmk__action_optional)) {
new_notify_action(rsc, rsc->private->assigned_node, n_data->pre,
n_data->pre_done, n_data);
}
new_post_notify_action(rsc, rsc->private->assigned_node, n_data);
}
}
/*!
* \internal
* \brief Create notification data and actions for one clone action
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
void
pe__create_action_notifications(pcmk_resource_t *rsc, notify_data_t *n_data)
{
if ((rsc == NULL) || (n_data == NULL)) {
return;
}
collect_resource_data(rsc, true, n_data);
add_notif_keys(rsc, n_data);
create_notify_actions(rsc, n_data);
}
/*!
* \internal
* \brief Free notification data for one action
*
* \param[in,out] n_data Notification data to free
*/
void
pe__free_action_notification_data(notify_data_t *n_data)
{
if (n_data == NULL) {
return;
}
g_list_free_full(n_data->stop, free);
g_list_free_full(n_data->start, free);
g_list_free_full(n_data->demote, free);
g_list_free_full(n_data->promote, free);
g_list_free_full(n_data->promoted, free);
g_list_free_full(n_data->unpromoted, free);
g_list_free_full(n_data->active, free);
g_list_free_full(n_data->inactive, free);
pcmk_free_nvpairs(n_data->keys);
free(n_data);
}
/*!
* \internal
* \brief Order clone "notifications complete" pseudo-action after fencing
*
* If a stop action is implied by fencing, the usual notification pseudo-actions
* will not be sufficient to order things properly, or even create all needed
* notifications if the clone is also stopping on another node, and another
* clone is ordered after it. This function creates new notification
* pseudo-actions relative to the fencing to ensure everything works properly.
*
* \param[in] stop Stop action implied by fencing
* \param[in,out] rsc Clone resource that notification is for
* \param[in,out] stonith_op Fencing action that implies \p stop
*/
void
pe__order_notifs_after_fencing(const pcmk_action_t *stop, pcmk_resource_t *rsc,
pcmk_action_t *stonith_op)
{
notify_data_t *n_data;
crm_info("Ordering notifications for implied %s after fencing", stop->uuid);
n_data = pe__action_notif_pseudo_ops(rsc, PCMK_ACTION_STOP, NULL,
stonith_op);
if (n_data != NULL) {
collect_resource_data(rsc, false, n_data);
add_notify_env(n_data, "notify_stop_resource", rsc->id);
add_notify_env(n_data, "notify_stop_uname", stop->node->private->name);
create_notify_actions(uber_parent(rsc), n_data);
pe__free_action_notification_data(n_data);
}
}
diff --git a/lib/pengine/status.c b/lib/pengine/status.c
index 83e6365baf..4d6dbdcab5 100644
--- a/lib/pengine/status.c
+++ b/lib/pengine/status.c
@@ -1,523 +1,524 @@
/*
* 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 <sys/param.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/cib_internal.h>
#include <glib.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
/*!
* \brief Create a new object to hold scheduler data
*
* \return New, initialized scheduler data on success, else NULL (and set errno)
* \note Only pcmk_scheduler_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.
*/
pcmk_scheduler_t *
pe_new_working_set(void)
{
pcmk_scheduler_t *scheduler = calloc(1, sizeof(pcmk_scheduler_t));
if (scheduler != NULL) {
set_working_set_defaults(scheduler);
}
return scheduler;
}
/*!
* \brief Free scheduler data
*
* \param[in,out] scheduler Scheduler data to free
*/
void
pe_free_working_set(pcmk_scheduler_t *scheduler)
{
if (scheduler != NULL) {
pe_reset_working_set(scheduler);
scheduler->priv = NULL;
free(scheduler);
}
}
#define XPATH_DEPRECATED_RULES \
"//" PCMK_XE_OP_DEFAULTS "//" PCMK_XE_EXPRESSION \
"|//" PCMK_XE_OP "//" PCMK_XE_EXPRESSION
/*!
* \internal
* \brief Log a warning for deprecated rule syntax in operations
*
* \param[in] scheduler Scheduler data
*/
static void
check_for_deprecated_rules(pcmk_scheduler_t *scheduler)
{
// @COMPAT Drop this function when support for the syntax is dropped
xmlNode *deprecated = get_xpath_object(XPATH_DEPRECATED_RULES,
scheduler->input, LOG_NEVER);
if (deprecated != NULL) {
pcmk__warn_once(pcmk__wo_op_attr_expr,
"Support for rules with node attribute expressions in "
PCMK_XE_OP " or " PCMK_XE_OP_DEFAULTS " is deprecated "
"and will be dropped in a future release");
}
}
/*
* 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(pcmk_scheduler_t * scheduler)
{
const char *new_version = NULL;
xmlNode *section = NULL;
if ((scheduler == NULL) || (scheduler->input == NULL)) {
return FALSE;
}
new_version = crm_element_value(scheduler->input, PCMK_XA_CRM_FEATURE_SET);
if (pcmk__check_feature_set(new_version) != pcmk_rc_ok) {
pcmk__config_err("Can't process CIB with feature set '%s' greater than our own '%s'",
new_version, CRM_FEATURE_SET);
return FALSE;
}
crm_trace("Beginning unpack");
if (scheduler->failed != NULL) {
pcmk__xml_free(scheduler->failed);
}
scheduler->failed = pcmk__xe_create(NULL, "failed-ops");
if (scheduler->now == NULL) {
scheduler->now = crm_time_new(NULL);
}
if (scheduler->dc_uuid == NULL) {
scheduler->dc_uuid = crm_element_value_copy(scheduler->input,
PCMK_XA_DC_UUID);
}
if (pcmk__xe_attr_is_true(scheduler->input, PCMK_XA_HAVE_QUORUM)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_quorate);
} else {
pcmk__clear_scheduler_flags(scheduler, pcmk_sched_quorate);
}
scheduler->op_defaults = get_xpath_object("//" PCMK_XE_OP_DEFAULTS,
scheduler->input, LOG_NEVER);
check_for_deprecated_rules(scheduler);
scheduler->rsc_defaults = get_xpath_object("//" PCMK_XE_RSC_DEFAULTS,
scheduler->input, LOG_NEVER);
section = get_xpath_object("//" PCMK_XE_CRM_CONFIG, scheduler->input,
LOG_TRACE);
unpack_config(section, scheduler);
if (!pcmk_any_flags_set(scheduler->flags,
pcmk_sched_location_only|pcmk_sched_quorate)
&& (scheduler->no_quorum_policy != pcmk_no_quorum_ignore)) {
- pcmk__sched_warn("Fencing and resource management disabled "
+ pcmk__sched_warn(scheduler,
+ "Fencing and resource management disabled "
"due to lack of quorum");
}
section = get_xpath_object("//" PCMK_XE_NODES, scheduler->input, LOG_TRACE);
unpack_nodes(section, scheduler);
section = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input,
LOG_TRACE);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
unpack_remote_nodes(section, scheduler);
}
unpack_resources(section, scheduler);
section = get_xpath_object("//" PCMK_XE_TAGS, scheduler->input, LOG_NEVER);
unpack_tags(section, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
section = get_xpath_object("//" PCMK_XE_STATUS, scheduler->input,
LOG_TRACE);
unpack_status(section, scheduler);
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_no_counts)) {
for (GList *item = scheduler->resources; item != NULL;
item = item->next) {
pcmk_resource_t *rsc = item->data;
rsc->private->fns->count(item->data);
}
crm_trace("Cluster resource count: %d (%d disabled, %d blocked)",
scheduler->ninstances, scheduler->disabled_resources,
scheduler->blocked_resources);
}
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_status);
return TRUE;
}
/*!
* \internal
* \brief Free a list of pcmk_resource_t
*
* \param[in,out] resources List to free
*
* \note When the scheduler's resource list is freed, that includes the original
* storage for the uname and id of any Pacemaker Remote nodes in the
* scheduler's node list, so take care not to use those afterward.
* \todo Refactor pcmk_node_t to strdup() the node name.
*/
static void
pe_free_resources(GList *resources)
{
pcmk_resource_t *rsc = NULL;
GList *iterator = resources;
while (iterator != NULL) {
rsc = (pcmk_resource_t *) iterator->data;
iterator = iterator->next;
rsc->private->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) {
pcmk_node_t *node = (pcmk_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->private->name for Pacemaker Remote nodes.
*/
crm_trace("Freeing node %s", (pcmk__is_pacemaker_remote_node(node)?
"(guest or remote)" : pcmk__node_name(node)));
if (node->private->attrs != NULL) {
g_hash_table_destroy(node->private->attrs);
}
if (node->private->utilization != NULL) {
g_hash_table_destroy(node->private->utilization);
}
if (node->private->digest_cache != NULL) {
g_hash_table_destroy(node->private->digest_cache);
}
g_list_free(node->details->running_rsc);
g_list_free(node->private->assigned_resources);
free(node->private);
free(node->details);
free(node->assign);
free(node);
}
if (nodes != NULL) {
g_list_free(nodes);
}
}
static void
pe__free_ordering(GList *constraints)
{
GList *iterator = constraints;
while (iterator != NULL) {
pcmk__action_relation_t *order = iterator->data;
iterator = iterator->next;
free(order->task1);
free(order->task2);
free(order);
}
if (constraints != NULL) {
g_list_free(constraints);
}
}
static void
pe__free_location(GList *constraints)
{
GList *iterator = constraints;
while (iterator != NULL) {
pcmk__location_t *cons = iterator->data;
iterator = iterator->next;
g_list_free_full(cons->nodes, free);
free(cons->id);
free(cons);
}
if (constraints != NULL) {
g_list_free(constraints);
}
}
/*!
* \brief Reset scheduler data to defaults without freeing it or constraints
*
* \param[in,out] scheduler Scheduler data to reset
*
* \deprecated This function is deprecated as part of the API;
* pe_reset_working_set() should be used instead.
*/
void
cleanup_calculations(pcmk_scheduler_t *scheduler)
{
if (scheduler == NULL) {
return;
}
pcmk__clear_scheduler_flags(scheduler, pcmk_sched_have_status);
if (scheduler->config_hash != NULL) {
g_hash_table_destroy(scheduler->config_hash);
}
if (scheduler->singletons != NULL) {
g_hash_table_destroy(scheduler->singletons);
}
if (scheduler->tickets) {
g_hash_table_destroy(scheduler->tickets);
}
if (scheduler->template_rsc_sets) {
g_hash_table_destroy(scheduler->template_rsc_sets);
}
if (scheduler->tags) {
g_hash_table_destroy(scheduler->tags);
}
free(scheduler->dc_uuid);
crm_trace("deleting resources");
pe_free_resources(scheduler->resources);
crm_trace("deleting actions");
pe_free_actions(scheduler->actions);
crm_trace("deleting nodes");
pe_free_nodes(scheduler->nodes);
pe__free_param_checks(scheduler);
g_list_free(scheduler->stop_needed);
pcmk__xml_free(scheduler->graph);
crm_time_free(scheduler->now);
pcmk__xml_free(scheduler->input);
pcmk__xml_free(scheduler->failed);
set_working_set_defaults(scheduler);
CRM_CHECK(scheduler->ordering_constraints == NULL,;
);
CRM_CHECK(scheduler->placement_constraints == NULL,;
);
}
/*!
* \brief Reset scheduler data to default state without freeing it
*
* \param[in,out] scheduler Scheduler data to reset
*/
void
pe_reset_working_set(pcmk_scheduler_t *scheduler)
{
if (scheduler == NULL) {
return;
}
crm_trace("Deleting %d ordering constraints",
g_list_length(scheduler->ordering_constraints));
pe__free_ordering(scheduler->ordering_constraints);
scheduler->ordering_constraints = NULL;
crm_trace("Deleting %d location constraints",
g_list_length(scheduler->placement_constraints));
pe__free_location(scheduler->placement_constraints);
scheduler->placement_constraints = NULL;
crm_trace("Deleting %d colocation constraints",
g_list_length(scheduler->colocation_constraints));
g_list_free_full(scheduler->colocation_constraints, free);
scheduler->colocation_constraints = NULL;
crm_trace("Deleting %d ticket constraints",
g_list_length(scheduler->ticket_constraints));
g_list_free_full(scheduler->ticket_constraints, free);
scheduler->ticket_constraints = NULL;
cleanup_calculations(scheduler);
}
void
set_working_set_defaults(pcmk_scheduler_t *scheduler)
{
void *priv = scheduler->priv;
memset(scheduler, 0, sizeof(pcmk_scheduler_t));
scheduler->priv = priv;
scheduler->order_id = 1;
scheduler->action_id = 1;
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
scheduler->flags = 0x0ULL;
pcmk__set_scheduler_flags(scheduler,
pcmk_sched_symmetric_cluster
|pcmk_sched_stop_removed_resources
|pcmk_sched_cancel_removed_actions);
if (!strcmp(PCMK__CONCURRENT_FENCING_DEFAULT, PCMK_VALUE_TRUE)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_concurrent_fencing);
}
}
pcmk_resource_t *
pe_find_resource(GList *rsc_list, const char *id)
{
return pe_find_resource_with_flags(rsc_list, id, pcmk_rsc_match_history);
}
pcmk_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) {
pcmk_resource_t *parent = rIter->data;
pcmk_resource_t *match = parent->private->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 pcmk_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
*/
pcmk_node_t *
pe_find_node_any(const GList *nodes, const char *id, const char *uname)
{
pcmk_node_t *match = NULL;
if (id != NULL) {
match = pe_find_node_id(nodes, id);
}
if ((match == NULL) && (uname != NULL)) {
match = pcmk__find_node_in_list(nodes, uname);
}
return match;
}
/*!
* \brief Find a node by ID in a list of nodes
*
* \param[in] nodes List of nodes (as pcmk_node_t*)
* \param[in] id ID of node to find
*
* \return Node from \p nodes that matches \p id if any, otherwise NULL
*/
pcmk_node_t *
pe_find_node_id(const GList *nodes, const char *id)
{
for (const GList *iter = nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_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->private->id, id, pcmk__str_casei)) {
return node;
}
}
return NULL;
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/pengine/status_compat.h>
/*!
* \brief Find a node by name in a list of nodes
*
* \param[in] nodes List of nodes (as pcmk_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
*/
pcmk_node_t *
pe_find_node(const GList *nodes, const char *node_name)
{
return pcmk__find_node_in_list(nodes, node_name);
}
// LCOV_EXCL_STOP
// End deprecated API
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index 254c905328..daa88449cb 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5080 +1,5087 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
// A (parsed) resource action history entry
struct action_history {
pcmk_resource_t *rsc; // Resource that history is for
pcmk_node_t *node; // Node that history is for
xmlNode *xml; // History entry XML
// Parsed from entry XML
const char *id; // XML ID of history entry
const char *key; // Operation key of action
const char *task; // Action name
const char *exit_reason; // Exit reason given for result
guint interval_ms; // Action interval
int call_id; // Call ID of action
int expected_exit_status; // Expected exit status of action
int exit_status; // Actual exit status of action
int execution_status; // Execution status of action
};
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(scheduler, option, flag) do { \
GHashTable *config_hash = (scheduler)->config_hash; \
const char *scf_value = pcmk__cluster_option(config_hash, (option)); \
\
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} else { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node,
xmlNode *xml_op, xmlNode **last_failure,
enum pcmk__on_fail *failed);
static void determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node);
static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node,
bool overwrite, pcmk_scheduler_t *scheduler);
static void determine_online_status(const xmlNode *node_state,
pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler);
static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler);
/*!
* \internal
* \brief Check whether a node is a dangling guest node
*
* \param[in] node Node to check
*
* \return true if \p node had a Pacemaker Remote connection resource with a
* launcher that was removed from the CIB, otherwise false.
*/
static bool
is_dangling_guest_node(pcmk_node_t *node)
{
return pcmk__is_pacemaker_remote_node(node)
&& (node->private->remote != NULL)
&& (node->private->remote->private->launcher == NULL)
&& pcmk_is_set(node->private->remote->flags,
pcmk__rsc_removed_launched);
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider
* \c PCMK_OPT_PRIORITY_FENCING_DELAY
*/
void
pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node,
const char *reason, bool priority_delay)
{
CRM_CHECK(node, return);
if (pcmk__is_guest_or_bundle_node(node)) {
// Fence a guest or bundle node by marking its launcher as failed
pcmk_resource_t *rsc = node->private->remote->private->launcher;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing guest node %s "
"(otherwise would because %s): "
"its guest resource %s is unmanaged",
pcmk__node_name(node), reason, rsc->id);
} else {
- pcmk__sched_warn("Guest node %s will be fenced "
+ pcmk__sched_warn(scheduler,
+ "Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pcmk__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.
*/
pcmk__set_node_flags(node, pcmk__node_remote_reset);
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
}
} else if (is_dangling_guest_node(node)) {
crm_info("Cleaning up dangling connection for guest node %s: "
"fencing was already done because %s, "
"and guest resource no longer exists",
pcmk__node_name(node), reason);
pcmk__set_rsc_flags(node->private->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
} else if (pcmk__is_remote_node(node)) {
pcmk_resource_t *rsc = node->private->remote;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing remote node %s "
"(otherwise would because %s): connection is unmanaged",
pcmk__node_name(node), reason);
} else if (!pcmk_is_set(node->private->flags, pcmk__node_remote_reset)) {
pcmk__set_node_flags(node, pcmk__node_remote_reset);
- pcmk__sched_warn("Remote node %s %s: %s",
+ pcmk__sched_warn(scheduler, "Remote node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
- pcmk__sched_warn("Cluster node %s %s: %s",
+ pcmk__sched_warn(scheduler, "Cluster node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \
"[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \
"or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \
"and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \
"/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(scheduler->flags, flag)) {
result = xpath_search(scheduler->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pcmk__set_scheduler_flags(scheduler, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
guint interval_ms = 0U;
GHashTable *config_hash = pcmk__strkey_table(free, free);
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
scheduler->config_hash = config_hash;
pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_data,
config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS,
FALSE, scheduler);
pcmk__validate_cluster_options(config_hash);
set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES,
pcmk_sched_probe_resources);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" is nonzero");
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING,
scheduler);
value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT);
pcmk_parse_interval_spec(value, &interval_ms);
if (interval_ms >= INT_MAX) {
scheduler->stonith_timeout = INT_MAX;
} else {
scheduler->stonith_timeout = (int) interval_ms;
}
crm_debug("STONITH timeout: %d", scheduler->stonith_timeout);
set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED,
pcmk_sched_fencing_enabled);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
crm_debug("STONITH of failed nodes is enabled");
} else {
crm_debug("STONITH of failed nodes is disabled");
}
scheduler->stonith_action = pcmk__cluster_option(config_hash,
PCMK_OPT_STONITH_ACTION);
if (!strcmp(scheduler->stonith_action, PCMK__ACTION_POWEROFF)) {
pcmk__warn_once(pcmk__wo_poweroff,
"Support for " PCMK_OPT_STONITH_ACTION " of "
"'" PCMK__ACTION_POWEROFF "' is deprecated and will be "
"removed in a future release "
"(use '" PCMK_ACTION_OFF "' instead)");
scheduler->stonith_action = PCMK_ACTION_OFF;
}
crm_trace("STONITH will %s nodes", scheduler->stonith_action);
set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING,
pcmk_sched_concurrent_fencing);
if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) {
crm_debug("Concurrent fencing is enabled");
} else {
crm_debug("Concurrent fencing is disabled");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY);
if (value) {
pcmk_parse_interval_spec(value, &interval_ms);
scheduler->priority_fencing_delay = (int) (interval_ms / 1000);
crm_trace("Priority fencing delay is %ds",
scheduler->priority_fencing_delay);
}
set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES,
pcmk_sched_stop_all);
crm_debug("Stop all active resources: %s",
pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all));
set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER,
pcmk_sched_symmetric_cluster);
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY);
if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei)) {
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
int do_panic = 0;
crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC,
&do_panic);
if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
scheduler->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop': cluster has never had quorum");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop' because fencing is disabled");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
crm_debug("On loss of quorum: Freeze resources");
break;
case pcmk_no_quorum_stop:
crm_debug("On loss of quorum: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case pcmk_no_quorum_fence:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case pcmk_no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES,
pcmk_sched_stop_removed_resources);
if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
crm_trace("Orphan resources are stopped");
} else {
crm_trace("Orphan resources are ignored");
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS,
pcmk_sched_cancel_removed_actions);
if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) {
crm_trace("Orphan resource actions are stopped");
} else {
crm_trace("Orphan resource actions are ignored");
}
value = pcmk__cluster_option(config_hash, PCMK__OPT_REMOVE_AFTER_STOP);
if (value != NULL) {
if (crm_is_true(value)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop);
pcmk__warn_once(pcmk__wo_remove_after,
"Support for the " PCMK__OPT_REMOVE_AFTER_STOP
" cluster property is deprecated and will be "
"removed in a future release");
} else {
pcmk__clear_scheduler_flags(scheduler,
pcmk_sched_remove_after_stop);
}
}
set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE,
pcmk_sched_in_maintenance);
crm_trace("Maintenance mode: %s",
pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk_sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk_sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->placement_strategy =
pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY);
crm_trace("Placement strategy: %s", scheduler->placement_strategy);
set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK,
pcmk_sched_shutdown_lock);
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT);
pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock));
scheduler->shutdown_lock /= 1000;
crm_trace("Resources will be locked to nodes that were cleanly "
"shut down (locks expire after %s)",
pcmk__readable_interval(scheduler->shutdown_lock));
} else {
crm_trace("Resources will not be locked to nodes that were cleanly "
"shut down");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout));
scheduler->node_pending_timeout /= 1000;
if (scheduler->node_pending_timeout == 0) {
crm_trace("Do not fence pending nodes");
} else {
crm_trace("Fence pending nodes after %s",
pcmk__readable_interval(scheduler->node_pending_timeout
* 1000));
}
return TRUE;
}
pcmk_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pcmk_scheduler_t *scheduler)
{
pcmk_node_t *new_node = NULL;
if (pcmk_find_node(scheduler, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
new_node = calloc(1, sizeof(pcmk_node_t));
if (new_node == NULL) {
- pcmk__sched_err("Could not allocate memory for node %s", uname);
+ pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
+ uname);
return NULL;
}
new_node->assign = calloc(1, sizeof(struct pcmk__node_assignment));
new_node->details = calloc(1, sizeof(struct pcmk__node_details));
new_node->private = calloc(1, sizeof(pcmk__node_private_t));
if ((new_node->assign == NULL) || (new_node->details == NULL)
|| (new_node->private == NULL)) {
free(new_node->assign);
free(new_node->details);
free(new_node->private);
free(new_node);
- pcmk__sched_err("Could not allocate memory for node %s", uname);
+ pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
+ uname);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->assign->score = char2score(score);
new_node->private->id = id;
new_node->private->name = uname;
new_node->private->flags = pcmk__node_probes_allowed;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->running_rsc = NULL;
new_node->private->scheduler = scheduler;
if (pcmk__str_eq(type, PCMK_VALUE_MEMBER,
pcmk__str_null_matches|pcmk__str_casei)) {
new_node->private->variant = pcmk__node_variant_cluster;
} else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) {
new_node->private->variant = pcmk__node_variant_remote;
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes);
} else {
/* @COMPAT 'ping' is the default for backward compatibility, but it
* should be changed to 'member' at a compatibility break
*/
if (!pcmk__str_eq(type, PCMK__VALUE_PING, pcmk__str_casei)) {
pcmk__config_warn("Node %s has unrecognized type '%s', "
"assuming '" PCMK__VALUE_PING "'",
pcmk__s(uname, "without name"), type);
}
pcmk__warn_once(pcmk__wo_ping_node,
"Support for nodes of type '" PCMK__VALUE_PING "' "
"(such as %s) is deprecated and will be removed in a "
"future release",
pcmk__s(uname, "unnamed node"));
new_node->private->variant = pcmk__node_variant_ping;
}
new_node->private->attrs = pcmk__strkey_table(free, free);
if (pcmk__is_pacemaker_remote_node(new_node)) {
pcmk__insert_dup(new_node->private->attrs, CRM_ATTR_KIND, "remote");
} else {
pcmk__insert_dup(new_node->private->attrs, CRM_ATTR_KIND, "cluster");
}
new_node->private->utilization = pcmk__strkey_table(free, free);
new_node->private->digest_cache = pcmk__strkey_table(free,
pe__free_digests);
scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = pcmk__xe_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, NULL, NULL, NULL);
attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) {
if (!pcmk__xe_is(attr_set, PCMK_XE_META_ATTRIBUTES)) {
continue;
}
for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL);
attr != NULL; attr = pcmk__xe_next(attr)) {
const char *value = crm_element_value(attr, PCMK_XA_VALUE);
const char *name = crm_element_value(attr, PCMK_XA_NAME);
if (name == NULL) { // Sanity
continue;
}
if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) {
remote_name = value;
} else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) {
remote_server = value;
} else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) {
remote_port = value;
} else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) {
connect_timeout = value;
} else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) {
remote_allow_migrate = value;
} else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node)
{
if ((new_node->private->variant == pcmk__node_variant_remote)
&& (new_node->private->remote == NULL)) {
/* Ignore fencing for remote nodes that don't have a connection resource
* associated with them. This happens when remote node entries get left
* in the nodes section after the connection resource is removed.
*/
return;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
// All nodes are unclean until we've seen their status entry
new_node->details->unclean = TRUE;
} else {
// Blind faith ...
new_node->details->unclean = FALSE;
}
}
gboolean
unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
pcmk_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
const char *score = NULL;
for (xml_obj = pcmk__xe_first_child(xml_nodes, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) {
if (pcmk__xe_is(xml_obj, PCMK_XE_NODE)) {
new_node = NULL;
id = crm_element_value(xml_obj, PCMK_XA_ID);
uname = crm_element_value(xml_obj, PCMK_XA_UNAME);
type = crm_element_value(xml_obj, PCMK_XA_TYPE);
score = crm_element_value(xml_obj, PCMK_XA_SCORE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" PCMK_XE_NODE
"> entry in configuration without id");
continue;
}
new_node = pe_create_node(id, uname, type, score, scheduler);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(scheduler, new_node);
add_node_attrs(xml_obj, new_node, FALSE, scheduler);
crm_trace("Done with node %s",
crm_element_value(xml_obj, PCMK_XA_UNAME));
}
}
if (scheduler->localhost
&& (pcmk_find_node(scheduler, scheduler->localhost) == NULL)) {
crm_info("Creating a fake local node");
pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0,
scheduler);
}
return TRUE;
}
static void
unpack_launcher(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *launcher_id = NULL;
if (rsc->private->children != NULL) {
g_list_foreach(rsc->private->children, (GFunc) unpack_launcher,
scheduler);
return;
}
launcher_id = g_hash_table_lookup(rsc->private->meta, PCMK__META_CONTAINER);
if ((launcher_id != NULL)
&& !pcmk__str_eq(launcher_id, rsc->id, pcmk__str_none)) {
pcmk_resource_t *launcher = pe_find_resource(scheduler->resources,
launcher_id);
if (launcher != NULL) {
rsc->private->launcher = launcher;
launcher->private->launched =
g_list_append(launcher->private->launched, rsc);
pcmk__rsc_trace(rsc, "Resource %s's launcher is %s",
rsc->id, launcher_id);
} else {
pcmk__config_err("Resource %s: Unknown " PCMK__META_CONTAINER " %s",
rsc->id, launcher_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
/* Create remote nodes and guest nodes from the resource configuration
* before unpacking resources.
*/
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
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 = pcmk__xe_id(xml_obj);
/* The pcmk_find_node() check ensures we don't iterate over an
* expanded node that has already been added to the node list
*/
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
NULL, scheduler);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) {
/* This will add an ocf:pacemaker:remote primitive to the
* configuration for the guest node's connection, to be unpacked
* later.
*/
new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
NULL, scheduler);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL);
xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, pcmk__xe_id(xml_obj2),
pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
NULL, scheduler);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc)
{
pcmk_node_t *remote_node = NULL;
if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) {
return;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pcmk_find_node(scheduler, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s",
new_rsc->id, pcmk__node_name(remote_node));
remote_node->private->remote = new_rsc;
if (new_rsc->private->launcher == NULL) {
/* Handle start-up fencing for remote nodes (as opposed to guest nodes)
* the same as is done for cluster nodes.
*/
handle_startup_fencing(scheduler, remote_node);
} else {
/* pe_create_node() marks the new node as "remote" or "cluster"; now
* that we know the node is a guest node, update it correctly.
*/
pcmk__insert_dup(remote_node->private->attrs,
CRM_ATTR_KIND, "container");
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] scheduler Scheduler data
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when pe__unpack_resource() calls resource_location()
*/
gboolean
unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
scheduler->template_rsc_sets = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) {
pcmk_resource_t *new_rsc = NULL;
const char *id = pcmk__xe_id(xml_obj);
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring <%s> resource without ID",
xml_obj->name);
continue;
}
if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) {
if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
pcmk__insert_dup(scheduler->template_rsc_sets, id, NULL);
}
continue;
}
crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id);
if (pe__unpack_resource(xml_obj, &new_rsc, NULL,
scheduler) == pcmk_rc_ok) {
scheduler->resources = g_list_append(scheduler->resources, new_rsc);
pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
xml_obj->name, id);
}
}
for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
unpack_launcher(rsc, scheduler);
link_rsc2remotenode(scheduler, rsc);
}
scheduler->resources = g_list_sort(scheduler->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
/* Ignore */
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) {
pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined");
pcmk__config_err("Either configure some or disable STONITH with the "
PCMK_OPT_STONITH_ENABLED " option");
pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity");
}
return TRUE;
}
gboolean
unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_tag = NULL;
scheduler->tags = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_tag = pcmk__xe_first_child(xml_tags, NULL, NULL, NULL);
xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = pcmk__xe_id(xml_tag);
if (!pcmk__xe_is(xml_tag, PCMK_XE_TAG)) {
continue;
}
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID,
(const char *) xml_tag->name);
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag, NULL, NULL, NULL);
xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) {
const char *obj_ref = pcmk__xe_id(xml_obj_ref);
if (!pcmk__xe_is(xml_obj_ref, PCMK_XE_OBJ_REF)) {
continue;
}
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID,
xml_obj_ref->name, tag_id);
continue;
}
pcmk__add_idref(scheduler->tags, tag_id, obj_ref);
}
}
return TRUE;
}
/* The ticket state section:
* "/cib/status/tickets/ticket_state" */
static gboolean
unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler)
{
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pcmk__ticket_t *ticket = NULL;
ticket_id = pcmk__xe_id(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return FALSE;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(scheduler->tickets, ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, scheduler);
if (ticket == NULL) {
return FALSE;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) {
continue;
}
pcmk__insert_dup(ticket->state, prop_name, prop_value);
}
granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED);
if (granted && crm_is_true(granted)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We have ticket '%s'", ticket->id);
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_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, PCMK_XA_STANDBY);
if (standby && crm_is_true(standby)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_standby);
if (pcmk_is_set(ticket->flags, pcmk__ticket_granted)) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_standby);
}
crm_trace("Done with ticket state for %s", ticket_id);
return TRUE;
}
static gboolean
unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
for (xml_obj = pcmk__xe_first_child(xml_tickets, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) {
if (!pcmk__xe_is(xml_obj, PCMK__XE_TICKET_STATE)) {
continue;
}
unpack_ticket_state(xml_obj, scheduler);
}
return TRUE;
}
static void
unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = NULL;
pcmk_resource_t *rsc = NULL;
int maint = 0;
if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
return;
}
if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) {
return;
}
crm_trace("Processing Pacemaker Remote node %s",
pcmk__node_name(this_node));
pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE),
&maint, 0);
if (maint) {
pcmk__set_node_flags(this_node, pcmk__node_remote_maint);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_maint);
}
rsc = this_node->private->remote;
if (!pcmk_is_set(this_node->private->flags, pcmk__node_remote_reset)) {
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
}
attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL,
NULL);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
}
if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL,
pcmk__rsc_node_current))
|| ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) {
crm_info("%s is in maintenance mode", pcmk__node_name(this_node));
this_node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(this_node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__warn_once(pcmk__wo_rdisc_enabled,
"Support for the "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" node attribute is deprecated and will be removed"
" (and behave as 'true') in a future release.");
if (pcmk__is_remote_node(this_node)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute on Pacemaker Remote node %s"
" because fencing is disabled",
pcmk__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",
pcmk__node_name(this_node));
pcmk__clear_node_flags(this_node, pcmk__node_probes_allowed);
}
}
}
/*!
* \internal
* \brief Unpack a cluster node's transient attributes
*
* \param[in] state CIB node state XML
* \param[in,out] node Cluster node whose attributes are being unpacked
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = pcmk__xe_first_child(state,
PCMK__XE_TRANSIENT_ATTRIBUTES,
NULL, NULL);
add_node_attrs(attrs, node, TRUE, scheduler);
if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(node));
pcmk__set_node_flags(node, pcmk__node_standby);
}
if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in maintenance mode", pcmk__node_name(node));
node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute for %s because disabling resource"
" discovery is not allowed for cluster nodes",
pcmk__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not
* the resource history inside it. Multiple passes through the status are needed
* to fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = crm_element_value(state, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without "
PCMK_XA_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = crm_element_value(state, PCMK_XA_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 PCMK__XE_NODE_STATE entry doesn't have
* a PCMK_XA_UNAME yet. We should recognize the node as `pending` and
* wait for it to join CPG.
*/
crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" "
"without " PCMK_XA_UNAME,
id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
crm_notice("Ignoring recorded state for removed node with name %s and "
PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id);
return;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
int remote_fenced = 0;
/* 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, PCMK__XA_NODE_FENCED),
&remote_fenced, 0);
if (remote_fenced) {
pcmk__set_node_flags(this_node, pcmk__node_remote_fenced);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_fenced);
}
return;
}
unpack_transient_attributes(state, this_node, scheduler);
/* Provisionally mark this cluster node as clean. We have at least seen it
* in the current cluster's lifetime.
*/
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
crm_trace("Determining online status of cluster node %s (id %s)",
pcmk__node_name(this_node), id);
determine_online_status(state, this_node, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)
&& this_node->details->online
&& (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(scheduler, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(const xmlNode *status, bool fence,
pcmk_scheduler_t *scheduler)
{
int rc = pcmk_rc_ok;
// Loop through all PCMK__XE_NODE_STATE entries in CIB status
for (const xmlNode *state = pcmk__xe_first_child(status,
PCMK__XE_NODE_STATE, NULL,
NULL);
state != NULL; state = pcmk__xe_next_same(state)) {
const char *id = pcmk__xe_id(state);
const char *uname = crm_element_value(state, PCMK_XA_UNAME);
pcmk_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
PCMK__XE_NODE_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (pcmk_is_set(this_node->private->flags, pcmk__node_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 (pcmk__is_guest_or_bundle_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.
*/
const pcmk_resource_t *remote = this_node->private->remote;
const pcmk_resource_t *launcher = remote->private->launcher;
if ((remote->private->orig_role != pcmk_role_started)
|| (launcher->private->orig_role != pcmk_role_started)) {
crm_trace("Not unpacking resource history for guest node %s "
"because launcher and connection are not known to "
"be up", id);
continue;
}
} else if (pcmk__is_remote_node(this_node)) {
/* We can unpack a remote node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection is up, with the exception of when shutdown locks are
* in use.
*/
pcmk_resource_t *rsc = this_node->private->remote;
if ((rsc == NULL)
|| (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)
&& (rsc->private->orig_role != pcmk_role_started))) {
crm_trace("Not unpacking resource history for remote node %s "
"because connection is not known to be up", id);
continue;
}
/* If fencing and shutdown locks are disabled and we're not processing
* unseen nodes, then we don't want to unpack offline nodes until online
* nodes have been unpacked. This allows us to number active clone
* instances first.
*/
} else if (!pcmk_any_flags_set(scheduler->flags,
pcmk_sched_fencing_enabled
|pcmk_sched_shutdown_lock)
&& !this_node->details->online) {
crm_trace("Not unpacking resource history for offline "
"cluster node %s", id);
continue;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
unpack_handle_remote_attrs(this_node, state, scheduler);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
pcmk__set_node_flags(this_node, pcmk__node_unpacked);
unpack_node_lrm(this_node, state, scheduler);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (scheduler->tickets == NULL) {
scheduler->tickets = pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL;
state = pcmk__xe_next(state)) {
if (pcmk__xe_is(state, PCMK_XE_TICKETS)) {
unpack_tickets_state((xmlNode *) state, scheduler);
} else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
unpack_node_state(state, scheduler);
}
}
while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(scheduler->flags,
pcmk_sched_fencing_enabled),
scheduler);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (scheduler->stop_needed != NULL) {
for (GList *item = scheduler->stop_needed; item; item = item->next) {
pcmk_resource_t *container = item->data;
pcmk_node_t *node = pcmk__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(scheduler->stop_needed);
scheduler->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *this_node = gIter->data;
if (!pcmk__is_pacemaker_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->private->remote != NULL)) {
pe__set_next_role(this_node->private->remote, pcmk_role_stopped,
"remote shutdown");
}
if (!pcmk_is_set(this_node->private->flags, pcmk__node_unpacked)) {
determine_remote_online_status(scheduler, this_node);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack node's time when it became a member at the cluster layer
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM);
int member = 0;
if (member_time == NULL) {
return -1LL;
} else if (crm_str_to_boolean(member_time, &member) == 1) {
/* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was
* recorded as a boolean for a DC < 2.1.7, or the node is pending
* shutdown and has left the CPG, in which case it was set to 1 to avoid
* fencing for PCMK_OPT_NODE_PENDING_TIMEOUT.
*
* We return the effective time for in_ccm=1 because what's important to
* avoid fencing is that effective time minus this value is less than
* the pending node timeout.
*/
return member? (long long) get_effective_time(scheduler) : 0LL;
} else {
long long when_member = 0LL;
if ((pcmk__scan_ll(member_time, &when_member,
0LL) != pcmk_rc_ok) || (when_member < 0LL)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM
" in " PCMK__XE_NODE_STATE " entry", member_time);
return -1LL;
}
return when_member;
}
}
/*!
* \internal
* \brief Unpack node's time when it became online in process group
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD);
// @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline"
if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE,
pcmk__str_casei|pcmk__str_null_matches)) {
return 0LL;
} else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) {
return 1LL;
} else {
long long when_online = 0LL;
if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok)
|| (when_online < 0)) {
crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in "
PCMK__XE_NODE_STATE " entry, assuming offline", peer_time);
return 0LL;
}
return when_online;
}
}
/*!
* \internal
* \brief Unpack node attribute for user-requested fencing
*
* \param[in] node Node to check
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
long long value = 0LL;
int value_i = 0;
const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE,
NULL, pcmk__rsc_node_current);
// Value may be boolean or an epoch time
if (crm_str_to_boolean(value_s, &value_i) == 1) {
return (value_i != 0);
}
if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) {
return (value > 0);
}
crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE
"node attribute for %s", value_s, pcmk__node_name(node));
return false;
}
static gboolean
determine_online_status_no_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
if (when_member <= 0) {
crm_trace("Node %s is %sdown", pcmk__node_name(this_node),
((when_member < 0)? "presumed " : ""));
} else if (when_online > 0) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node %s is not ready to run resources: %s",
pcmk__node_name(this_node), join);
}
} else if (!pcmk_is_set(this_node->private->flags,
pcmk__node_expected_up)) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE);
crm_info("Node %s member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->node_pending_timeout > 0)
&& (when_member > 0) && (when_online <= 0)) {
// There is a timeout on pending nodes, and node is pending
time_t timeout = when_member + scheduler->node_pending_timeout;
if (get_effective_time(node->private->scheduler) >= timeout) {
return true; // Node has timed out
}
// Node is pending, but still has time
pe__update_recheck_time(timeout, scheduler, "pending node timeout");
}
return false;
}
static bool
determine_online_status_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
bool termination_requested = unpack_node_terminate(this_node, node_state);
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
/*
- PCMK__XA_JOIN ::= member|down|pending|banned
- PCMK_XA_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- PCMK__XA_IN_CCM ::= true|false
- PCMK_XA_CRMD ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- PCMK__XA_IN_CCM ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- PCMK_XA_CRMD ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"),
(termination_requested? " (termination requested)" : ""));
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", pcmk__node_name(this_node));
/* Slightly different criteria since we can't shut down a dead peer */
return (when_online > 0);
}
if (when_member < 0) {
pe_fence_node(scheduler, this_node,
"peer has not been seen by the cluster", FALSE);
return false;
}
if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) {
pe_fence_node(scheduler, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (termination_requested) {
if ((when_member <= 0) && (when_online <= 0)
&& pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) {
crm_info("%s was fenced as requested", pcmk__node_name(this_node));
return false;
}
pe_fence_node(scheduler, this_node, "fencing was requested", false);
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN,
pcmk__str_null_matches)) {
if (pending_too_long(scheduler, this_node, when_member, when_online)) {
pe_fence_node(scheduler, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if ((when_member > 0) || (when_online > 0)) {
crm_info("- %s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pcmk__node_name(this_node));
}
} else if (when_member <= 0) {
// Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes
pe_fence_node(scheduler, this_node,
"peer is no longer part of the cluster", TRUE);
} else if (when_online <= 0) {
pe_fence_node(scheduler, this_node,
"peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) {
crm_info("%s is active", pcmk__node_name(this_node));
} else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING,
CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("%s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
pe_fence_node(scheduler, this_node, "peer was in an unknown state",
FALSE);
}
return (when_member > 0);
}
static void
determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node)
{
pcmk_resource_t *rsc = this_node->private->remote;
pcmk_resource_t *launcher = NULL;
pcmk_node_t *host = NULL;
const char *node_type = "Remote";
if (rsc == NULL) {
/* This is a leftover node state entry for a former Pacemaker Remote
* node whose connection resource was removed. Consider it offline.
*/
crm_trace("Pacemaker Remote node %s is considered OFFLINE because "
"its connection resource has been removed from the CIB",
this_node->private->id);
this_node->details->online = FALSE;
return;
}
launcher = rsc->private->launcher;
if (launcher != NULL) {
node_type = "Guest";
if (pcmk__list_of_1(rsc->private->active_nodes)) {
host = rsc->private->active_nodes->data;
}
}
/* If the resource is currently started, mark it online. */
if (rsc->private->orig_role == pcmk_role_started) {
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if ((rsc->private->orig_role == pcmk_role_started)
&& (rsc->private->next_role == pcmk_role_stopped)) {
crm_trace("%s node %s shutting down because connection resource is stopping",
node_type, this_node->private->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if ((launcher != NULL) && pcmk_is_set(launcher->flags, pcmk__rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->private->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
node_type, this_node->private->id);
this_node->details->online = FALSE;
} else if ((rsc->private->orig_role == pcmk_role_stopped)
|| ((launcher != NULL)
&& (launcher->private->orig_role == pcmk_role_stopped))) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
node_type, this_node->private->id);
this_node->details->online = FALSE;
pcmk__clear_node_flags(this_node, pcmk__node_remote_reset);
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->private->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else {
crm_trace("%s node %s is %s",
node_type, this_node->private->id,
this_node->details->online? "ONLINE" : "OFFLINE");
}
}
static void
determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
if (pe__shutdown_requested(this_node)) {
this_node->details->shutdown = TRUE;
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
pcmk__set_node_flags(this_node, pcmk__node_expected_up);
}
if (this_node->private->variant == pcmk__node_variant_ping) {
this_node->details->unclean = FALSE;
online = FALSE; /* As far as resource management is concerned,
* the node is safely offline.
* Anyone caught abusing this logic will be shot
*/
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
online = determine_online_status_no_fencing(scheduler, node_state,
this_node);
} else {
online = determine_online_status_fencing(scheduler, node_state,
this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (this_node->private->variant == pcmk__node_variant_ping) {
crm_info("%s is not a Pacemaker node", pcmk__node_name(this_node));
} else if (this_node->details->unclean) {
- pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node));
+ pcmk__sched_warn(scheduler, "%s is unclean",
+ pcmk__node_name(this_node));
} else if (!this_node->details->online) {
crm_trace("%s is offline", pcmk__node_name(this_node));
} else if (this_node->details->shutdown) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
} else if (this_node->details->pending) {
crm_info("%s is pending", pcmk__node_name(this_node));
} else if (pcmk_is_set(this_node->private->flags, pcmk__node_standby)) {
crm_info("%s is in standby", pcmk__node_name(this_node));
} else if (this_node->details->maintenance) {
crm_info("%s is in maintenance", pcmk__node_name(this_node));
} else {
crm_info("%s is online", pcmk__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 = pcmk__assert_alloc(base_name_len + 3, sizeof(char));
memcpy(zero, last_rsc_id, base_name_len);
zero[base_name_len] = ':';
zero[base_name_len + 1] = '0';
return zero;
}
static pcmk_resource_t *
create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE);
pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none);
crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Orphan resource");
if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pcmk_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pcmk_find_node(scheduler, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, NULL,
scheduler);
}
link_rsc2remotenode(scheduler, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) {
// This removed resource needs to be mapped to a launcher
crm_trace("Launched resource %s was removed from the configuration",
rsc_id);
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_launched);
}
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed);
scheduler->resources = g_list_append(scheduler->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create orphan instance for anonymous clone resource history
*
* \param[in,out] parent Clone resource that orphan will be added to
* \param[in] rsc_id Orphan's resource ID
* \param[in] node Where orphan is active (for logging only)
* \param[in,out] scheduler Scheduler data
*
* \return Newly added orphaned instance of \p parent
*/
static pcmk_resource_t *
create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *top = pe__create_clone_child(parent, scheduler);
pcmk_resource_t *orphan = NULL;
// find_rsc() because we might be a cloned group
orphan = top->private->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s",
top->id, parent->id, rsc_id, pcmk__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 \c PCMK_META_CLONE_MAX
* instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX
* instances are already active).
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node on which to check for instance
* \param[in,out] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (no instance)
*/
static pcmk_resource_t *
find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
pcmk_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
CRM_ASSERT(pcmk__is_anonymous_clone(parent));
// Check for active (or partially active, for cloned groups) instance
pcmk__rsc_trace(parent, "Looking for %s on %s in %s",
rsc_id, pcmk__node_name(node), parent->id);
for (rIter = parent->private->children;
(rIter != NULL) && (rsc == NULL); rIter = rIter->next) {
GList *locations = NULL;
pcmk_resource_t *child = rIter->data;
/* Check whether this instance is already known to be active or pending
* anywhere, at this stage of unpacking. Because this function is called
* for a resource before the resource's individual operation history
* entries are unpacked, locations will generally not contain the
* desired node.
*
* However, there are three exceptions:
* (1) when child is a cloned group and we have already unpacked the
* history of another member of the group on the same node;
* (2) when we've already unpacked the history of another numbered
* instance on the same node (which can happen if
* PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and
* (3) when we re-run calculations on the same scheduler data as part of
* a simulation.
*/
child->private->fns->location(child, &locations, 2);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (pcmk__same_node((pcmk_node_t *) locations->data, node)) {
/* 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->private->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 PCMK_META_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->private->active_nodes != NULL) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple (orphan) instance histories on %s",
parent->id, pcmk__node_name(node));
skip_inactive = TRUE;
rsc = NULL;
} else {
pcmk__rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pcmk__rsc_blocked)) {
// Remember one inactive instance in case we don't find active
inactive_instance =
parent->private->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 != NULL) {
const pcmk_node_t *pending_node = NULL;
pending_node = inactive_instance->private->pending_node;
if ((pending_node != NULL)
&& !pcmk__same_node(pending_node, node)) {
inactive_instance = NULL;
}
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pcmk__rsc_trace(parent, "Resource %s, empty slot",
inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or
* PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we
* don't want to consume a valid instance number for unclean nodes. Such
* instances may appear to be active according to the history, but should be
* considered inactive, so we can start an instance elsewhere. Treat such
* instances as orphans.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pcmk__is_guest_or_bundle_node(node)
&& !pe__is_universal_clone(parent, scheduler)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler);
pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pcmk_resource_t *
unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
const char *rsc_id)
{
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(scheduler->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0,
* we create a single :0 orphan to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources,
clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (orphan)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->private->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 (pcmk__is_anonymous_clone(parent)) {
if (pcmk__is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->private->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(scheduler, node, parent, base);
free(base);
CRM_ASSERT(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none)
&& !pcmk__str_eq(rsc_id, rsc->private->history_id, pcmk__str_none)) {
pcmk__str_update(&(rsc->private->history_id), rsc_id);
pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pcmk__node_name(node), rsc->id,
pcmk_is_set(rsc->flags, pcmk__rsc_removed)? " (ORPHAN)" : "");
}
return rsc;
}
static pcmk_resource_t *
process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
crm_debug("Detected orphan resource %s on %s",
rsc_id, pcmk__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -PCMK_SCORE_INFINITY,
"__orphan_do_not_run__", scheduler);
}
return rsc;
}
static void
process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node,
enum pcmk__on_fail on_fail)
{
pcmk_node_t *tmpnode = NULL;
char *reason = NULL;
enum pcmk__on_fail save_on_fail = pcmk__on_fail_ignore;
pcmk_scheduler_t *scheduler = NULL;
bool known_active = false;
CRM_ASSERT(rsc);
scheduler = rsc->private->scheduler;
known_active = (rsc->private->orig_role > pcmk_role_stopped);
pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, pcmk_role_text(rsc->private->orig_role),
pcmk__node_name(node), pcmk__on_fail_text(on_fail));
/* process current state */
if (rsc->private->orig_role != pcmk_role_unknown) {
pcmk_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->private->probed_nodes,
node->private->id) == NULL) {
pcmk_node_t *n = pe__copy_node(node);
pcmk__rsc_trace(rsc, "%s (%s in history) known on %s",
rsc->id,
pcmk__s(rsc->private->history_id, "the same"),
pcmk__node_name(n));
g_hash_table_insert(iter->private->probed_nodes,
(gpointer) n->private->id, n);
}
if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) {
break;
}
iter = iter->private->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if (known_active && !node->details->online && !node->details->maintenance
&& pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
gboolean should_fence = FALSE;
/* If this is a guest node, fence it (regardless of whether fencing is
* enabled, because guest node fencing is done by recovery of the
* container resource rather than by the fencer). Mark the resource
* we're processing as failed. When the guest comes back up, its
* operation history in the CIB will be cleared, freeing the affected
* resource to run again once we are sure we know its state.
*/
if (pcmk__is_guest_or_bundle_node(node)) {
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
should_fence = TRUE;
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
if (pcmk__is_remote_node(node)
&& (node->private->remote != NULL)
&& !pcmk_is_set(node->private->remote->flags,
pcmk__rsc_failed)) {
/* Setting unseen means that fencing of the remote node will
* occur only if the connection resource is not going to start
* somewhere. This allows connection resources on a failed
* cluster node to move to another node without requiring the
* remote nodes to be fenced as well.
*/
pcmk__clear_node_flags(node, pcmk__node_seen);
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(scheduler, 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:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
demote_action(rsc, node, FALSE);
break;
case pcmk__on_fail_fence_node:
/* treat it as if it is still running
* but also mark the node as unclean
*/
reason = crm_strdup_printf("%s failed there", rsc->id);
pe_fence_node(scheduler, node, reason, FALSE);
free(reason);
break;
case pcmk__on_fail_standby_node:
pcmk__set_node_flags(node,
pcmk__node_standby|pcmk__node_fail_standby);
break;
case pcmk__on_fail_block:
/* is_managed == FALSE will prevent any
* actions being sent for the resource
*/
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked);
break;
case pcmk__on_fail_ban:
/* make sure it comes up somewhere else
* or not at all
*/
resource_location(rsc, node, -PCMK_SCORE_INFINITY,
"__action_migration_auto__", scheduler);
break;
case pcmk__on_fail_stop:
pe__set_next_role(rsc, pcmk_role_stopped,
PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP);
break;
case pcmk__on_fail_restart:
if (known_active) {
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_restart_container:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if ((rsc->private->launcher != NULL) && pcmk__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.
*/
scheduler->stop_needed = g_list_prepend(scheduler->stop_needed,
rsc->private->launcher);
} else if (rsc->private->launcher != NULL) {
stop_action(rsc->private->launcher, node, FALSE);
} else if (known_active) {
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_reset_remote:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
tmpnode = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
}
if (pcmk__is_remote_node(tmpnode)
&& !pcmk_is_set(tmpnode->private->flags,
pcmk__node_remote_fenced)) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(scheduler, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (known_active) {
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->private->remote_reconnect_ms > 0U) {
pe__set_next_role(rsc, pcmk_role_stopped, "remote reset");
}
break;
}
/* Ensure a remote connection failure forces an unclean Pacemaker Remote
* node to be fenced. By marking the node as seen, the failure will result
* in a fencing operation regardless if we're going to attempt to reconnect
* in this transition.
*/
if (pcmk_all_flags_set(rsc->flags,
pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
pcmk__set_node_flags(tmpnode, pcmk__node_seen);
}
}
if (known_active) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Removed resource %s is active on %s and will be "
"stopped when possible",
rsc->id, pcmk__node_name(node));
} else {
crm_notice("Removed resource %s must be stopped manually on %s "
"because " PCMK_OPT_STOP_ORPHAN_RESOURCES
" is set to false", rsc->id, pcmk__node_name(node));
}
}
native_add_running(rsc, node, scheduler,
(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:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
break;
default:
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
break;
}
} else if ((rsc->private->history_id != NULL)
&& (strchr(rsc->private->history_id, ':') != NULL)) {
/* Only do this for older status sections that included instance numbers
* Otherwise stopped instances will appear as orphans
*/
pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)",
rsc->private->history_id, rsc->id);
free(rsc->private->history_id);
rsc->private->history_id = NULL;
} else {
GList *possible_matches = pe__resource_actions(rsc, node,
PCMK_ACTION_STOP, FALSE);
GList *gIter = possible_matches;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *stop = (pcmk_action_t *) gIter->data;
pcmk__set_action_flags(stop, pcmk__action_optional);
}
g_list_free(possible_matches);
}
/* A successful stop after migrate_to on the migration source doesn't make
* the partially migrated resource stopped on the migration target.
*/
if ((rsc->private->orig_role == pcmk_role_stopped)
&& (rsc->private->active_nodes != NULL)
&& (rsc->private->partial_migration_target != NULL)
&& pcmk__same_node(rsc->private->partial_migration_source, node)) {
rsc->private->orig_role = pcmk_role_started;
}
}
/* create active recurring operations as optional */
static void
process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc,
int start_index, int stop_index,
GList *sorted_op_list, pcmk_scheduler_t *scheduler)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
CRM_ASSERT(rsc);
pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d",
rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = pcmk__xe_id(rsc_op);
counter++;
if (node->details->online == FALSE) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline",
rsc->id, pcmk__node_name(node));
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pcmk__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pcmk__node_name(node), counter);
continue;
}
crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if (interval_ms == 0) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring",
id, pcmk__node_name(node));
continue;
}
status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: status",
id, pcmk__node_name(node));
continue;
}
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
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, PCMK__XA_RC_CODE);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((scheduler->shutdown_lock > 0)
&& (get_effective_time(scheduler)
> (lock_time + scheduler->shutdown_lock))) {
pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pcmk__node_name(node));
pe__clear_resource_history(rsc, node);
} else {
rsc->private->lock_node = node;
rsc->private->lock_time = lock_time;
}
}
}
/*!
* \internal
* \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked
* \param[in,out] scheduler Scheduler data
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pcmk_resource_t *
unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource,
pcmk_scheduler_t *scheduler)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = pcmk_role_unknown;
const char *rsc_id = pcmk__xe_id(lrm_resource);
pcmk_resource_t *rsc = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
xmlNode *rsc_op = NULL;
xmlNode *last_failure = NULL;
enum pcmk__on_fail on_fail = pcmk__on_fail_ignore;
enum rsc_role_e saved_role = pcmk_role_unknown;
if (rsc_id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE
" entry: No " PCMK_XA_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s",
rsc_id, pcmk__node_name(node));
/* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort
* them
*/
for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next_same(rsc_op)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(scheduler, node, rsc_id);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_orphan_resource(lrm_resource, node, scheduler);
}
}
CRM_ASSERT(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, scheduler);
}
/* process operations */
saved_role = rsc->private->orig_role;
rsc->private->orig_role = pcmk_role_unknown;
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail);
}
/* create active recurring operations as optional */
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
process_recurring(node, rsc, start_index, stop_index, sorted_op_list,
scheduler);
/* no need to free the contents */
g_list_free(sorted_op_list);
process_rsc_state(rsc, node, on_fail);
if (get_target_role(rsc, &req_role)) {
if ((rsc->private->next_role == pcmk_role_unknown)
|| (req_role < rsc->private->next_role)) {
pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE);
} else if (req_role > rsc->private->next_role) {
pcmk__rsc_info(rsc,
"%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, pcmk_role_text(rsc->private->next_role),
pcmk_role_text(req_role));
}
}
if (saved_role > rsc->private->orig_role) {
rsc->private->orig_role = saved_role;
}
return rsc;
}
static void
handle_removed_launched_resources(const xmlNode *lrm_rsc_list,
pcmk_scheduler_t *scheduler)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list, NULL,
NULL, NULL);
rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) {
pcmk_resource_t *rsc;
pcmk_resource_t *launcher = NULL;
const char *rsc_id;
const char *launcher_id = NULL;
if (!pcmk__xe_is(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
continue;
}
launcher_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER);
rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
if ((launcher_id == NULL) || (rsc_id == NULL)) {
continue;
}
launcher = pe_find_resource(scheduler->resources, launcher_id);
if (launcher == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->resources, rsc_id);
if ((rsc == NULL) || (rsc->private->launcher != NULL)
|| !pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
continue;
}
pcmk__rsc_trace(rsc, "Mapped launcher of removed resource %s to %s",
rsc->id, launcher_id);
rsc->private->launcher = launcher;
launcher->private->launched = g_list_append(launcher->private->launched,
rsc);
}
}
/*!
* \internal
* \brief Unpack one node's lrm status section
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] xml CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler)
{
bool found_removed_launched_resource = false;
// Drill down to PCMK__XE_LRM_RESOURCES section
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL);
if (xml == NULL) {
return;
}
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL);
if (xml == NULL) {
return;
}
// Unpack each PCMK__XE_LRM_RESOURCE entry
for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL; rsc_entry = pcmk__xe_next_same(rsc_entry)) {
pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
found_removed_launched_resource = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* removed launched resources to their launchers.
*/
if (found_removed_launched_resource) {
handle_removed_launched_resources(xml, scheduler);
}
}
static void
set_active(pcmk_resource_t *rsc)
{
const pcmk_resource_t *top = pe__const_top_resource(rsc, false);
if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) {
rsc->private->orig_role = pcmk_role_unpromoted;
} else {
rsc->private->orig_role = pcmk_role_started;
}
}
static void
set_node_score(gpointer key, gpointer value, gpointer user_data)
{
pcmk_node_t *node = value;
int *score = user_data;
node->assign->score = *score;
}
#define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE
#define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \
"/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
int target_rc, pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL),
return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']"
SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'",
NULL);
/* Need to check against transition_magic too? */
if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
if (xml && target_rc >= 0) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_EXEC_ERROR;
crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc);
crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status);
if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) {
return NULL;
}
}
return xml;
}
static xmlNode *
find_lrm_resource(const char *rsc_id, const char *node_name,
pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']",
NULL);
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
return xml;
}
/*!
* \internal
* \brief Check whether a resource has no completed action history on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node_name Node to check
*
* \return true if \p rsc_id is unknown on \p node_name, otherwise false
*/
static bool
unknown_on_node(pcmk_resource_t *rsc, const char *node_name)
{
bool result = false;
xmlXPathObjectPtr search;
char *xpath = NULL;
xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
SUB_XPATH_LRM_RSC_OP
"[@" PCMK__XA_RC_CODE "!='%d']",
node_name, rsc->id, PCMK_OCF_UNKNOWN);
search = xpath_search(rsc->private->scheduler->input, xpath);
result = (numXpathResults(search) == 0);
freeXpathObject(search);
free(xpath);
return result;
}
/*!
* \internal
* \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,out] scheduler Scheduler data
*
* \return true if such a monitor happened after event, false otherwise
*/
static bool
monitor_not_running_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
/* Any probe/monitor operation on the node indicating it was not running
* there
*/
xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name,
NULL, PCMK_OCF_NOT_RUNNING, scheduler);
return (monitor != NULL) && (pe__is_newer_op(monitor, xml_op) > 0);
}
/*!
* \internal
* \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,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
non_monitor_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
xmlNode *lrm_resource = NULL;
lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler);
if (lrm_resource == NULL) {
return false;
}
for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP,
NULL, NULL);
op != NULL; op = pcmk__xe_next_same(op)) {
const char * task = NULL;
if (op == xml_op) {
continue;
}
task = crm_element_value(op, PCMK_XA_OPERATION);
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) > 0) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether the resource has newer state on a node after a migration
* attempt
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] migrate_to Any migrate_to event that is being compared to
* \param[in] migrate_from Any migrate_from event that is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
newer_state_after_migrate(const char *rsc_id, const char *node_name,
const xmlNode *migrate_to,
const xmlNode *migrate_from,
pcmk_scheduler_t *scheduler)
{
const xmlNode *xml_op = (migrate_from != NULL)? migrate_from : migrate_to;
const char *source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE);
/* It's preferred to compare to the migrate event on the same node if
* existing, since call ids are more reliable.
*/
if ((xml_op != migrate_to) && (migrate_to != NULL)
&& pcmk__str_eq(node_name, source, pcmk__str_casei)) {
xml_op = migrate_to;
}
/* 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, scheduler)
|| monitor_not_running_after(rsc_id, node_name, xml_op, scheduler);
}
/*!
* \internal
* \brief Parse migration source and target node names from history entry
*
* \param[in] entry Resource history entry for a migration action
* \param[in] source_node If not NULL, source must match this node
* \param[in] target_node If not NULL, target must match this node
* \param[out] source_name Where to store migration source node name
* \param[out] target_name Where to store migration target node name
*
* \return Standard Pacemaker return code
*/
static int
get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node,
const pcmk_node_t *target_node,
const char **source_name, const char **target_name)
{
*source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE);
*target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
PCMK__META_MIGRATE_SOURCE " and "
PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry));
return pcmk_rc_unpack_error;
}
if ((source_node != NULL)
&& !pcmk__str_eq(*source_name, source_node->private->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_SOURCE "='%s' does not match %s",
pcmk__xe_id(entry), *source_name,
pcmk__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->private->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_TARGET "='%s' does not match %s",
pcmk__xe_id(entry), *target_name,
pcmk__node_name(target_node));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*
* \internal
* \brief Add a migration source to a resource's list of dangling migrations
*
* If the migrate_to and migrate_from actions in a live migration both
* succeeded, but there is no stop on the source, the migration is considered
* "dangling." Add the source to the resource's dangling migration list, which
* will be used to schedule a stop on the source without affecting the target.
*
* \param[in,out] rsc Resource involved in migration
* \param[in] node Migration source
*/
static void
add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s",
rsc->id, pcmk__node_name(node));
rsc->private->orig_role = pcmk_role_stopped;
rsc->private->dangling_migration_sources =
g_list_prepend(rsc->private->dangling_migration_sources,
(gpointer) node);
}
/*!
* \internal
* \brief Update resource role etc. after a successful migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_success(struct action_history *history)
{
/* A complete migration sequence is:
* 1. migrate_to on source node (which succeeded if we get to this function)
* 2. migrate_from on target node
* 3. stop on source node
*
* If no migrate_from has happened, the migration is considered to be
* "partial". If the migrate_from succeeded but no stop has happened, the
* migration is considered to be "dangling".
*
* If a successful migrate_to and stop have happened on the source node, we
* still need to check for a partial migration, due to scenarios (easier to
* produce with batch-limit=1) like:
*
* - A resource is migrating from node1 to node2, and a migrate_to is
* initiated for it on node1.
*
* - node2 goes into standby mode while the migrate_to is pending, which
* aborts the transition.
*
* - Upon completion of the migrate_to, a new transition schedules a stop
* on both nodes and a start on node1.
*
* - If the new transition is aborted for any reason while the resource is
* stopping on node1, the transition after that stop completes will see
* the migrate_to and stop on the source, but it's still a partial
* migration, and the resource must be stopped on node2 because it is
* potentially active there due to the migrate_to.
*
* We also need to take into account that either node's history may be
* cleared at any point in the migration process.
*/
int from_rc = PCMK_OCF_OK;
int from_status = PCMK_EXEC_PENDING;
pcmk_node_t *target_node = NULL;
xmlNode *migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
bool source_newer_op = false;
bool target_newer_state = false;
bool active_on_target = false;
pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
// 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,
scheduler);
// 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, scheduler);
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, PCMK__XA_RC_CODE, &from_rc);
crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &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,
scheduler);
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->private->orig_role = pcmk_role_started;
target_node = pcmk_find_node(scheduler, 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, scheduler, TRUE);
} else {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
return;
}
// The migrate_from is pending, complete but erased, or to be scheduled
/* If there is no history at all for the resource on an online target, then
* it was likely cleaned. Just return, and we'll schedule a probe. Once we
* have the probe result, it will be reflected in target_newer_state.
*/
if ((target_node != NULL) && target_node->details->online
&& unknown_on_node(history->rsc, target)) {
return;
}
if (active_on_target) {
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
native_add_running(history->rsc, target_node, scheduler, 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->private->partial_migration_target = target_node;
history->rsc->private->partial_migration_source = source_node;
}
} else if (!source_newer_op) {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_failure(struct action_history *history)
{
xmlNode *target_migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
// 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->private->orig_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, scheduler);
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, scheduler)) {
/* The resource has no newer state on the target, so assume it's still
* active there.
* (if it is up).
*/
pcmk_node_t *target_node = pcmk_find_node(scheduler, target);
if (target_node && target_node->details->online) {
native_add_running(history->rsc, target_node, scheduler, FALSE);
}
} else if (!non_monitor_after(history->rsc->id, source, history->xml,
scheduler)) {
/* 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->private->dangling_migration_sources =
g_list_prepend(history->rsc->private->dangling_migration_sources,
(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;
pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
// 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->private->orig_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, scheduler);
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,
scheduler)) {
/* The resource has no newer state on the source, so assume it's still
* active there (if it is up).
*/
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
if (source_node && source_node->details->online) {
native_add_running(history->rsc, source_node, scheduler, 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)
{
const pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
if (!(history->node->details->online)) {
return;
}
for (const xmlNode *xIter = scheduler->failed->children;
xIter != NULL; xIter = xIter->next) {
const char *key = pcmk__xe_history_key(xIter);
const char *uname = crm_element_value(xIter, PCMK_XA_UNAME);
if (pcmk__str_eq(history->key, key, pcmk__str_none)
&& pcmk__str_eq(uname, history->node->private->name,
pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s",
history->key, pcmk__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pcmk__node_name(history->node));
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->private->name);
crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id);
pcmk__xml_copy(scheduler->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, PCMK_XA_LAST_RC_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 = pcmk__str_copy(p);
}
free(when_s);
}
if (result == NULL) {
result = pcmk__str_copy("unknown_time");
}
return result;
}
/*!
* \internal
* \brief Ban a resource (or its clone if an anonymous instance) from all nodes
*
* \param[in,out] rsc Resource to ban
*/
static void
ban_from_all_nodes(pcmk_resource_t *rsc)
{
int score = -PCMK_SCORE_INFINITY;
const pcmk_scheduler_t *scheduler = rsc->private->scheduler;
if (rsc->private->parent != NULL) {
pcmk_resource_t *parent = uber_parent(rsc);
if (pcmk__is_anonymous_clone(parent)) {
/* For anonymous clones, if an operation with
* PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the
* entire clone must stop.
*/
rsc = parent;
}
}
// Ban the resource from all nodes
crm_notice("%s will not be started under current conditions", rsc->id);
if (rsc->private->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->private->allowed_nodes);
}
rsc->private->allowed_nodes = pe__node_list2table(scheduler->nodes);
g_hash_table_foreach(rsc->private->allowed_nodes, set_node_score, &score);
}
/*!
* \internal
* \brief Get configured failure handling and role after failure for an action
*
* \param[in,out] history Unpacked action history entry
* \param[out] on_fail Where to set configured failure handling
* \param[out] fail_role Where to set to role after failure
*/
static void
unpack_failure_handling(struct action_history *history,
enum pcmk__on_fail *on_fail,
enum rsc_role_e *fail_role)
{
xmlNode *config = pcmk__find_action_config(history->rsc, history->task,
history->interval_ms, true);
GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node,
history->task,
history->interval_ms, config);
const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL);
*on_fail = pcmk__parse_on_fail(history->rsc, history->task,
history->interval_ms, on_fail_str);
*fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail,
meta);
g_hash_table_destroy(meta);
}
/*!
* \internal
* \brief Update resource role, failure handling, etc., after a failed action
*
* \param[in,out] history Parsed action result history
* \param[in] config_on_fail Action failure handling from configuration
* \param[in] fail_role Resource's role after failure of this action
* \param[out] last_failure This will be set to the history XML
* \param[in,out] on_fail Actual handling of action result
*/
static void
unpack_rsc_op_failure(struct action_history *history,
enum pcmk__on_fail config_on_fail,
enum rsc_role_e fail_role, xmlNode **last_failure,
enum pcmk__on_fail *on_fail)
{
bool is_probe = false;
char *last_change_s = NULL;
pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
*last_failure = history->xml;
is_probe = pcmk_xe_is_probe(history->xml);
last_change_s = last_change_str(history->xml);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)
&& (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
crm_trace("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " QB_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,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
} else {
- pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of "
+ pcmk__sched_warn(scheduler,
+ "Unexpected result (%s%s%s) was recorded for %s of "
"%s on %s at %s " QB_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,
pcmk__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 " PCMK_XA_RESOURCE_DISCOVERY " option for location "
"constraints",
history->rsc->id, pcmk__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (*on_fail < config_on_fail) {
pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s",
pcmk__on_fail_text(*on_fail),
pcmk__on_fail_text(config_on_fail), history->key);
*on_fail = config_on_fail;
}
if (strcmp(history->task, PCMK_ACTION_STOP) == 0) {
resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY,
"__stop_fail__", scheduler);
} 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->private->orig_role = pcmk_role_promoted;
} else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) {
if (config_on_fail == pcmk__on_fail_block) {
history->rsc->private->orig_role = pcmk_role_promoted;
pe__set_next_role(history->rsc, pcmk_role_stopped,
"demote with " PCMK_META_ON_FAIL "=block");
} else if (history->exit_status == PCMK_OCF_NOT_RUNNING) {
history->rsc->private->orig_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->private->orig_role = pcmk_role_unpromoted;
}
}
if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
/* leave stopped */
pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id);
history->rsc->private->orig_role = pcmk_role_stopped;
} else if (history->rsc->private->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id);
set_active(history->rsc);
}
pcmk__rsc_trace(history->rsc,
"Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s",
history->rsc->id,
pcmk_role_text(history->rsc->private->orig_role),
pcmk__btoa(history->node->details->unclean),
pcmk__on_fail_text(config_on_fail),
pcmk_role_text(fail_role));
if ((fail_role != pcmk_role_started)
&& (history->rsc->private->next_role < fail_role)) {
pe__set_next_role(history->rsc, fail_role, "failure");
}
if (fail_role == pcmk_role_stopped) {
ban_from_all_nodes(history->rsc);
}
}
/*!
* \internal
* \brief Block a resource with a failed action if it cannot be recovered
*
* If resource action is a failed stop and fencing is not possible, mark the
* resource as unmanaged and blocked, since recovery cannot be done.
*
* \param[in,out] history Parsed action history entry
*/
static void
block_if_unrecoverable(struct action_history *history)
{
char *last_change_s = NULL;
if (strcmp(history->task, PCMK_ACTION_STOP) != 0) {
return; // All actions besides stop are always recoverable
}
if (pe_can_fence(history->node->private->scheduler, history->node)) {
return; // Failed stops are recoverable via fencing
}
last_change_s = last_change_str(history->xml);
- pcmk__sched_err("No further recovery can be attempted for %s "
+ pcmk__sched_err(history->node->private->scheduler,
+ "No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
QB_XS " rc=%d id=%s",
history->rsc->id, history->task,
pcmk__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);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked);
}
/*!
* \internal
* \brief Update action history's execution status and why
*
* \param[in,out] history Parsed action history entry
* \param[out] why Where to store reason for update
* \param[in] value New value
* \param[in] reason Description of why value was changed
*/
static inline void
remap_because(struct action_history *history, const char **why, int value,
const char *reason)
{
if (history->execution_status != value) {
history->execution_status = value;
*why = reason;
}
}
/*!
* \internal
* \brief Remap informational monitor results and operation status
*
* For the monitor results, certain OCF codes are for providing extended information
* to the user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by Pacemaker.
*
* For operation status, the action result can be used to determine an appropriate
* status for the purposes of responding to the action. The status provided by the
* executor is not directly usable since the executor does not know what was expected.
*
* \param[in,out] history Parsed action history entry
* \param[in,out] on_fail What should be done about the result
* \param[in] expired Whether result is expired
*
* \note If the result is remapped and the node is not shutting down or failed,
* the operation will be recorded in the scheduler data's list of failed
* operations to highlight it for the user.
*
* \note This may update the resource's current and next role.
*/
static void
remap_operation(struct action_history *history,
enum pcmk__on_fail *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 (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& ((history->execution_status != PCMK_EXEC_DONE)
|| (history->exit_status != PCMK_OCF_NOT_RUNNING))) {
history->execution_status = PCMK_EXEC_DONE;
history->exit_status = PCMK_OCF_NOT_RUNNING;
why = "equivalent probe result";
}
/* If the executor reported an execution status of anything but done or
* error, consider that final. But for done or error, we know better whether
* it should be treated as a failure or not, because we know the expected
* result.
*/
switch (history->execution_status) {
case PCMK_EXEC_DONE:
case PCMK_EXEC_ERROR:
break;
// These should be treated as node-fatal
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"node-fatal error");
goto remap_done;
default:
goto remap_done;
}
is_probe = pcmk_xe_is_probe(history->xml);
if (is_probe) {
task = "probe";
}
if (history->expected_exit_status < 0) {
/* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with
* Heartbeat 2.0.7 or earlier as the cluster layer, did not include the
* expected exit status in the transition key, which (along with the
* similar case of a corrupted transition key in the CIB) will be
* reported to this function as -1. Pacemaker 2.0+ does not support
* rolling upgrades from those versions or processing of saved CIB files
* from those versions, so we do not need to care much about this case.
*/
remap_because(history, &why, PCMK_EXEC_ERROR,
"obsolete history format");
pcmk__config_warn("Expected result not found for %s on %s "
"(corrupt or obsolete CIB?)",
history->key, pcmk__node_name(history->node));
} else if (history->exit_status == history->expected_exit_status) {
remap_because(history, &why, PCMK_EXEC_DONE, "expected result");
} else {
remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result");
pcmk__rsc_debug(history->rsc,
"%s on %s: expected %d (%s), got %d (%s%s%s)",
history->key, pcmk__node_name(history->node),
history->expected_exit_status,
services_ocf_exitcode_str(history->expected_exit_status),
history->exit_status,
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""));
}
switch (history->exit_status) {
case PCMK_OCF_OK:
if (is_probe
&& (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active on %s at %s",
history->rsc->id, pcmk__node_name(history->node),
last_change_s);
free(last_change_s);
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe
|| (history->expected_exit_status == history->exit_status)
|| !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) {
/* For probes, recurring monitors for the Stopped role, and
* unmanaged resources, "not running" is not considered a
* failure.
*/
remap_because(history, &why, PCMK_EXEC_DONE, "exit status");
history->rsc->private->orig_role = pcmk_role_stopped;
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe
&& (history->exit_status != history->expected_exit_status)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active and promoted on %s at %s",
history->rsc->id,
pcmk__node_name(history->node), last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->private->orig_role = pcmk_role_promoted;
}
break;
case PCMK_OCF_FAILED_PROMOTED:
if (!expired) {
history->rsc->private->orig_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, PCMK_META_INTERVAL,
&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,
pcmk__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure,
rsc->private->scheduler);
} else {
pcmk__op_digest_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->private->scheduler);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, pcmk__xe_history_key(xml_op),
node->private->id);
break;
case pcmk__digest_match:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn,
pcmk_scheduler_t *scheduler)
{
pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id);
if (remote_node) {
pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, scheduler);
order_actions(fence, action, pcmk__ar_first_implies_then);
}
}
static bool
should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task,
guint interval_ms, bool is_last_failure)
{
/* Clearing failures of recurring monitors has special concerns. The
* executor reports only changes in the monitor result, so if the
* monitor is still active and still getting the same failure result,
* that will go undetected after the failure is cleared.
*
* Also, the operation history will have the time when the recurring
* monitor result changed to the given code, not the time when the
* result last happened.
*
* @TODO We probably should clear such failures only when the failure
* timeout has passed since the last occurrence of the failed result.
* However we don't record that information. We could maybe approximate
* that by clearing only if there is a more recent successful monitor or
* stop result, but we don't even have that information at this point
* since we are still unpacking the resource's operation history.
*
* This is especially important for remote connection resources with a
* reconnect interval, so in that case, we skip clearing failures
* if the remote node hasn't been fenced.
*/
if ((rsc->private->remote_reconnect_ms > 0U)
&& pcmk_is_set(rsc->private->scheduler->flags,
pcmk_sched_fencing_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pcmk_node_t *remote_node = pcmk_find_node(rsc->private->scheduler,
rsc->id);
if (remote_node && !pcmk_is_set(remote_node->private->flags,
pcmk__node_remote_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;
const guint expiration_sec =
history->rsc->private->failure_expiration_ms / 1000;
pcmk_scheduler_t *scheduler = history->rsc->private->scheduler;
if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) {
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not expired: "
"Not Installed does not expire",
history->id, pcmk__node_name(history->node));
return false; // "Not installed" must always be cleared manually
}
if ((expiration_sec > 0)
&& (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE,
&last_run) == 0)) {
/* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a
* timestamp
*/
time_t now = get_effective_time(scheduler);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + expiration_sec))
&& !should_ignore_failure_timeout(history->rsc, history->task,
history->interval_ms,
is_last_failure)) {
expired = true;
}
// Does the resource as a whole have an unexpired fail count?
unexpired_fail_count = pe_get_failcount(history->node, history->rsc,
&last_failure,
pcmk__fc_effective,
history->xml);
// Update scheduler recheck time according to *last* failure
crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d "
"expiration=%s last-failure@%lld",
history->id, (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count,
pcmk__readable_interval(expiration_sec * 1000),
(long long) last_failure);
last_failure += expiration_sec + 1;
if (unexpired_fail_count && (now < last_failure)) {
pe__update_recheck_time(last_failure, scheduler,
"fail count expiration");
}
}
if (expired) {
if (pe_get_failcount(history->node, history->rsc, NULL,
pcmk__fc_default, history->xml)) {
// There is a fail count ignoring timeout
if (unexpired_fail_count == 0) {
// There is no fail count considering timeout
clear_reason = "it expired";
} else {
/* This operation is old, but there is an unexpired fail count.
* In a properly functioning cluster, this should only be
* possible if this operation is not a failure (otherwise the
* fail count should be expired too), so this is really just a
* failsafe.
*/
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Unexpired fail count",
history->id, pcmk__node_name(history->node));
expired = false;
}
} else if (is_last_failure
&& (history->rsc->private->remote_reconnect_ms > 0U)) {
/* Clear any expired last failure when reconnect interval is set,
* even if there is no fail count.
*/
clear_reason = "reconnect interval is set";
}
}
if (!expired && is_last_failure
&& should_clear_for_param_change(history->xml, history->task,
history->rsc, history->node)) {
clear_reason = "resource parameters have changed";
}
if (clear_reason != NULL) {
pcmk_action_t *clear_op = NULL;
// Schedule clearing of the fail count
clear_op = pe__clear_failcount(history->rsc, history->node,
clear_reason, scheduler);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)
&& (history->rsc->private->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, scheduler);
}
}
if (expired && (history->interval_ms == 0)
&& pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
switch (history->exit_status) {
case PCMK_OCF_OK:
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Don't expire probes that return these values
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Probe result",
history->id, pcmk__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, PCMK__XA_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
/*!
* \internal
* \brief Update a resource's state for an action result
*
* \param[in,out] history Parsed action history entry
* \param[in] exit_status Exit status to base new state on
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
update_resource_state(struct action_history *history, int exit_status,
const xmlNode *last_failure,
enum pcmk__on_fail *on_fail)
{
bool clear_past_failure = false;
if ((exit_status == PCMK_OCF_NOT_INSTALLED)
|| (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml))) {
history->rsc->private->orig_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, pcmk__xe_history_key(last_failure),
pcmk__str_none)) {
clear_past_failure = true;
}
if (history->rsc->private->orig_role < pcmk_role_started) {
set_active(history->rsc);
}
} else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) {
history->rsc->private->orig_role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) {
history->rsc->private->orig_role = pcmk_role_stopped;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
history->rsc->private->orig_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
* PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE
*/
clear_past_failure = true;
}
history->rsc->private->orig_role = pcmk_role_unpromoted;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
history->rsc->private->orig_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->private->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "%s active on %s",
history->rsc->id, pcmk__node_name(history->node));
set_active(history->rsc);
}
if (!clear_past_failure) {
return;
}
switch (*on_fail) {
case pcmk__on_fail_stop:
case pcmk__on_fail_ban:
case pcmk__on_fail_standby_node:
case pcmk__on_fail_fence_node:
pcmk__rsc_trace(history->rsc,
"%s (%s) is not cleared by a completed %s",
history->rsc->id, pcmk__on_fail_text(*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->private->remote_reconnect_ms == 0U) {
/* 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, PCMK__XA_OP_STATUS,
&(history->execution_status)) < 0)
|| (history->execution_status < PCMK_EXEC_PENDING)
|| (history->execution_status > PCMK_EXEC_MAX)
|| (history->execution_status == PCMK_EXEC_CANCELLED)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_OP_STATUS " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_OP_STATUS),
""));
return pcmk_rc_unpack_error;
}
if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE,
&(history->exit_status)) < 0)
|| (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) {
#if 0
/* @COMPAT We should ignore malformed entries, but since that would
* change behavior, it should be done at a major or minor series
* release.
*/
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_RC_CODE " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_RC_CODE),
""));
return pcmk_rc_unpack_error;
#else
history->exit_status = CRM_EX_ERROR;
#endif
}
history->exit_reason = crm_element_value(history->xml, PCMK_XA_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 (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& (orig_exit_status != history->expected_exit_status)) {
if (history->rsc->private->orig_role <= pcmk_role_stopped) {
history->rsc->private->orig_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,
pcmk__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,
pcmk__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 a PCMK__XA_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, pcmk__node_name(history->node));
crm_xml_add(history->xml, PCMK__XA_OP_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 pcmk__on_fail *on_fail)
{
pcmk_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) {
ban_rsc = uber_parent(history->rsc);
}
crm_notice("Treating probe result '%s' for %s on %s as 'not running'",
services_ocf_exitcode_str(orig_exit_status), history->rsc->id,
pcmk__node_name(history->node));
update_resource_state(history, history->expected_exit_status, last_failure,
on_fail);
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->private->name);
record_failed_op(history);
resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY,
"masked-probe-failure", ban_rsc->private->scheduler);
}
/*!
* \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
* \c PCMK__XE_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, PCMK_XA_OPERATION),
pcmk__str_none)) {
return false; // last_failure is for different action
}
if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL,
&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, PCMK_XA_LAST_RC_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(crm_element_value(last_failure,
PCMK_XA_LAST_RC_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) {
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending);
set_active(history->rsc);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->private->orig_role = pcmk_role_promoted;
} else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0)
&& history->node->details->unclean) {
/* A migrate_to action is pending on a unclean source, so force a stop
* on the target.
*/
const char *migrate_target = NULL;
pcmk_node_t *target = NULL;
migrate_target = crm_element_value(history->xml,
PCMK__META_MIGRATE_TARGET);
target = pcmk_find_node(history->rsc->private->scheduler,
migrate_target);
if (target != NULL) {
stop_action(history->rsc, target, FALSE);
}
}
if (history->rsc->private->pending_action != 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_action().
*/
#if 0
history->rsc->private->pending_action = strdup("probe");
history->rsc->private->pending_node = history->node;
#endif
} else {
history->rsc->private->pending_action = strdup(history->task);
history->rsc->private->pending_node = history->node;
}
}
static void
unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum pcmk__on_fail *on_fail)
{
int old_rc = 0;
bool expired = false;
pcmk_resource_t *parent = rsc;
enum rsc_role_e fail_role = pcmk_role_unknown;
enum pcmk__on_fail 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 = pcmk__xe_id(xml_op);
if (history.id == NULL) {
pcmk__config_err("Ignoring resource history entry for %s on %s "
"without ID", rsc->id, pcmk__node_name(node));
return;
}
// Task and interval
history.task = crm_element_value(xml_op, PCMK_XA_OPERATION);
if (history.task == NULL) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"without " PCMK_XA_OPERATION,
history.id, rsc->id, pcmk__node_name(node));
return;
}
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms));
if (!can_affect_state(&history)) {
pcmk__rsc_trace(rsc,
"Ignoring resource history entry %s for %s on %s "
"with irrelevant action '%s'",
history.id, rsc->id, pcmk__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 = pcmk__xe_history_key(xml_op);
crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pcmk__node_name(node),
pcmk_exec_status_str(history.execution_status),
crm_exit_str(history.exit_status));
if (node->details->unclean) {
pcmk__rsc_trace(rsc,
"%s is running on %s, which is unclean (further action "
"depends on value of stop's on-fail attribute)",
rsc->id, pcmk__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 (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) {
mask_probe_failure(&history, old_rc, *last_failure, on_fail);
goto done;
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
parent = uber_parent(rsc);
}
switch (history.execution_status) {
case PCMK_EXEC_PENDING:
process_pending_action(&history, *last_failure);
goto done;
case PCMK_EXEC_DONE:
update_resource_state(&history, history.exit_status, *last_failure,
on_fail);
goto done;
case PCMK_EXEC_NOT_INSTALLED:
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if (failure_strategy == pcmk__on_fail_ignore) {
crm_warn("Cannot ignore failed %s of %s on %s: "
"Resource agent doesn't exist "
QB_XS " status=%d rc=%d id=%s",
history.task, rsc->id, pcmk__node_name(node),
history.execution_status, history.exit_status,
history.id);
/* Also for printing it as "FAILED" by marking it as
* pcmk__rsc_failed later
*/
*on_fail = pcmk__on_fail_ban;
}
resource_location(parent, node, -PCMK_SCORE_INFINITY,
"hard-error", rsc->private->scheduler);
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
goto done;
case PCMK_EXEC_NOT_CONNECTED:
if (pcmk__is_pacemaker_remote_node(node)
&& pcmk_is_set(node->private->remote->flags,
pcmk__rsc_managed)) {
/* We should never get into a situation where a managed remote
* connection resource is considered OK but a resource action
* behind the connection gets a "not connected" status. But as a
* fail-safe in case a bug or unusual circumstances do lead to
* that, ensure the remote connection is considered failed.
*/
pcmk__set_rsc_flags(node->private->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
break; // Not done, do error handling
case PCMK_EXEC_ERROR:
case PCMK_EXEC_ERROR_HARD:
case PCMK_EXEC_ERROR_FATAL:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_INVALID:
break; // Not done, do error handling
default: // No other value should be possible at this point
break;
}
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if ((failure_strategy == pcmk__on_fail_ignore)
|| ((failure_strategy == pcmk__on_fail_restart_container)
&& (strcmp(history.task, PCMK_ACTION_STOP) == 0))) {
char *last_change_s = last_change_str(xml_op);
crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded "
QB_XS " %s",
history.task, services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), rsc->id,
pcmk__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, PCMK_XA_UNAME, node->private->name);
pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure);
record_failed_op(&history);
if ((failure_strategy == pcmk__on_fail_restart_container)
&& (*on_fail <= pcmk__on_fail_restart)) {
*on_fail = failure_strategy;
}
} else {
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
if (history.execution_status == PCMK_EXEC_ERROR_HARD) {
uint8_t log_level = LOG_ERR;
if (history.exit_status == PCMK_OCF_NOT_INSTALLED) {
log_level = LOG_NOTICE;
}
do_crm_log(log_level,
"Preventing %s from restarting on %s because "
"of hard failure (%s%s%s) " QB_XS " %s",
parent->id, pcmk__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, -PCMK_SCORE_INFINITY,
"hard-error", rsc->private->scheduler);
} else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) {
- pcmk__sched_err("Preventing %s from restarting anywhere because "
+ pcmk__sched_err(rsc->private->scheduler,
+ "Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " QB_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, -PCMK_SCORE_INFINITY,
"fatal-error", rsc->private->scheduler);
}
}
done:
pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)",
rsc->id, pcmk__node_name(node), history.id,
pcmk_role_text(rsc->private->orig_role),
pcmk_role_text(rsc->private->next_role));
}
static void
add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite,
pcmk_scheduler_t *scheduler)
{
const char *cluster_name = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
pcmk__insert_dup(node->private->attrs,
CRM_ATTR_UNAME, node->private->name);
pcmk__insert_dup(node->private->attrs, CRM_ATTR_ID, node->private->id);
if (pcmk__str_eq(node->private->id, scheduler->dc_uuid, pcmk__str_casei)) {
scheduler->dc_node = node;
pcmk__insert_dup(node->private->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_TRUE);
} else {
pcmk__insert_dup(node->private->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_FALSE);
}
cluster_name = g_hash_table_lookup(scheduler->config_hash,
PCMK_OPT_CLUSTER_NAME);
if (cluster_name) {
pcmk__insert_dup(node->private->attrs, CRM_ATTR_CLUSTER_NAME,
cluster_name);
}
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data,
node->private->attrs, NULL, overwrite,
scheduler);
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_data,
node->private->utilization, NULL,
FALSE, scheduler);
if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL,
pcmk__rsc_node_current) == NULL) {
const char *site_name = pcmk__node_attr(node, "site-name", NULL,
pcmk__rsc_node_current);
if (site_name) {
pcmk__insert_dup(node->private->attrs,
CRM_ATTR_SITE_NAME, site_name);
} else if (cluster_name) {
/* Default to cluster-name if unset */
pcmk__insert_dup(node->private->attrs,
CRM_ATTR_SITE_NAME, 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, NULL, NULL, NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) {
if (pcmk__xe_is(rsc_op, PCMK__XE_LRM_RSC_OP)) {
crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc);
crm_xml_add(rsc_op, PCMK_XA_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", pcmk__xe_id(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op));
continue;
}
op_list = g_list_append(op_list, rsc_op);
}
g_list_free(sorted_op_list);
return op_list;
}
GList *
find_operations(const char *rsc, const char *node, gboolean active_filter,
pcmk_scheduler_t *scheduler)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS,
NULL, NULL);
pcmk_node_t *this_node = NULL;
xmlNode *node_state = NULL;
CRM_CHECK(status != NULL, return NULL);
for (node_state = pcmk__xe_first_child(status, NULL, NULL, NULL);
node_state != NULL; node_state = pcmk__xe_next(node_state)) {
if (pcmk__xe_is(node_state, PCMK__XE_NODE_STATE)) {
const char *uname = crm_element_value(node_state, PCMK_XA_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pcmk_find_node(scheduler, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
} else {
determine_online_status(node_state, this_node, scheduler);
}
if (this_node->details->online
|| pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
/* offline nodes run no resources...
* unless stonith is enabled in which case we need to
* make sure rsc start events happen after the stonith
*/
xmlNode *lrm_rsc = NULL;
tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL,
NULL);
tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL,
NULL);
for (lrm_rsc = pcmk__xe_first_child(tmp, NULL, NULL, NULL);
lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) {
if (pcmk__xe_is(lrm_rsc, PCMK__XE_LRM_RESOURCE)) {
const char *rsc_id = crm_element_value(lrm_rsc,
PCMK_XA_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 d798a49eb6..d647a3b390 100644
--- a/lib/pengine/utils.c
+++ b/lib/pengine/utils.c
@@ -1,905 +1,906 @@
/*
* 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 <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/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] scheduler Scheduler data
* \param[in] node Name of node to check
*
* \return true if node can be fenced, false otherwise
*/
bool
pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node)
{
if (pcmk__is_guest_or_bundle_node(node)) {
/* A guest or bundle node is fenced by stopping its launcher, which is
* possible if the launcher's host is either online or fenceable.
*/
pcmk_resource_t *rsc = node->private->remote->private->launcher;
for (GList *n = rsc->private->active_nodes; n != NULL; n = n->next) {
pcmk_node_t *launcher_node = n->data;
if (!launcher_node->details->online
&& !pe_can_fence(scheduler, launcher_node)) {
return false;
}
}
return true;
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
return false; /* Turned off */
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) {
return false; /* No devices */
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
return true;
} else if (scheduler->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",
pcmk__node_name(node));
return true;
}
crm_trace("Cannot fence %s", pcmk__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.
*/
pcmk_node_t *
pe__copy_node(const pcmk_node_t *this_node)
{
pcmk_node_t *new_node = NULL;
CRM_ASSERT(this_node != NULL);
new_node = pcmk__assert_alloc(1, sizeof(pcmk_node_t));
new_node->assign = pcmk__assert_alloc(1,
sizeof(struct pcmk__node_assignment));
new_node->assign->probe_mode = this_node->assign->probe_mode;
new_node->assign->score = this_node->assign->score;
new_node->assign->count = this_node->assign->count;
new_node->details = this_node->details;
new_node->private = this_node->private;
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) {
pcmk_node_t *new_node = NULL;
new_node = pe__copy_node((const pcmk_node_t *) gIter->data);
g_hash_table_insert(result, (gpointer) new_node->private->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 pcmk_node_t *node1 = (const pcmk_node_t *) a;
const pcmk_node_t *node2 = (const pcmk_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->private->name, node2->private->name);
}
/*!
* \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] scheduler Scheduler data
*/
static void
pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes, pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->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 pcmk_node_t *node = (const pcmk_node_t *) gIter->data;
out->message(out, "node-weight", rsc, comment, node->private->name,
pcmk_readable_score(node->assign->score));
}
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 pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes)
{
GHashTableIter iter;
pcmk_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,
pcmk__node_name(node),
pcmk_readable_score(node->assign->score));
} else {
qb_log_from_external_source(function, file, "%s: %s = %s",
LOG_TRACE, line, 0,
comment, pcmk__node_name(node),
pcmk_readable_score(node->assign->score));
}
}
}
/*!
* \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] scheduler Scheduler data
*/
void
pe__show_node_scores_as(const char *file, const char *function, int line,
bool to_log, const pcmk_resource_t *rsc,
const char *comment, GHashTable *nodes,
pcmk_scheduler_t *scheduler)
{
if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
// 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, scheduler);
}
if (rsc == NULL) {
return;
}
// If this resource has children, repeat recursively for each
for (GList *gIter = rsc->private->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) gIter->data;
pe__show_node_scores_as(file, function, line, to_log, child, comment,
child->private->allowed_nodes, scheduler);
}
}
/*!
* \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 a's priority > b's priority (or \c b is \c NULL and \c a is not)
* \retval 0 a's priority == b's priority (or both \c a and \c b are \c NULL)
* \retval 1 a's priority < b's priority (or \c a is \c NULL and \c b is not)
*/
gint
pe__cmp_rsc_priority(gconstpointer a, gconstpointer b)
{
const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a;
const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b;
if (a == NULL && b == NULL) {
return 0;
}
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (resource1->private->priority > resource2->private->priority) {
return -1;
}
if (resource1->private->priority < resource2->private->priority) {
return 1;
}
return 0;
}
static void
resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag)
{
pcmk_node_t *match = NULL;
if ((pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes)
|| (node->assign->probe_mode == pcmk__probe_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 {
for (GList *gIter = rsc->private->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
resource_node_score(child_rsc, node, score, tag);
}
}
match = g_hash_table_lookup(rsc->private->allowed_nodes, node->private->id);
if (match == NULL) {
match = pe__copy_node(node);
g_hash_table_insert(rsc->private->allowed_nodes,
(gpointer) match->private->id, match);
}
match->assign->score = pcmk__add_scores(match->assign->score, score);
pcmk__rsc_trace(rsc,
"Enabling %s preference (%s) for %s on %s (now %s)",
tag, pcmk_readable_score(score), rsc->id,
pcmk__node_name(node),
pcmk_readable_score(match->assign->score));
}
void
resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag, pcmk_scheduler_t *scheduler)
{
if (node != NULL) {
resource_node_score(rsc, node, score, tag);
} else if (scheduler != NULL) {
GList *gIter = scheduler->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data;
resource_node_score(rsc, node_iter, score, tag);
}
} else {
GHashTableIter iter;
pcmk_node_t *node_iter = NULL;
g_hash_table_iter_init(&iter, rsc->private->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 == -PCMK_SCORE_INFINITY)
&& (rsc->private->assigned_node != NULL)) {
// @TODO Should this be more like pcmk__unassign_resource()?
crm_info("Unassigning %s from %s",
rsc->id, pcmk__node_name(rsc->private->assigned_node));
free(rsc->private->assigned_node);
rsc->private->assigned_node = NULL;
}
}
time_t
get_effective_time(pcmk_scheduler_t *scheduler)
{
if(scheduler) {
if (scheduler->now == NULL) {
crm_trace("Recording a new 'now'");
scheduler->now = crm_time_new(NULL);
}
return crm_time_get_seconds_since_epoch(scheduler->now);
}
crm_trace("Defaulting to 'now'");
return time(NULL);
}
gboolean
get_target_role(const pcmk_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->private->meta,
PCMK_META_TARGET_ROLE);
CRM_CHECK(role != NULL, return FALSE);
if (pcmk__str_eq(value, PCMK_ROLE_STARTED,
pcmk__str_null_matches|pcmk__str_casei)) {
return FALSE;
}
if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting " PCMK_META_TARGET_ROLE
" to the explicit value '" PCMK_VALUE_DEFAULT
"' is deprecated and will be removed in a "
"future release (just leave it unset)");
return FALSE;
}
local_role = pcmk_parse_role(value);
if (local_role == pcmk_role_unknown) {
pcmk__config_err("Ignoring '" PCMK_META_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,
pcmk__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 '" PCMK_META_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(pcmk_action_t *lh_action, pcmk_action_t *rh_action,
uint32_t flags)
{
GList *gIter = NULL;
pcmk__related_action_t *wrapper = NULL;
GList *list = NULL;
if (flags == pcmk__ar_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) {
pcmk__related_action_t *after = gIter->data;
if ((after->action == rh_action)
&& pcmk_any_flags_set(after->flags, flags)) {
return FALSE;
}
}
wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t));
wrapper->action = rh_action;
wrapper->flags = flags;
list = lh_action->actions_after;
list = g_list_prepend(list, wrapper);
lh_action->actions_after = list;
wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t));
wrapper->action = lh_action;
wrapper->flags = flags;
list = rh_action->actions_before;
list = g_list_prepend(list, wrapper);
rh_action->actions_before = list;
return TRUE;
}
void
destroy_ticket(gpointer data)
{
pcmk__ticket_t *ticket = data;
if (ticket->state) {
g_hash_table_destroy(ticket->state);
}
free(ticket->id);
free(ticket);
}
pcmk__ticket_t *
ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler)
{
pcmk__ticket_t *ticket = NULL;
if (pcmk__str_empty(ticket_id)) {
return NULL;
}
if (scheduler->tickets == NULL) {
scheduler->tickets = pcmk__strkey_table(free, destroy_ticket);
}
ticket = g_hash_table_lookup(scheduler->tickets, ticket_id);
if (ticket == NULL) {
ticket = calloc(1, sizeof(pcmk__ticket_t));
if (ticket == NULL) {
- pcmk__sched_err("Cannot allocate ticket '%s'", ticket_id);
+ pcmk__sched_err(scheduler, "Cannot allocate ticket '%s'",
+ ticket_id);
return NULL;
}
crm_trace("Creating ticket entry for %s", ticket_id);
ticket->id = strdup(ticket_id);
ticket->last_granted = -1;
ticket->state = pcmk__strkey_table(free, free);
g_hash_table_insert(scheduler->tickets, strdup(ticket->id), ticket);
}
return ticket;
}
const char *
rsc_printable_id(const pcmk_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
return rsc->id;
}
return pcmk__xe_id(rsc->private->xml);
}
void
pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__clear_rsc_flags(rsc, flags);
for (GList *gIter = rsc->private->children;
gIter != NULL; gIter = gIter->next) {
pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag)
{
for (GList *lpc = scheduler->resources; lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *r = (pcmk_resource_t *) lpc->data;
pe__clear_resource_flags_recursive(r, flag);
}
}
void
pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__set_rsc_flags(rsc, flags);
for (GList *gIter = rsc->private->children;
gIter != NULL; gIter = gIter->next) {
pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason,
pcmk_action_t *dependency, pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) {
/* No resources require it */
return;
} else if ((rsc != NULL)
&& !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) {
/* Wasn't a stonith device */
return;
} else if(node
&& node->details->online
&& node->details->unclean == FALSE
&& node->details->shutdown == FALSE) {
pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE,
reason, FALSE, scheduler);
if(dependency) {
order_actions(unfence, dependency, pcmk__ar_ordered);
}
} else if(rsc) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->private->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, scheduler);
}
}
}
}
/*!
* \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 pcmk_node_t *node)
{
const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL,
pcmk__rsc_node_current);
return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches);
}
/*!
* \internal
* \brief Update "recheck by" time in scheduler data
*
* \param[in] recheck Epoch time when recheck should happen
* \param[in,out] scheduler Scheduler data
* \param[in] reason What time is being updated for (for logs)
*/
void
pe__update_recheck_time(time_t recheck, pcmk_scheduler_t *scheduler,
const char *reason)
{
if ((recheck > get_effective_time(scheduler))
&& ((scheduler->recheck_by == 0)
|| (scheduler->recheck_by > recheck))) {
scheduler->recheck_by = recheck;
crm_debug("Updated next scheduler recheck to %s for %s",
pcmk__trim(ctime(&recheck)), reason);
}
}
/*!
* \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] scheduler Scheduler data 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, pcmk_scheduler_t *scheduler)
{
crm_time_t *next_change = crm_time_new_undefined();
pe_eval_nvpairs(scheduler->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, scheduler, "rule evaluation");
}
crm_time_free(next_change);
}
bool
pe__resource_is_disabled(const pcmk_resource_t *rsc)
{
const char *target_role = NULL;
CRM_CHECK(rsc != NULL, return false);
target_role = g_hash_table_lookup(rsc->private->meta,
PCMK_META_TARGET_ROLE);
if (target_role) {
// If invalid, we've already logged an error when unpacking
enum rsc_role_e target_role_e = pcmk_parse_role(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,
pcmk__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 pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return (rsc != NULL) && pcmk__list_of_1(rsc->private->active_nodes)
&& pcmk__same_node((const pcmk_node_t *)
rsc->private->active_nodes->data, node);
}
bool
pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list)
{
if (rsc != NULL) {
for (GList *ele = rsc->private->active_nodes; ele; ele = ele->next) {
pcmk_node_t *node = (pcmk_node_t *) ele->data;
if (pcmk__str_in_list(node->private->name, node_list,
pcmk__str_star_matches|pcmk__str_casei)) {
return true;
}
}
}
return false;
}
bool
pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node)
{
return rsc->private->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) {
pcmk_resource_t *rsc = (pcmk_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->private->parent != NULL)
&& pcmk__str_in_list(rsc_printable_id(rsc->private->parent),
filter, pcmk__str_star_matches))) {
retval = g_list_prepend(retval, rsc);
}
}
return retval;
}
GList *
pe__build_node_name_list(pcmk_scheduler_t *scheduler, 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 {
pcmk_node_t *node = pcmk_find_node(scheduler, 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(scheduler, s);
}
}
return nodes;
}
GList *
pe__build_rsc_list(pcmk_scheduler_t *scheduler, 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;
pcmk_resource_t *rsc = pe_find_resource_with_flags(scheduler->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(scheduler, s);
}
}
return resources;
}
xmlNode *
pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name)
{
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
const char *rsc_id = rsc->id;
if (pcmk__is_clone(parent)) {
rsc_id = pe__clone_child_id(parent);
}
for (xmlNode *xml_op = pcmk__xe_first_child(rsc->private->scheduler->failed,
NULL, NULL, NULL);
xml_op != NULL; xml_op = pcmk__xe_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, PCMK__META_ON_NODE);
if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) {
continue;
}
if (!parse_op_key(pcmk__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;
}