diff --git a/include/crm/common/actions.h b/include/crm/common/actions.h
index 3109e307dd..c0390e868b 100644
--- a/include/crm/common/actions.h
+++ b/include/crm/common/actions.h
@@ -1,338 +1,341 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_ACTIONS__H
#define PCMK__CRM_COMMON_ACTIONS__H
#include <stdbool.h> // bool
#include <strings.h> // strcasecmp()
#include <glib.h> // gboolean, guint
#include <libxml/tree.h> // xmlNode
#include <crm/lrmd_events.h> // lrmd_event_data_t
#ifdef __cplusplus
extern "C" {
#endif
/*!
* \file
* \brief APIs related to actions
* \ingroup core
*/
//! Default timeout (in milliseconds) for non-metadata actions
#define PCMK_DEFAULT_ACTION_TIMEOUT_MS 20000
// @COMPAT We don't need a separate timeout for metadata, much less a longer one
//! \deprecated Default timeout (in milliseconds) for metadata actions
#define PCMK_DEFAULT_METADATA_TIMEOUT_MS 30000
// Action names as strings
#define PCMK_ACTION_CANCEL "cancel"
#define PCMK_ACTION_CLEAR_FAILCOUNT "clear_failcount"
#define PCMK_ACTION_CLONE_ONE_OR_MORE "clone-one-or-more"
#define PCMK_ACTION_DELETE "delete"
#define PCMK_ACTION_DEMOTE "demote"
#define PCMK_ACTION_DEMOTED "demoted"
#define PCMK_ACTION_DO_SHUTDOWN "do_shutdown"
#define PCMK_ACTION_LIST "list"
#define PCMK_ACTION_LRM_DELETE "lrm_delete"
#define PCMK_ACTION_LOAD_STOPPED "load_stopped"
#define PCMK_ACTION_MAINTENANCE_NODES "maintenance_nodes"
#define PCMK_ACTION_META_DATA "meta-data"
#define PCMK_ACTION_MIGRATE_FROM "migrate_from"
#define PCMK_ACTION_MIGRATE_TO "migrate_to"
#define PCMK_ACTION_MONITOR "monitor"
#define PCMK_ACTION_NOTIFIED "notified"
#define PCMK_ACTION_NOTIFY "notify"
#define PCMK_ACTION_OFF "off"
#define PCMK_ACTION_ON "on"
#define PCMK_ACTION_ONE_OR_MORE "one-or-more"
#define PCMK_ACTION_PROMOTE "promote"
#define PCMK_ACTION_PROMOTED "promoted"
#define PCMK_ACTION_REBOOT "reboot"
#define PCMK_ACTION_RELOAD "reload"
#define PCMK_ACTION_RELOAD_AGENT "reload-agent"
#define PCMK_ACTION_RUNNING "running"
#define PCMK_ACTION_START "start"
#define PCMK_ACTION_STATUS "status"
#define PCMK_ACTION_STONITH "stonith"
#define PCMK_ACTION_STOP "stop"
#define PCMK_ACTION_STOPPED "stopped"
#define PCMK_ACTION_VALIDATE_ALL "validate-all"
//! Possible actions (including some pseudo-actions)
enum action_tasks {
pcmk_action_unspecified = 0, //!< Unspecified or unknown action
pcmk_action_monitor, //!< Monitor
// Each "completed" action must be the regular action plus 1
pcmk_action_stop, //!< Stop
pcmk_action_stopped, //!< Stop completed
pcmk_action_start, //!< Start
pcmk_action_started, //!< Start completed
pcmk_action_notify, //!< Notify
pcmk_action_notified, //!< Notify completed
pcmk_action_promote, //!< Promote
pcmk_action_promoted, //!< Promoted
pcmk_action_demote, //!< Demote
pcmk_action_demoted, //!< Demoted
pcmk_action_shutdown, //!< Shut down node
pcmk_action_fence, //!< Fence node
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
//! \deprecated Use pcmk_action_unspecified instead
no_action = pcmk_action_unspecified,
//! \deprecated Use pcmk_action_monitor instead
monitor_rsc = pcmk_action_monitor,
//! \deprecated Use pcmk_action_stop instead
stop_rsc = pcmk_action_stop,
//! \deprecated Use pcmk_action_stopped instead
stopped_rsc = pcmk_action_stopped,
//! \deprecated Use pcmk_action_start instead
start_rsc = pcmk_action_start,
//! \deprecated Use pcmk_action_started instead
started_rsc = pcmk_action_started,
//! \deprecated Use pcmk_action_notify instead
action_notify = pcmk_action_notify,
//! \deprecated Use pcmk_action_notified instead
action_notified = pcmk_action_notified,
//! \deprecated Use pcmk_action_promote instead
action_promote = pcmk_action_promote,
//! \deprecated Use pcmk_action_promoted instead
action_promoted = pcmk_action_promoted,
//! \deprecated Use pcmk_action_demote instead
action_demote = pcmk_action_demote,
//! \deprecated Use pcmk_action_demoted instead
action_demoted = pcmk_action_demoted,
//! \deprecated Use pcmk_action_shutdown instead
shutdown_crm = pcmk_action_shutdown,
//! \deprecated Use pcmk_action_fence instead
stonith_node = pcmk_action_fence,
#endif
};
//! Possible responses to a resource action failure
enum action_fail_response {
/* The order is (partially) significant here; the values from
* pcmk_on_fail_ignore through pcmk_on_fail_fence_node are in order of
* increasing severity.
*
* @COMPAT The values should be ordered and numbered per the "TODO" comments
* below, so all values are in order of severity and there is room for
* future additions, but that would break API compatibility.
* @TODO For now, we just use a function to compare the values specially, but
* at the next compatibility break, we should arrange things
* properly so we can compare with less than and greater than.
*/
// @TODO Define as 10
pcmk_on_fail_ignore = 0, //!< Act as if failure didn't happen
// @TODO Define as 30
pcmk_on_fail_restart = 1, //!< Restart resource
// @TODO Define as 60
pcmk_on_fail_ban = 2, //!< Ban resource from current node
// @TODO Define as 70
pcmk_on_fail_block = 3, //!< Treat resource as unmanaged
// @TODO Define as 80
pcmk_on_fail_stop = 4, //!< Stop resource and leave stopped
// @TODO Define as 90
pcmk_on_fail_standby_node = 5, //!< Put resource's node in standby
// @TODO Define as 100
pcmk_on_fail_fence_node = 6, //!< Fence resource's node
// @COMPAT Values below here are out of desired order for API compatibility
// @TODO Define as 50
pcmk_on_fail_restart_container = 7, //!< Restart resource's container
// @TODO Define as 40
/*!
* Fence the remote node created by the resource if fencing is enabled,
* otherwise attempt to restart the resource (used internally for some
* remote connection failures).
*/
pcmk_on_fail_reset_remote = 8,
// @TODO Define as 20
pcmk_on_fail_demote = 9, //!< Demote if promotable, else stop
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
//! \deprecated Use pcmk_on_fail_ignore instead
action_fail_ignore = pcmk_on_fail_ignore,
//! \deprecated Use pcmk_on_fail_restart instead
action_fail_recover = pcmk_on_fail_restart,
//! \deprecated Use pcmk_on_fail_ban instead
action_fail_migrate = pcmk_on_fail_ban,
//! \deprecated Use pcmk_on_fail_block instead
action_fail_block = pcmk_on_fail_block,
//! \deprecated Use pcmk_on_fail_stop instead
action_fail_stop = pcmk_on_fail_stop,
//! \deprecated Use pcmk_on_fail_standby_node instead
action_fail_standby = pcmk_on_fail_standby_node,
//! \deprecated Use pcmk_on_fail_fence_node instead
action_fail_fence = pcmk_on_fail_fence_node,
//! \deprecated Use pcmk_on_fail_restart_container instead
action_fail_restart_container = pcmk_on_fail_restart_container,
//! \deprecated Use pcmk_on_fail_reset_remote instead
action_fail_reset_remote = pcmk_on_fail_reset_remote,
//! \deprecated Use pcmk_on_fail_demote instead
action_fail_demote = pcmk_on_fail_demote,
#endif
};
//! Action scheduling flags
enum pe_action_flags {
//! No action flags set (compare with equality rather than bit set)
pcmk_no_action_flags = 0,
//! Whether action does not require invoking an agent
pcmk_action_pseudo = (1 << 0),
//! Whether action is runnable
pcmk_action_runnable = (1 << 1),
//! Whether action should not be executed
pcmk_action_optional = (1 << 2),
//! Whether action should be added to transition graph even if optional
pcmk_action_always_in_graph = (1 << 3),
//! Whether operation-specific instance attributes have been unpacked yet
pcmk_action_attrs_evaluated = (1 << 4),
//! Whether action can be related to a live migration
pcmk_action_migratable = (1 << 7),
//! Whether action has been added to transition graph
pcmk_action_added_to_graph = (1 << 8),
//! Whether action is a stop to abort a dangling migration
pcmk_action_migration_abort = (1 << 11),
/*!
* Whether action is an ordering point for minimum required instances
* (used to implement ordering after clones with clone-min configured,
* and ordered sets with require-all=false)
*/
pcmk_action_min_runnable = (1 << 12),
//! Whether action is recurring monitor that must be rescheduled if active
pcmk_action_reschedule = (1 << 13),
+ //! Whether action has already been processed by a recursive procedure
+ pcmk_action_detect_loop = (1 << 14),
+
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
//! \deprecated Use pcmk_action_pseudo instead
pe_action_pseudo = pcmk_action_pseudo,
//! \deprecated Use pcmk_action_runnable instead
pe_action_runnable = pcmk_action_runnable,
//! \deprecated Use pcmk_action_optional instead
pe_action_optional = pcmk_action_optional,
//! \deprecated Use pcmk_action_always_in_graph instead
pe_action_print_always = pcmk_action_always_in_graph,
//! \deprecated Use pcmk_action_attrs_evaluated instead
pe_action_have_node_attrs = pcmk_action_attrs_evaluated,
//! \deprecated Do not use
pe_action_implied_by_stonith = (1 << 6),
//! \deprecated Use pcmk_action_migratable instead
pe_action_migrate_runnable = pcmk_action_migratable,
//! \deprecated Use pcmk_action_added_to_graph instead
pe_action_dumped = pcmk_action_added_to_graph,
//! \deprecated Do not use
pe_action_processed = (1 << 9),
//! \deprecated Do not use
pe_action_clear = (1 << 10),
//! \deprecated Use pcmk_action_migration_abort instead
pe_action_dangle = pcmk_action_migration_abort,
//! \deprecated Use pcmk_action_min_runnable instead
pe_action_requires_any = pcmk_action_min_runnable,
//! \deprecated Use pcmk_action_reschedule instead
pe_action_reschedule = pcmk_action_reschedule,
#endif
- pe_action_tracking = 0x04000,
+ pe_action_tracking = pcmk_action_detect_loop,
pe_action_dedup = 0x08000, //! Internal state tracking when creating graph
pe_action_dc = 0x10000, //! Action may run on DC instead of target
};
// For parsing various action-related string specifications
gboolean parse_op_key(const char *key, char **rsc_id, char **op_type,
guint *interval_ms);
gboolean decode_transition_key(const char *key, char **uuid, int *transition_id,
int *action_id, int *target_rc);
gboolean decode_transition_magic(const char *magic, char **uuid,
int *transition_id, int *action_id,
int *op_status, int *op_rc, int *target_rc);
// @COMPAT Either these shouldn't be in libcrmcommon or lrmd_event_data_t should
int rsc_op_expected_rc(const lrmd_event_data_t *event);
gboolean did_rsc_op_fail(lrmd_event_data_t *event, int target_rc);
bool crm_op_needs_metadata(const char *rsc_class, const char *op);
xmlNode *crm_create_op_xml(xmlNode *parent, const char *prefix,
const char *task, const char *interval_spec,
const char *timeout);
bool pcmk_is_probe(const char *task, guint interval);
bool pcmk_xe_is_probe(const xmlNode *xml_op);
bool pcmk_xe_mask_probe_failure(const xmlNode *xml_op);
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_ACTIONS__H
diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c
index 69e1d4e0fc..c1465207ec 100644
--- a/lib/pacemaker/pcmk_graph_producer.c
+++ b/lib/pacemaker/pcmk_graph_producer.c
@@ -1,1095 +1,1095 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <glib.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), 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->details->uname)
/*!
* \internal
* \brief Add an XML node tag for a specified ID
*
* \param[in] id Node UUID to add
* \param[in,out] xml Parent XML tag to add to
*/
static xmlNode*
add_node_to_xml_by_id(const char *id, xmlNode *xml)
{
xmlNode *node_xml;
node_xml = create_xml_node(xml, XML_CIB_TAG_NODE);
crm_xml_add(node_xml, XML_ATTR_ID, id);
return node_xml;
}
/*!
* \internal
* \brief Add an XML node tag for a specified node
*
* \param[in] node Node to add
* \param[in,out] xml XML to add node to
*/
static void
add_node_to_xml(const pe_node_t *node, void *xml)
{
add_node_to_xml_by_id(node->details->id, (xmlNode *) xml);
}
/*!
* \internal
* \brief Count (optionally add to XML) nodes needing maintenance state update
*
* \param[in,out] xml Parent XML tag to add to, if any
* \param[in] data_set Working set for cluster
*
* \return Count of nodes added
* \note Only Pacemaker Remote nodes are considered currently
*/
static int
add_maintenance_nodes(xmlNode *xml, const pe_working_set_t *data_set)
{
xmlNode *maintenance = NULL;
int count = 0;
if (xml != NULL) {
maintenance = create_xml_node(xml, XML_GRAPH_TAG_MAINTENANCE);
}
for (const GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
const pe_node_t *node = iter->data;
if (pe__is_guest_or_remote_node(node) &&
(node->details->maintenance != node->details->remote_maintenance)) {
if (maintenance != NULL) {
crm_xml_add(add_node_to_xml_by_id(node->details->id,
maintenance),
XML_NODE_IS_MAINTENANCE,
(node->details->maintenance? "1" : "0"));
}
count++;
}
}
crm_trace("%s %d nodes in need of maintenance mode update in state",
((maintenance == NULL)? "Counted" : "Added"), count);
return count;
}
/*!
* \internal
* \brief Add pseudo action with nodes needing maintenance state update
*
* \param[in,out] data_set Working set for cluster
*/
static void
add_maintenance_update(pe_working_set_t *data_set)
{
pe_action_t *action = NULL;
if (add_maintenance_nodes(NULL, data_set) != 0) {
action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, data_set);
pe__set_action_flags(action, 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 pe_action_t *action)
{
CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL),
return);
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) {
/* Shutdown makes the action's node down */
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
/* Fencing makes the action's node and any hosted guest nodes down */
const char *fence = g_hash_table_lookup(action->meta, "stonith_action");
if (pcmk__is_fencing_action(fence)) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
pe_foreach_guest_node(action->node->details->data_set, action->node,
add_node_to_xml, downed);
}
} else if (action->rsc && action->rsc->is_remote_node
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
/* Stopping a remote connection resource makes connected node down,
* unless it's part of a migration
*/
GList *iter;
pe_action_t *input;
bool migrating = false;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((pe_action_wrapper_t *) iter->data)->action;
if ((input->rsc != NULL)
&& pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none)
&& pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
migrating = true;
break;
}
}
if (!migrating) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->rsc->id, downed);
}
}
}
/*!
* \internal
* \brief Create a transition graph operation key for a clone action
*
* \param[in] action Clone action
* \param[in] interval_ms Action interval in milliseconds
*
* \return Newly allocated string with transition graph operation key
*/
static char *
clone_op_key(const pe_action_t *action, guint interval_ms)
{
if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = g_hash_table_lookup(action->meta, "notify_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_operation");
CRM_LOG_ASSERT((n_type != NULL) && (n_task != NULL));
return pcmk__notify_key(action->rsc->clone_name, n_type, n_task);
} else if (action->cancel_task != NULL) {
return pcmk__op_key(action->rsc->clone_name, action->cancel_task,
interval_ms);
} else {
return pcmk__op_key(action->rsc->clone_name, action->task, interval_ms);
}
}
/*!
* \internal
* \brief Add node details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] xml Transition graph action XML for \p action
*/
static void
add_node_details(const pe_action_t *action, xmlNode *xml)
{
pe_node_t *router_node = pcmk__connection_host_for_action(action);
crm_xml_add(xml, XML_LRM_ATTR_TARGET, action->node->details->uname);
crm_xml_add(xml, XML_LRM_ATTR_TARGET_UUID, action->node->details->id);
if (router_node != NULL) {
crm_xml_add(xml, XML_LRM_ATTR_ROUTER_NODE, router_node->details->uname);
}
}
/*!
* \internal
* \brief Add resource details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_resource_details(const pe_action_t *action, xmlNode *action_xml)
{
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER,
XML_ATTR_TYPE
};
/* If a resource is locked to a node via shutdown-lock, mark its actions
* so the controller can preserve the lock when the action completes.
*/
if (pcmk__action_locks_rsc_to_node(action)) {
crm_xml_add_ll(action_xml, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
(long long) action->rsc->lock_time);
}
// List affected resource
rsc_xml = create_xml_node(action_xml,
(const char *) action->rsc->xml->name);
if (pcmk_is_set(action->rsc->flags, pcmk_rsc_removed)
&& (action->rsc->clone_name != NULL)) {
/* Use the numbered instance name here, because if there is more
* than one instance on a node, we need to make sure the command
* goes to the right one.
*
* This is important even for anonymous clones, because the clone's
* unique meta-attribute might have just been toggled from on to
* off.
*/
crm_debug("Using orphan clone name %s instead of %s",
action->rsc->id, action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->clone_name);
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
} else if (!pcmk_is_set(action->rsc->flags, pcmk_rsc_unique)) {
const char *xml_id = ID(action->rsc->xml);
crm_debug("Using anonymous clone name %s for %s (aka %s)",
xml_id, action->rsc->id, action->rsc->clone_name);
/* ID is what we'd like client to use
* ID_LONG is what they might know it as instead
*
* ID_LONG is only strictly needed /here/ during the
* transition period until all nodes in the cluster
* are running the new software /and/ have rebooted
* once (meaning that they've only ever spoken to a DC
* supporting this feature).
*
* If anyone toggles the unique flag to 'on', the
* 'instance free' name will correspond to an orphan
* and fall into the clause above instead
*/
crm_xml_add(rsc_xml, XML_ATTR_ID, xml_id);
if ((action->rsc->clone_name != NULL)
&& !pcmk__str_eq(xml_id, action->rsc->clone_name,
pcmk__str_none)) {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->clone_name);
} else {
crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id);
}
} else {
CRM_ASSERT(action->rsc->clone_name == NULL);
crm_xml_add(rsc_xml, XML_ATTR_ID, action->rsc->id);
}
for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) {
crm_xml_add(rsc_xml, attr_list[lpc],
g_hash_table_lookup(action->rsc->meta, attr_list[lpc]));
}
}
/*!
* \internal
* \brief Add action attributes to transition graph action XML
*
* \param[in,out] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_action_attributes(pe_action_t *action, xmlNode *action_xml)
{
xmlNode *args_xml = NULL;
/* We create free-standing XML to start, so we can sort the attributes
* before adding it to action_xml, which keeps the scheduler regression
* test graphs comparable.
*/
args_xml = create_xml_node(NULL, XML_TAG_ATTRS);
crm_xml_add(args_xml, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
g_hash_table_foreach(action->extra, hash2field, args_xml);
if ((action->rsc != NULL) && (action->node != NULL)) {
// Get the resource instance attributes, evaluated properly for node
GHashTable *params = pe_rsc_params(action->rsc, action->node,
action->rsc->cluster);
pcmk__substitute_remote_addr(action->rsc, params);
g_hash_table_foreach(params, hash2smartfield, args_xml);
} else if ((action->rsc != NULL)
&& (action->rsc->variant <= pcmk_rsc_variant_primitive)) {
GHashTable *params = pe_rsc_params(action->rsc, NULL,
action->rsc->cluster);
g_hash_table_foreach(params, hash2smartfield, args_xml);
}
g_hash_table_foreach(action->meta, hash2metafield, args_xml);
if (action->rsc != NULL) {
pe_resource_t *parent = action->rsc;
while (parent != NULL) {
parent->cmds->add_graph_meta(parent, args_xml);
parent = parent->parent;
}
pcmk__add_bundle_meta_to_xml(args_xml, action);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)
&& (action->node != NULL)) {
/* Pass the node's attributes as meta-attributes.
*
* @TODO: Determine whether it is still necessary to do this. It was
* added in 33d99707, probably for the libfence-based implementation in
* c9a90bd, which is no longer used.
*/
g_hash_table_foreach(action->node->details->attrs, hash2metafield,
args_xml);
}
sorted_xml(args_xml, action_xml, FALSE);
free_xml(args_xml);
}
/*!
* \internal
* \brief Create the transition graph XML for a scheduled action
*
* \param[in,out] parent Parent XML element to add action to
* \param[in,out] action Scheduled action
* \param[in] skip_details If false, add action details as sub-elements
* \param[in] data_set Cluster working set
*/
static void
create_graph_action(xmlNode *parent, pe_action_t *action, bool skip_details,
const pe_working_set_t *data_set)
{
bool needs_node_info = true;
bool needs_maintenance_info = false;
xmlNode *action_xml = NULL;
if ((action == NULL) || (data_set == NULL)) {
return;
}
// Create the top-level element based on task
if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) {
/* All fences need node info; guest node fences are pseudo-events */
if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT);
} else {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
}
} else if (pcmk__str_any_of(action->task,
PCMK_ACTION_DO_SHUTDOWN,
PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE,
pcmk__str_none)) {
// CIB-only clean-up for shutdown locks
action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT);
crm_xml_add(action_xml, PCMK__XA_MODE, XML_TAG_CIB);
} else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES,
pcmk__str_none)) {
needs_maintenance_info = true;
}
action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT);
needs_node_info = false;
} else {
action_xml = create_xml_node(parent, XML_GRAPH_TAG_RSC_OP);
}
crm_xml_add_int(action_xml, XML_ATTR_ID, action->id);
crm_xml_add(action_xml, XML_LRM_ATTR_TASK, action->task);
if ((action->rsc != NULL) && (action->rsc->clone_name != NULL)) {
char *clone_key = NULL;
guint interval_ms;
if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
clone_key = clone_op_key(action, interval_ms);
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, clone_key);
crm_xml_add(action_xml, "internal_" XML_LRM_ATTR_TASK_KEY,
action->uuid);
free(clone_key);
} else {
crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, action->uuid);
}
if (needs_node_info && (action->node != NULL)) {
add_node_details(action, action_xml);
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET),
strdup(action->node->details->uname));
g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET_UUID),
strdup(action->node->details->id));
}
if (skip_details) {
return;
}
if ((action->rsc != NULL)
&& !pcmk_is_set(action->flags, 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, data_set);
}
}
/*!
* \internal
* \brief Check whether an action should be added to the transition graph
*
* \param[in] action Action to check
*
* \return true if action should be added to graph, otherwise false
*/
static bool
should_add_action_to_graph(const pe_action_t *action)
{
if (!pcmk_is_set(action->flags, 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,
XML_LRM_ATTR_INTERVAL_MS);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) {
crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)",
action->uuid, action->id, action->rsc->id);
return false;
}
}
/* Always add pseudo-actions, fence actions, and shutdown actions (already
* determined to be required and runnable by this point)
*/
if (pcmk_is_set(action->flags, pcmk_action_pseudo)
|| pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH,
PCMK_ACTION_DO_SHUTDOWN, NULL)) {
return true;
}
if (action->node == NULL) {
pe_err("Skipping action %s (%d) "
"because it was not assigned to a node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unassigned", action, false);
return false;
}
if (pcmk_is_set(action->flags, pe_action_dc)) {
crm_trace("Action %s (%d) should be dumped: "
"can run on DC instead of %s",
action->uuid, action->id, pe__node_name(action->node));
} else if (pe__is_guest_node(action->node)
&& !action->node->details->remote_requires_reset) {
crm_trace("Action %s (%d) should be dumped: "
"assuming will be runnable on guest %s",
action->uuid, action->id, pe__node_name(action->node));
} else if (!action->node->details->online) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for offline node (bug?)",
action->uuid, action->id);
pcmk__log_action("Offline node", action, false);
return false;
} else if (action->node->details->unclean) {
pe_err("Skipping action %s (%d) "
"because it was scheduled for unclean node (bug?)",
action->uuid, action->id);
pcmk__log_action("Unclean node", action, false);
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether an ordering's flags can change an action
*
* \param[in] ordering Ordering to check
*
* \return true if ordering has flags that can change an action, false otherwise
*/
static bool
ordering_can_change_actions(const pe_action_wrapper_t *ordering)
{
return pcmk_any_flags_set(ordering->type, ~(pe_order_implies_first_printed
|pe_order_implies_then_printed
|pe_order_optional));
}
/*!
* \internal
* \brief Check whether an action input should be in the transition graph
*
* \param[in] action Action to check
* \param[in,out] input Action input to check
*
* \return true if input should be in graph, false otherwise
* \note This function may not only check an input, but disable it under certian
* circumstances (load or anti-colocation orderings that are not needed).
*/
static bool
should_add_input_to_graph(const pe_action_t *action, pe_action_wrapper_t *input)
{
if (input->state == pe_link_dumped) {
return true;
}
if (input->type == pe_order_none) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"ordering disabled",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, 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->type, pe_order_one_or_more)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"one-or-more and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_implies_first_migratable)
&& !pcmk_is_set(input->action->flags, pcmk_action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"implies input migratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_apply_first_non_migratable)
&& pcmk_is_set(input->action->flags, pcmk_action_migratable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"only if input unmigratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->type == pe_order_optional)
&& pcmk_is_set(input->action->flags, 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->type == pe_order_load) {
pe_node_t *input_node = input->action->node;
// load orderings are relevant only if actions are for same node
if ((action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
pe_node_t *assigned = action->rsc->allocated_to;
/* For load_stopped -> migrate_to orderings, we care about where it
* has been assigned to, not where it will be executed.
*/
if (!pe__same_node(input_node, assigned)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(assigned? assigned->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
}
} else if (!pe__same_node(input_node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->type == pe_order_anti_colocation) {
if (input->action->node && action->node
&& !pe__same_node(input->action->node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
pe__node_name(action->node),
pe__node_name(input->action->node));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& pcmk_is_set(input->action->rsc->flags, pcmk_rsc_failed)
&& !pcmk_is_set(input->action->rsc->flags, pcmk_rsc_managed)
&& pcmk__ends_with(input->action->uuid, "_stop_0")
&& action->rsc && pe_rsc_is_clone(action->rsc)) {
crm_warn("Ignoring requirement that %s complete before %s:"
" unmanaged failed resources cannot prevent clone shutdown",
input->action->uuid, action->uuid);
return false;
} else if (pcmk_is_set(input->action->flags, 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->type);
return true;
}
/*!
* \internal
* \brief Check whether an ordering creates an ordering loop
*
* \param[in] init_action "First" action in ordering
* \param[in] action Callers should always set this the same as
* \p init_action (this function may use a different
* value for recursive calls)
* \param[in,out] input Action wrapper for "then" action in ordering
*
* \return true if the ordering creates a loop, otherwise false
*/
bool
pcmk__graph_has_loop(const pe_action_t *init_action, const pe_action_t *action,
pe_action_wrapper_t *input)
{
bool has_loop = false;
- if (pcmk_is_set(input->action->flags, pe_action_tracking)) {
+ 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->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
return false;
}
// Don't need to check inputs that won't be used
if (!should_add_input_to_graph(action, input)) {
return false;
}
if (input->action == init_action) {
crm_debug("Input loop found in %s@%s ->...-> %s@%s",
action->uuid,
action->node? action->node->details->uname : "",
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
return true;
}
- pe__set_action_flags(input->action, pe_action_tracking);
+ pe__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->details->uname : "",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
input->type,
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
// Recursively check input itself for loops
for (GList *iter = input->action->actions_before;
iter != NULL; iter = iter->next) {
if (pcmk__graph_has_loop(init_action, input->action,
(pe_action_wrapper_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
break;
}
}
- pe__clear_action_flags(input->action, pe_action_tracking);
+ pe__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->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
}
return has_loop;
}
/*!
* \internal
* \brief Create a synapse XML element for a transition graph
*
* \param[in] action Action that synapse is for
* \param[in,out] data_set Cluster working set containing graph
*
* \return Newly added XML element for new graph synapse
*/
static xmlNode *
create_graph_synapse(const pe_action_t *action, pe_working_set_t *data_set)
{
int synapse_priority = 0;
xmlNode *syn = create_xml_node(data_set->graph, "synapse");
crm_xml_add_int(syn, XML_ATTR_ID, data_set->num_synapse);
data_set->num_synapse++;
if (action->rsc != NULL) {
synapse_priority = action->rsc->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, XML_CIB_ATTR_PRIORITY, synapse_priority);
}
return syn;
}
/*!
* \internal
* \brief Add an action to the transition graph XML if appropriate
*
* \param[in,out] data Action to possibly add
* \param[in,out] user_data Cluster working set
*
* \note This will de-duplicate the action inputs, meaning that the
* pe_action_wrapper_t:type flags can no longer be relied on to retain
* their original settings. That means this MUST be called after
* pcmk__apply_orderings() is complete, and nothing after this should rely
* on those type flags. (For example, some code looks for type equal to
* some flag rather than whether the flag is set, and some code looks for
* particular combinations of flags -- such code must be done before
* pcmk__create_graph().)
*/
static void
add_action_to_graph(gpointer data, gpointer user_data)
{
pe_action_t *action = (pe_action_t *) data;
pe_working_set_t *data_set = (pe_working_set_t *) user_data;
xmlNode *syn = NULL;
xmlNode *set = NULL;
xmlNode *in = NULL;
/* If we haven't already, de-duplicate inputs (even if we won't be adding
* the action to the graph, so that crm_simulate's dot graphs don't have
* duplicates).
*/
if (!pcmk_is_set(action->flags, pe_action_dedup)) {
pcmk__deduplicate_action_inputs(action);
pe__set_action_flags(action, pe_action_dedup);
}
if (pcmk_is_set(action->flags, pcmk_action_added_to_graph)
|| !should_add_action_to_graph(action)) {
return; // Already added, or shouldn't be
}
pe__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->details->uname));
syn = create_graph_synapse(action, data_set);
set = create_xml_node(syn, "action_set");
in = create_xml_node(syn, "inputs");
create_graph_action(set, action, false, data_set);
for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) lpc->data;
if (should_add_input_to_graph(action, input)) {
xmlNode *input_xml = create_xml_node(in, "trigger");
input->state = pe_link_dumped;
create_graph_action(input_xml, input->action, true, data_set);
}
}
}
static int transition_id = -1;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const char *filename)
{
if (was_processing_error) {
crm_err("Calculated transition %d (with errors)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else if (was_processing_warning) {
crm_warn("Calculated transition %d (with warnings)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else {
crm_notice("Calculated transition %d%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
}
if (crm_config_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*!
* \internal
* \brief Add a resource's actions to the transition graph
*
* \param[in,out] rsc Resource whose actions should be added
*/
void
pcmk__add_rsc_actions_to_graph(pe_resource_t *rsc)
{
GList *iter = NULL;
CRM_ASSERT(rsc != NULL);
pe_rsc_trace(rsc, "Adding actions for %s to graph", rsc->id);
// First add the resource's own actions
g_list_foreach(rsc->actions, add_action_to_graph, rsc->cluster);
// Then recursively add its children's actions (appropriate to variant)
for (iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) iter->data;
child_rsc->cmds->add_actions_to_graph(child_rsc);
}
}
/*!
* \internal
* \brief Create a transition graph with all cluster actions needed
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__create_graph(pe_working_set_t *data_set)
{
GList *iter = NULL;
const char *value = NULL;
long long limit = 0LL;
transition_id++;
crm_trace("Creating transition graph %d", transition_id);
data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
value = pe_pref(data_set->config_hash, "cluster-delay");
crm_xml_add(data_set->graph, "cluster-delay", value);
value = pe_pref(data_set->config_hash, "stonith-timeout");
crm_xml_add(data_set->graph, "stonith-timeout", value);
crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY");
if (pcmk_is_set(data_set->flags, pcmk_sched_start_failure_fatal)) {
crm_xml_add(data_set->graph, "failed-start-offset", "INFINITY");
} else {
crm_xml_add(data_set->graph, "failed-start-offset", "1");
}
value = pe_pref(data_set->config_hash, "batch-limit");
crm_xml_add(data_set->graph, "batch-limit", value);
crm_xml_add_int(data_set->graph, "transition_id", transition_id);
value = pe_pref(data_set->config_hash, "migration-limit");
if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) {
crm_xml_add(data_set->graph, "migration-limit", value);
}
if (data_set->recheck_by > 0) {
char *recheck_epoch = NULL;
recheck_epoch = crm_strdup_printf("%llu",
(long long) data_set->recheck_by);
crm_xml_add(data_set->graph, "recheck-by", recheck_epoch);
free(recheck_epoch);
}
/* The following code will de-duplicate action inputs, so nothing past this
* should rely on the action input type flags retaining their original
* values.
*/
// Add resource actions to graph
for (iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
pe_rsc_trace(rsc, "Processing actions for %s", rsc->id);
rsc->cmds->add_actions_to_graph(rsc);
}
// Add pseudo-action for list of nodes with maintenance state update
add_maintenance_update(data_set);
// Add non-resource (node) actions
for (iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = (pe_action_t *) iter->data;
if ((action->rsc != NULL)
&& (action->node != NULL)
&& action->node->details->shutdown
&& !pcmk_is_set(action->rsc->flags, 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(data_set->flags, pcmk_sched_quorate)
|| (data_set->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",
pe__node_name(action->node), action->rsc->id,
(managed? " blocked" : " unmanaged"),
(failed? " failed" : ""), action->uuid);
}
}
add_action_to_graph((gpointer) action, (gpointer) data_set);
}
crm_log_xml_trace(data_set->graph, "graph");
}
diff --git a/lib/pacemaker/pcmk_sched_probes.c b/lib/pacemaker/pcmk_sched_probes.c
index 821ab63972..f452e06381 100644
--- a/lib/pacemaker/pcmk_sched_probes.c
+++ b/lib/pacemaker/pcmk_sched_probes.c
@@ -1,903 +1,903 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <crm/crm.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Add the expected result to a newly created probe
*
* \param[in,out] probe Probe action to add expected result to
* \param[in] rsc Resource that probe is for
* \param[in] node Node that probe will run on
*/
static void
add_expected_result(pe_action_t *probe, const pe_resource_t *rsc,
const pe_node_t *node)
{
// Check whether resource is currently active on node
pe_node_t *running = pe_find_node_id(rsc->running_on, node->details->id);
// The expected result is what we think the resource's current state is
if (running == NULL) {
pe__add_action_expected_result(probe, CRM_EX_NOT_RUNNING);
} else if (rsc->role == pcmk_role_promoted) {
pe__add_action_expected_result(probe, CRM_EX_PROMOTED);
}
}
/*!
* \internal
* \brief Create any needed robes on a node for a list of resources
*
* \param[in,out] rscs List of resources to create probes for
* \param[in,out] node Node to create probes on
*
* \return true if any probe was created, otherwise false
*/
bool
pcmk__probe_resource_list(GList *rscs, pe_node_t *node)
{
bool any_created = false;
for (GList *iter = rscs; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (rsc->cmds->create_probe(rsc, node)) {
any_created = true;
}
}
return any_created;
}
/*!
* \internal
* \brief Order one resource's start after another's start-up probe
*
* \param[in,out] rsc1 Resource that might get start-up probe
* \param[in] rsc2 Resource that might be started
*/
static void
probe_then_start(pe_resource_t *rsc1, pe_resource_t *rsc2)
{
if ((rsc1->allocated_to != NULL)
&& (g_hash_table_lookup(rsc1->known_on,
rsc1->allocated_to->details->id) == NULL)) {
pcmk__new_ordering(rsc1,
pcmk__op_key(rsc1->id, PCMK_ACTION_MONITOR, 0),
NULL,
rsc2, pcmk__op_key(rsc2->id, PCMK_ACTION_START, 0),
NULL,
pe_order_optional, rsc1->cluster);
}
}
/*!
* \internal
* \brief Check whether a guest resource will stop
*
* \param[in] node Guest node to check
*
* \return true if guest resource will likely stop, otherwise false
*/
static bool
guest_resource_will_stop(const pe_node_t *node)
{
const pe_resource_t *guest_rsc = node->details->remote_rsc->container;
/* Ideally, we'd check whether the guest has a required stop, but that
* information doesn't exist yet, so approximate it ...
*/
return node->details->remote_requires_reset
|| node->details->unclean
|| pcmk_is_set(guest_rsc->flags, pcmk_rsc_failed)
|| (guest_rsc->next_role == pcmk_role_stopped)
// Guest is moving
|| ((guest_rsc->role > pcmk_role_stopped)
&& (guest_rsc->allocated_to != NULL)
&& (pe_find_node(guest_rsc->running_on,
guest_rsc->allocated_to->details->uname) == NULL));
}
/*!
* \internal
* \brief Create a probe action for a resource on a node
*
* \param[in,out] rsc Resource to create probe for
* \param[in,out] node Node to create probe on
*
* \return Newly created probe action
*/
static pe_action_t *
probe_action(pe_resource_t *rsc, pe_node_t *node)
{
pe_action_t *probe = NULL;
char *key = pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0);
crm_debug("Scheduling probe of %s %s on %s",
role2text(rsc->role), rsc->id, pe__node_name(node));
probe = custom_action(rsc, key, PCMK_ACTION_MONITOR, node, FALSE, TRUE,
rsc->cluster);
pe__clear_action_flags(probe, pcmk_action_optional);
pcmk__order_vs_unfence(rsc, node, probe, pe_order_optional);
add_expected_result(probe, rsc, node);
return probe;
}
/*!
* \internal
* \brief Create probes for a resource on a node, if needed
*
* \brief Schedule any probes needed for a resource on a node
*
* \param[in,out] rsc Resource to create probe for
* \param[in,out] node Node to create probe on
*
* \return true if any probe was created, otherwise false
*/
bool
pcmk__probe_rsc_on_node(pe_resource_t *rsc, pe_node_t *node)
{
uint32_t flags = pe_order_optional;
pe_action_t *probe = NULL;
pe_node_t *allowed = NULL;
pe_resource_t *top = uber_parent(rsc);
const char *reason = NULL;
CRM_ASSERT((rsc != NULL) && (node != NULL));
if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_probe_resources)) {
reason = "start-up probes are disabled";
goto no_probe;
}
if (pe__is_guest_or_remote_node(node)) {
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_none)) {
reason = "Pacemaker Remote nodes cannot run stonith agents";
goto no_probe;
} else if (pe__is_guest_node(node)
&& pe__resource_contains_guest_node(rsc->cluster, rsc)) {
reason = "guest nodes cannot run resources containing guest nodes";
goto no_probe;
} else if (rsc->is_remote_node) {
reason = "Pacemaker Remote nodes cannot host remote connections";
goto no_probe;
}
}
// If this is a collective resource, probes are created for its children
if (rsc->children != NULL) {
return pcmk__probe_resource_list(rsc->children, node);
}
if ((rsc->container != NULL) && !rsc->is_remote_node) {
reason = "resource is inside a container";
goto no_probe;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
reason = "resource is orphaned";
goto no_probe;
} else if (g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) {
reason = "resource state is already known";
goto no_probe;
}
allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (rsc->exclusive_discover || top->exclusive_discover) {
// Exclusive discovery is enabled ...
if (allowed == NULL) {
// ... but this node is not allowed to run the resource
reason = "resource has exclusive discovery but is not allowed "
"on node";
goto no_probe;
} else if (allowed->rsc_discover_mode != pe_discover_exclusive) {
// ... but no constraint marks this node for discovery of resource
reason = "resource has exclusive discovery but is not enabled "
"on node";
goto no_probe;
}
}
if (allowed == NULL) {
allowed = node;
}
if (allowed->rsc_discover_mode == pe_discover_never) {
reason = "node has discovery disabled";
goto no_probe;
}
if (pe__is_guest_node(node)) {
pe_resource_t *guest = node->details->remote_rsc->container;
if (guest->role == pcmk_role_stopped) {
// The guest is stopped, so we know no resource is active there
reason = "node's guest is stopped";
probe_then_start(guest, top);
goto no_probe;
} else if (guest_resource_will_stop(node)) {
reason = "node's guest will stop";
// Order resource start after guest stop (in case it's restarting)
pcmk__new_ordering(guest,
pcmk__op_key(guest->id, PCMK_ACTION_STOP, 0),
NULL, top,
pcmk__op_key(top->id, PCMK_ACTION_START, 0),
NULL, pe_order_optional, rsc->cluster);
goto no_probe;
}
}
// We've eliminated all cases where a probe is not needed, so now it is
probe = probe_action(rsc, node);
/* Below, we will order the probe relative to start or reload. If this is a
* clone instance, the start or reload is for the entire clone rather than
* just the instance. Otherwise, the start or reload is for the resource
* itself.
*/
if (!pe_rsc_is_clone(top)) {
top = rsc;
}
/* Prevent a start if the resource can't be probed, but don't cause the
* resource or entire clone to stop if already active.
*/
if (!pcmk_is_set(probe->flags, pcmk_action_runnable)
&& (top->running_on == NULL)) {
pe__set_order_flags(flags, pe_order_runnable_left);
}
// Start or reload after probing the resource
pcmk__new_ordering(rsc, NULL, probe,
top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL,
flags, rsc->cluster);
pcmk__new_ordering(rsc, NULL, probe, top, reload_key(rsc), NULL,
pe_order_optional, rsc->cluster);
return true;
no_probe:
pe_rsc_trace(rsc,
"Skipping probe for %s on %s because %s",
rsc->id, node->details->id, reason);
return false;
}
/*!
* \internal
* \brief Check whether a probe should be ordered before another action
*
* \param[in] probe Probe action to check
* \param[in] then Other action to check
*
* \return true if \p probe should be ordered before \p then, otherwise false
*/
static bool
probe_needed_before_action(const pe_action_t *probe, const pe_action_t *then)
{
// Probes on a node are performed after unfencing it, not before
if (pcmk__str_eq(then->task, PCMK_ACTION_STONITH, pcmk__str_none)
&& pe__same_node(probe->node, then->node)) {
const char *op = g_hash_table_lookup(then->meta, "stonith_action");
if (pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) {
return false;
}
}
// Probes should be done on a node before shutting it down
if (pcmk__str_eq(then->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)
&& (probe->node != NULL) && (then->node != NULL)
&& !pe__same_node(probe->node, then->node)) {
return false;
}
// Otherwise probes should always be done before any other action
return true;
}
/*!
* \internal
* \brief Add implicit "probe then X" orderings for "stop then X" orderings
*
* If the state of a resource is not known yet, a probe will be scheduled,
* expecting a "not running" result. If the probe fails, a stop will not be
* scheduled until the next transition. Thus, if there are ordering constraints
* like "stop this resource then do something else that's not for the same
* resource", add implicit "probe this resource then do something" equivalents
* so the relation is upheld until we know whether a stop is needed.
*
* \param[in,out] data_set Cluster working set
*/
static void
add_probe_orderings_for_stops(pe_working_set_t *data_set)
{
for (GList *iter = data_set->ordering_constraints; iter != NULL;
iter = iter->next) {
pe__ordering_t *order = iter->data;
uint32_t order_flags = pe_order_optional;
GList *probes = NULL;
GList *then_actions = NULL;
pe_action_t *first = NULL;
pe_action_t *then = NULL;
// Skip disabled orderings
if (order->flags == pe_order_none) {
continue;
}
// Skip non-resource orderings, and orderings for the same resource
if ((order->lh_rsc == NULL) || (order->lh_rsc == order->rh_rsc)) {
continue;
}
// Skip invalid orderings (shouldn't be possible)
first = order->lh_action;
then = order->rh_action;
if (((first == NULL) && (order->lh_action_task == NULL))
|| ((then == NULL) && (order->rh_action_task == NULL))) {
continue;
}
// Skip orderings for first actions other than stop
if ((first != NULL) && !pcmk__str_eq(first->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
continue;
} else if ((first == NULL)
&& !pcmk__ends_with(order->lh_action_task,
"_" PCMK_ACTION_STOP "_0")) {
continue;
}
/* Do not imply a probe ordering for a resource inside of a stopping
* container. Otherwise, it might introduce a transition loop, since a
* probe could be scheduled after the container starts again.
*/
if ((order->rh_rsc != NULL)
&& (order->lh_rsc->container == order->rh_rsc)) {
if ((then != NULL) && pcmk__str_eq(then->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
continue;
} else if ((then == NULL)
&& pcmk__ends_with(order->rh_action_task,
"_" PCMK_ACTION_STOP "_0")) {
continue;
}
}
// Preserve certain order options for future filtering
if (pcmk_is_set(order->flags, pe_order_apply_first_non_migratable)) {
pe__set_order_flags(order_flags,
pe_order_apply_first_non_migratable);
}
if (pcmk_is_set(order->flags, pe_order_same_node)) {
pe__set_order_flags(order_flags, pe_order_same_node);
}
// Preserve certain order types for future filtering
if ((order->flags == pe_order_anti_colocation)
|| (order->flags == pe_order_load)) {
order_flags = order->flags;
}
// List all scheduled probes for the first resource
probes = pe__resource_actions(order->lh_rsc, NULL, PCMK_ACTION_MONITOR,
FALSE);
if (probes == NULL) { // There aren't any
continue;
}
// List all relevant "then" actions
if (then != NULL) {
then_actions = g_list_prepend(NULL, then);
} else if (order->rh_rsc != NULL) {
then_actions = find_actions(order->rh_rsc->actions,
order->rh_action_task, NULL);
if (then_actions == NULL) { // There aren't any
g_list_free(probes);
continue;
}
}
crm_trace("Implying 'probe then' orderings for '%s then %s' "
"(id=%d, type=%.6x)",
((first == NULL)? order->lh_action_task : first->uuid),
((then == NULL)? order->rh_action_task : then->uuid),
order->id, order->flags);
for (GList *probe_iter = probes; probe_iter != NULL;
probe_iter = probe_iter->next) {
pe_action_t *probe = (pe_action_t *) probe_iter->data;
for (GList *then_iter = then_actions; then_iter != NULL;
then_iter = then_iter->next) {
pe_action_t *then = (pe_action_t *) then_iter->data;
if (probe_needed_before_action(probe, then)) {
order_actions(probe, then, order_flags);
}
}
}
g_list_free(then_actions);
g_list_free(probes);
}
}
/*!
* \internal
* \brief Add necessary orderings between probe and starts of clone instances
*
* , in additon to the ordering with the parent resource added upon creating
* the probe.
*
* \param[in,out] probe Probe as 'first' action in an ordering
* \param[in,out] after 'then' action wrapper in the ordering
*/
static void
add_start_orderings_for_probe(pe_action_t *probe, pe_action_wrapper_t *after)
{
uint32_t flags = pe_order_optional|pe_order_runnable_left;
/* Although the ordering between the probe of the clone instance and the
* start of its parent has been added in pcmk__probe_rsc_on_node(), we
* avoided enforcing `pe_order_runnable_left` order type for that as long as
* any of the clone instances are running to prevent them from being
* unexpectedly stopped.
*
* On the other hand, we still need to prevent any inactive instances from
* starting unless the probe is runnable so that we don't risk starting too
* many instances before we know the state on all nodes.
*/
if ((after->action->rsc->variant <= pcmk_rsc_variant_group)
|| pcmk_is_set(probe->flags, pcmk_action_runnable)
// The order type is already enforced for its parent.
|| pcmk_is_set(after->type, pe_order_runnable_left)
|| (pe__const_top_resource(probe->rsc, false) != after->action->rsc)
|| !pcmk__str_eq(after->action->task, PCMK_ACTION_START,
pcmk__str_none)) {
return;
}
crm_trace("Adding probe start orderings for 'unrunnable %s@%s "
"then instances of %s@%s'",
probe->uuid, pe__node_name(probe->node),
after->action->uuid, pe__node_name(after->action->node));
for (GList *then_iter = after->action->actions_after; then_iter != NULL;
then_iter = then_iter->next) {
pe_action_wrapper_t *then = (pe_action_wrapper_t *) then_iter->data;
if (then->action->rsc->running_on
|| (pe__const_top_resource(then->action->rsc, false)
!= after->action->rsc)
|| !pcmk__str_eq(then->action->task, PCMK_ACTION_START,
pcmk__str_none)) {
continue;
}
crm_trace("Adding probe start ordering for 'unrunnable %s@%s "
"then %s@%s' (type=%#.6x)",
probe->uuid, pe__node_name(probe->node),
then->action->uuid, pe__node_name(then->action->node), flags);
/* Prevent the instance from starting if the instance can't, but don't
* cause any other intances to stop if already active.
*/
order_actions(probe, then->action, flags);
}
return;
}
/*!
* \internal
* \brief Order probes before restarts and re-promotes
*
* If a given ordering is a "probe then start" or "probe then promote" ordering,
* add an implicit "probe then stop/demote" ordering in case the action is part
* of a restart/re-promote, and do the same recursively for all actions ordered
* after the "then" action.
*
* \param[in,out] probe Probe as 'first' action in an ordering
* \param[in,out] after 'then' action in the ordering
*/
static void
add_restart_orderings_for_probe(pe_action_t *probe, pe_action_t *after)
{
GList *iter = NULL;
bool interleave = false;
pe_resource_t *compatible_rsc = NULL;
// Validate that this is a resource probe followed by some action
if ((after == NULL) || (probe == NULL) || (probe->rsc == NULL)
|| (probe->rsc->variant != pcmk_rsc_variant_primitive)
|| !pcmk__str_eq(probe->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
return;
}
// Avoid running into any possible loop
- if (pcmk_is_set(after->flags, pe_action_tracking)) {
+ if (pcmk_is_set(after->flags, pcmk_action_detect_loop)) {
return;
}
- pe__set_action_flags(after, pe_action_tracking);
+ pe__set_action_flags(after, pcmk_action_detect_loop);
crm_trace("Adding probe restart orderings for '%s@%s then %s@%s'",
probe->uuid, pe__node_name(probe->node),
after->uuid, pe__node_name(after->node));
/* Add restart orderings if "then" is for a different primitive.
* Orderings for collective resources will be added later.
*/
if ((after->rsc != NULL)
&& (after->rsc->variant == pcmk_rsc_variant_primitive)
&& (probe->rsc != after->rsc)) {
GList *then_actions = NULL;
if (pcmk__str_eq(after->task, PCMK_ACTION_START, pcmk__str_none)) {
then_actions = pe__resource_actions(after->rsc, NULL,
PCMK_ACTION_STOP, FALSE);
} else if (pcmk__str_eq(after->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
then_actions = pe__resource_actions(after->rsc, NULL,
PCMK_ACTION_DEMOTE, FALSE);
}
for (iter = then_actions; iter != NULL; iter = iter->next) {
pe_action_t *then = (pe_action_t *) iter->data;
// Skip pseudo-actions (for example, those implied by fencing)
if (!pcmk_is_set(then->flags, pcmk_action_pseudo)) {
order_actions(probe, then, pe_order_optional);
}
}
g_list_free(then_actions);
}
/* Detect whether "then" is an interleaved clone action. For these, we want
* to add orderings only for the relevant instance.
*/
if ((after->rsc != NULL)
&& (after->rsc->variant > pcmk_rsc_variant_group)) {
const char *interleave_s = g_hash_table_lookup(after->rsc->meta,
XML_RSC_ATTR_INTERLEAVE);
interleave = crm_is_true(interleave_s);
if (interleave) {
compatible_rsc = pcmk__find_compatible_instance(probe->rsc,
after->rsc,
pcmk_role_unknown,
false);
}
}
/* Now recursively do the same for all actions ordered after "then". This
* also handles collective resources since the collective action will be
* ordered before its individual instances' actions.
*/
for (iter = after->actions_after; iter != NULL; iter = iter->next) {
pe_action_wrapper_t *after_wrapper = (pe_action_wrapper_t *) iter->data;
/* pe_order_implies_then is the reason why a required A.start
* implies/enforces B.start to be required too, which is the cause of
* B.restart/re-promote.
*
* Not sure about pe_order_implies_then_on_node though. It's now only
* used for unfencing case, which tends to introduce transition
* loops...
*/
if (!pcmk_is_set(after_wrapper->type, pe_order_implies_then)) {
/* The order type between a group/clone and its child such as
* B.start-> B_child.start is:
* pe_order_implies_first_printed | pe_order_runnable_left
*
* Proceed through the ordering chain and build dependencies with
* its children.
*/
if ((after->rsc == NULL)
|| (after->rsc->variant < pcmk_rsc_variant_group)
|| (probe->rsc->parent == after->rsc)
|| (after_wrapper->action->rsc == NULL)
|| (after_wrapper->action->rsc->variant > pcmk_rsc_variant_group)
|| (after->rsc != after_wrapper->action->rsc->parent)) {
continue;
}
/* Proceed to the children of a group or a non-interleaved clone.
* For an interleaved clone, proceed only to the relevant child.
*/
if ((after->rsc->variant > pcmk_rsc_variant_group) && interleave
&& ((compatible_rsc == NULL)
|| (compatible_rsc != after_wrapper->action->rsc))) {
continue;
}
}
crm_trace("Recursively adding probe restart orderings for "
"'%s@%s then %s@%s' (type=%#.6x)",
after->uuid, pe__node_name(after->node),
after_wrapper->action->uuid,
pe__node_name(after_wrapper->action->node),
after_wrapper->type);
add_restart_orderings_for_probe(probe, after_wrapper->action);
}
}
/*!
* \internal
* \brief Clear the tracking flag on all scheduled actions
*
* \param[in,out] data_set Cluster working set
*/
static void
clear_actions_tracking_flag(pe_working_set_t *data_set)
{
for (GList *iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = iter->data;
- pe__clear_action_flags(action, pe_action_tracking);
+ pe__clear_action_flags(action, pcmk_action_detect_loop);
}
}
/*!
* \internal
* \brief Add start and restart orderings for probes scheduled for a resource
*
* \param[in,out] data Resource whose probes should be ordered
* \param[in] user_data Unused
*/
static void
add_start_restart_orderings_for_rsc(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
GList *probes = NULL;
// For collective resources, order each instance recursively
if (rsc->variant != pcmk_rsc_variant_primitive) {
g_list_foreach(rsc->children, add_start_restart_orderings_for_rsc,
NULL);
return;
}
// Find all probes for given resource
probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE);
// Add probe restart orderings for each probe found
for (GList *iter = probes; iter != NULL; iter = iter->next) {
pe_action_t *probe = (pe_action_t *) iter->data;
for (GList *then_iter = probe->actions_after; then_iter != NULL;
then_iter = then_iter->next) {
pe_action_wrapper_t *then = (pe_action_wrapper_t *) then_iter->data;
add_start_orderings_for_probe(probe, then);
add_restart_orderings_for_probe(probe, then->action);
clear_actions_tracking_flag(rsc->cluster);
}
}
g_list_free(probes);
}
/*!
* \internal
* \brief Add "A then probe B" orderings for "A then B" orderings
*
* \param[in,out] data_set Cluster working set
*
* \note This function is currently disabled (see next comment).
*/
static void
order_then_probes(pe_working_set_t *data_set)
{
#if 0
/* Given an ordering "A then B", we would prefer to wait for A to be started
* before probing B.
*
* For example, if A is a filesystem which B can't even run without, it
* would be helpful if the author of B's agent could assume that A is
* running before B.monitor will be called.
*
* However, we can't _only_ probe after A is running, otherwise we wouldn't
* detect the state of B if A could not be started. We can't even do an
* opportunistic version of this, because B may be moving:
*
* A.stop -> A.start -> B.probe -> B.stop -> B.start
*
* and if we add B.stop -> A.stop here, we get a loop:
*
* A.stop -> A.start -> B.probe -> B.stop -> A.stop
*
* We could kill the "B.probe -> B.stop" dependency, but that could mean
* stopping B "too" soon, because B.start must wait for the probe, and
* we don't want to stop B if we can't start it.
*
* We could add the ordering only if A is an anonymous clone with
* clone-max == node-max (since we'll never be moving it). However, we could
* still be stopping one instance at the same time as starting another.
*
* The complexity of checking for allowed conditions combined with the ever
* narrowing use case suggests that this code should remain disabled until
* someone gets smarter.
*/
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
pe_action_t *start = NULL;
GList *actions = NULL;
GList *probes = NULL;
actions = pe__resource_actions(rsc, NULL, PCMK_ACTION_START, FALSE);
if (actions) {
start = actions->data;
g_list_free(actions);
}
if (start == NULL) {
crm_err("No start action for %s", rsc->id);
continue;
}
probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE);
for (actions = start->actions_before; actions != NULL;
actions = actions->next) {
pe_action_wrapper_t *before = (pe_action_wrapper_t *) actions->data;
pe_action_t *first = before->action;
pe_resource_t *first_rsc = first->rsc;
if (first->required_runnable_before) {
for (GList *clone_actions = first->actions_before;
clone_actions != NULL;
clone_actions = clone_actions->next) {
before = (pe_action_wrapper_t *) clone_actions->data;
crm_trace("Testing '%s then %s' for %s",
first->uuid, before->action->uuid, start->uuid);
CRM_ASSERT(before->action->rsc != NULL);
first_rsc = before->action->rsc;
break;
}
} else if (!pcmk__str_eq(first->task, PCMK_ACTION_START,
pcmk__str_none)) {
crm_trace("Not a start op %s for %s", first->uuid, start->uuid);
}
if (first_rsc == NULL) {
continue;
} else if (pe__const_top_resource(first_rsc, false)
== pe__const_top_resource(start->rsc, false)) {
crm_trace("Same parent %s for %s", first_rsc->id, start->uuid);
continue;
} else if (!pe_rsc_is_clone(pe__const_top_resource(first_rsc,
false))) {
crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid);
continue;
}
crm_err("Applying %s before %s %d", first->uuid, start->uuid,
pe__const_top_resource(first_rsc, false)->variant);
for (GList *probe_iter = probes; probe_iter != NULL;
probe_iter = probe_iter->next) {
pe_action_t *probe = (pe_action_t *) probe_iter->data;
crm_err("Ordering %s before %s", first->uuid, probe->uuid);
order_actions(first, probe, pe_order_optional);
}
}
}
#endif
}
void
pcmk__order_probes(pe_working_set_t *data_set)
{
// Add orderings for "probe then X"
g_list_foreach(data_set->resources, add_start_restart_orderings_for_rsc,
NULL);
add_probe_orderings_for_stops(data_set);
order_then_probes(data_set);
}
/*!
* \internal
* \brief Schedule any probes needed
*
* \param[in,out] data_set Cluster working set
*
* \note This may also schedule fencing of failed remote nodes.
*/
void
pcmk__schedule_probes(pe_working_set_t *data_set)
{
// Schedule probes on each node in the cluster as needed
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
const char *probed = NULL;
if (!node->details->online) { // Don't probe offline nodes
if (pcmk__is_failed_remote_node(node)) {
pe_fence_node(data_set, node,
"the connection is unrecoverable", FALSE);
}
continue;
} else if (node->details->unclean) { // ... or nodes that need fencing
continue;
} else if (!node->details->rsc_discovery_enabled) {
// The user requested that probes not be done on this node
continue;
}
/* This is no longer needed for live clusters, since the probe_complete
* node attribute will never be in the CIB. However this is still useful
* for processing old saved CIBs (< 1.1.14), including the
* reprobe-target_rc regression test.
*/
probed = pe_node_attribute_raw(node, CRM_OP_PROBED);
if (probed != NULL && crm_is_true(probed) == FALSE) {
pe_action_t *probe_op = NULL;
probe_op = custom_action(NULL,
crm_strdup_printf("%s-%s", CRM_OP_REPROBE,
node->details->uname),
CRM_OP_REPROBE, node, FALSE, TRUE,
data_set);
add_hash_param(probe_op->meta, XML_ATTR_TE_NOWAIT,
XML_BOOLEAN_TRUE);
continue;
}
// Probe each resource in the cluster on this node, as needed
pcmk__probe_resource_list(data_set->resources, node);
}
}