diff --git a/include/crm/common/actions.h b/include/crm/common/actions.h
index e5727d2eec..790f411463 100644
--- a/include/crm/common/actions.h
+++ b/include/crm/common/actions.h
@@ -1,285 +1,288 @@
/*
* 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,
- pe_action_pseudo = 0x00001,
+ //! Whether action does not require invoking an agent
+ pcmk_action_pseudo = (1 << 0),
+
+ pe_action_pseudo = pcmk_action_pseudo,
pe_action_runnable = 0x00002,
pe_action_optional = 0x00004,
pe_action_print_always = 0x00008,
pe_action_have_node_attrs = 0x00010,
pe_action_implied_by_stonith = 0x00040,
pe_action_migrate_runnable = 0x00080,
pe_action_dumped = 0x00100,
pe_action_processed = 0x00200,
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
pe_action_clear = 0x00400, //! \deprecated Unused
#endif
pe_action_dangle = 0x00800,
/* This action requires one or more of its dependencies to be runnable.
* We use this to clear the runnable flag before checking dependencies.
*/
pe_action_requires_any = 0x01000,
pe_action_reschedule = 0x02000,
pe_action_tracking = 0x04000,
pe_action_dedup = 0x08000, //! Internal state tracking when creating graph
pe_action_dc = 0x10000, //! Action may run on DC instead of target
};
// 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 509671f942..218c0b905c 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), pe_action_pseudo)? "pseudo-action" : "action")
+ (pcmk_is_set((flags), pcmk_action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pe_action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pe_action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->details->uname)
/*!
* \internal
* \brief Add an XML node tag for a specified ID
*
* \param[in] id Node UUID to add
* \param[in,out] xml Parent XML tag to add to
*/
static xmlNode*
add_node_to_xml_by_id(const char *id, xmlNode *xml)
{
xmlNode *node_xml;
node_xml = create_xml_node(xml, XML_CIB_TAG_NODE);
crm_xml_add(node_xml, XML_ATTR_ID, id);
return node_xml;
}
/*!
* \internal
* \brief Add an XML node tag for a specified node
*
* \param[in] node Node to add
* \param[in,out] xml XML to add node to
*/
static void
add_node_to_xml(const pe_node_t *node, void *xml)
{
add_node_to_xml_by_id(node->details->id, (xmlNode *) xml);
}
/*!
* \internal
* \brief Count (optionally add to XML) nodes needing maintenance state update
*
* \param[in,out] xml Parent XML tag to add to, if any
* \param[in] data_set Working set for cluster
*
* \return Count of nodes added
* \note Only Pacemaker Remote nodes are considered currently
*/
static int
add_maintenance_nodes(xmlNode *xml, const pe_working_set_t *data_set)
{
xmlNode *maintenance = NULL;
int count = 0;
if (xml != NULL) {
maintenance = create_xml_node(xml, XML_GRAPH_TAG_MAINTENANCE);
}
for (const GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
const pe_node_t *node = iter->data;
if (pe__is_guest_or_remote_node(node) &&
(node->details->maintenance != node->details->remote_maintenance)) {
if (maintenance != NULL) {
crm_xml_add(add_node_to_xml_by_id(node->details->id,
maintenance),
XML_NODE_IS_MAINTENANCE,
(node->details->maintenance? "1" : "0"));
}
count++;
}
}
crm_trace("%s %d nodes in need of maintenance mode update in state",
((maintenance == NULL)? "Counted" : "Added"), count);
return count;
}
/*!
* \internal
* \brief Add pseudo action with nodes needing maintenance state update
*
* \param[in,out] data_set Working set for cluster
*/
static void
add_maintenance_update(pe_working_set_t *data_set)
{
pe_action_t *action = NULL;
if (add_maintenance_nodes(NULL, data_set) != 0) {
action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, data_set);
pe__set_action_flags(action, pe_action_print_always);
}
}
/*!
* \internal
* \brief Add XML with nodes that an action is expected to bring down
*
* If a specified action is expected to bring any nodes down, add an XML block
* with their UUIDs. When a node is lost, this allows the controller to
* determine whether it was expected.
*
* \param[in,out] xml Parent XML tag to add to
* \param[in] action Action to check for downed nodes
*/
static void
add_downed_nodes(xmlNode *xml, const pe_action_t *action)
{
CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL),
return);
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) {
/* Shutdown makes the action's node down */
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
/* Fencing makes the action's node and any hosted guest nodes down */
const char *fence = g_hash_table_lookup(action->meta, "stonith_action");
if (pcmk__is_fencing_action(fence)) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->node->details->id, downed);
pe_foreach_guest_node(action->node->details->data_set, action->node,
add_node_to_xml, downed);
}
} else if (action->rsc && action->rsc->is_remote_node
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
/* Stopping a remote connection resource makes connected node down,
* unless it's part of a migration
*/
GList *iter;
pe_action_t *input;
bool migrating = false;
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
input = ((pe_action_wrapper_t *) iter->data)->action;
if ((input->rsc != NULL)
&& pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none)
&& pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
migrating = true;
break;
}
}
if (!migrating) {
xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED);
add_node_to_xml_by_id(action->rsc->id, downed);
}
}
}
/*!
* \internal
* \brief Create a transition graph operation key for a clone action
*
* \param[in] action Clone action
* \param[in] interval_ms Action interval in milliseconds
*
* \return Newly allocated string with transition graph operation key
*/
static char *
clone_op_key(const pe_action_t *action, guint interval_ms)
{
if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = g_hash_table_lookup(action->meta, "notify_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_operation");
CRM_LOG_ASSERT((n_type != NULL) && (n_task != NULL));
return pcmk__notify_key(action->rsc->clone_name, n_type, n_task);
} else if (action->cancel_task != NULL) {
return pcmk__op_key(action->rsc->clone_name, action->cancel_task,
interval_ms);
} else {
return pcmk__op_key(action->rsc->clone_name, action->task, interval_ms);
}
}
/*!
* \internal
* \brief Add node details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] xml Transition graph action XML for \p action
*/
static void
add_node_details(const pe_action_t *action, xmlNode *xml)
{
pe_node_t *router_node = pcmk__connection_host_for_action(action);
crm_xml_add(xml, XML_LRM_ATTR_TARGET, action->node->details->uname);
crm_xml_add(xml, XML_LRM_ATTR_TARGET_UUID, action->node->details->id);
if (router_node != NULL) {
crm_xml_add(xml, XML_LRM_ATTR_ROUTER_NODE, router_node->details->uname);
}
}
/*!
* \internal
* \brief Add resource details to transition graph action XML
*
* \param[in] action Scheduled action
* \param[in,out] action_xml Transition graph action XML for \p action
*/
static void
add_resource_details(const pe_action_t *action, xmlNode *action_xml)
{
xmlNode *rsc_xml = NULL;
const char *attr_list[] = {
XML_AGENT_ATTR_CLASS,
XML_AGENT_ATTR_PROVIDER,
XML_ATTR_TYPE
};
/* If a resource is locked to a node via shutdown-lock, mark its actions
* so the controller can preserve the lock when the action completes.
*/
if (pcmk__action_locks_rsc_to_node(action)) {
crm_xml_add_ll(action_xml, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
(long long) action->rsc->lock_time);
}
// List affected resource
rsc_xml = create_xml_node(action_xml,
(const char *) action->rsc->xml->name);
if (pcmk_is_set(action->rsc->flags, 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, pe_action_pseudo)) {
+ 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, pe_action_pseudo)) {
+ } 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, pe_action_pseudo)) {
+ && !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, pe_action_runnable)) {
crm_trace("Ignoring action %s (%d): unrunnable",
action->uuid, action->id);
return false;
}
if (pcmk_is_set(action->flags, pe_action_optional)
&& !pcmk_is_set(action->flags, pe_action_print_always)) {
crm_trace("Ignoring action %s (%d): optional",
action->uuid, action->id);
return false;
}
/* Actions for unmanaged resources should be excluded from the graph,
* with the exception of monitors and cancellation of recurring monitors.
*/
if ((action->rsc != NULL)
&& !pcmk_is_set(action->rsc->flags, 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, pe_action_pseudo)
+ 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, pe_action_runnable)
&& !ordering_can_change_actions(input)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (!pcmk_is_set(input->action->flags, pe_action_runnable)
&& pcmk_is_set(input->type, pe_order_one_or_more)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"one-or-more and input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_implies_first_migratable)
&& !pcmk_is_set(input->action->flags, pe_action_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"implies input migratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (pcmk_is_set(input->type, pe_order_apply_first_non_migratable)
&& pcmk_is_set(input->action->flags,
pe_action_migrate_runnable)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"only if input unmigratable but input unrunnable",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if ((input->type == pe_order_optional)
&& pcmk_is_set(input->action->flags, pe_action_migrate_runnable)
&& pcmk__ends_with(input->action->uuid, "_stop_0")) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"optional but stop in migration",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
} else if (input->type == pe_order_load) {
pe_node_t *input_node = input->action->node;
// load orderings are relevant only if actions are for same node
if ((action->rsc != NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
pe_node_t *assigned = action->rsc->allocated_to;
/* For load_stopped -> migrate_to orderings, we care about where it
* has been assigned to, not where it will be executed.
*/
if (!pe__same_node(input_node, assigned)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(assigned? assigned->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
}
} else if (!pe__same_node(input_node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
(action->node? action->node->details->uname : "<none>"),
(input_node? input_node->details->uname : "<none>"));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"load ordering input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->type == pe_order_anti_colocation) {
if (input->action->node && action->node
&& !pe__same_node(input->action->node, action->node)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation node mismatch %s vs %s",
action->uuid, action->id,
input->action->uuid, input->action->id,
pe__node_name(action->node),
pe__node_name(input->action->node));
input->type = pe_order_none;
return false;
} else if (pcmk_is_set(input->action->flags, pe_action_optional)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"anti-colocation input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
input->type = pe_order_none;
return false;
}
} else if (input->action->rsc
&& input->action->rsc != action->rsc
&& pcmk_is_set(input->action->rsc->flags, 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, pe_action_optional)
&& !pcmk_any_flags_set(input->action->flags,
pe_action_print_always|pe_action_dumped)
&& !should_add_action_to_graph(input->action)) {
crm_trace("Ignoring %s (%d) input %s (%d): "
"input optional",
action->uuid, action->id,
input->action->uuid, input->action->id);
return false;
}
crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x",
action->uuid, action->id, action_type_str(input->action->flags),
input->action->uuid, input->action->id,
action_node_str(input->action),
action_runnable_str(input->action->flags),
action_optional_str(input->action->flags), input->type);
return true;
}
/*!
* \internal
* \brief Check whether an ordering creates an ordering loop
*
* \param[in] init_action "First" action in ordering
* \param[in] action Callers should always set this the same as
* \p init_action (this function may use a different
* value for recursive calls)
* \param[in,out] input Action wrapper for "then" action in ordering
*
* \return true if the ordering creates a loop, otherwise false
*/
bool
pcmk__graph_has_loop(const pe_action_t *init_action, const pe_action_t *action,
pe_action_wrapper_t *input)
{
bool has_loop = false;
if (pcmk_is_set(input->action->flags, pe_action_tracking)) {
crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
return false;
}
// Don't need to check inputs that won't be used
if (!should_add_input_to_graph(action, input)) {
return false;
}
if (input->action == init_action) {
crm_debug("Input loop found in %s@%s ->...-> %s@%s",
action->uuid,
action->node? action->node->details->uname : "",
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
return true;
}
pe__set_action_flags(input->action, pe_action_tracking);
crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)"
"for graph loop with %s@%s ",
action->uuid,
action->node? action->node->details->uname : "",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
input->type,
init_action->uuid,
init_action->node? init_action->node->details->uname : "");
// Recursively check input itself for loops
for (GList *iter = input->action->actions_before;
iter != NULL; iter = iter->next) {
if (pcmk__graph_has_loop(init_action, input->action,
(pe_action_wrapper_t *) iter->data)) {
// Recursive call already logged a debug message
has_loop = true;
break;
}
}
pe__clear_action_flags(input->action, pe_action_tracking);
if (!has_loop) {
crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)",
input->action->uuid,
input->action->node? input->action->node->details->uname : "",
action->uuid,
action->node? action->node->details->uname : "",
input->type);
}
return has_loop;
}
/*!
* \internal
* \brief Create a synapse XML element for a transition graph
*
* \param[in] action Action that synapse is for
* \param[in,out] data_set Cluster working set containing graph
*
* \return Newly added XML element for new graph synapse
*/
static xmlNode *
create_graph_synapse(const pe_action_t *action, pe_working_set_t *data_set)
{
int synapse_priority = 0;
xmlNode *syn = create_xml_node(data_set->graph, "synapse");
crm_xml_add_int(syn, XML_ATTR_ID, data_set->num_synapse);
data_set->num_synapse++;
if (action->rsc != NULL) {
synapse_priority = action->rsc->priority;
}
if (action->priority > synapse_priority) {
synapse_priority = action->priority;
}
if (synapse_priority > 0) {
crm_xml_add_int(syn, XML_CIB_ATTR_PRIORITY, synapse_priority);
}
return syn;
}
/*!
* \internal
* \brief Add an action to the transition graph XML if appropriate
*
* \param[in,out] data Action to possibly add
* \param[in,out] user_data Cluster working set
*
* \note This will de-duplicate the action inputs, meaning that the
* pe_action_wrapper_t:type flags can no longer be relied on to retain
* their original settings. That means this MUST be called after
* pcmk__apply_orderings() is complete, and nothing after this should rely
* on those type flags. (For example, some code looks for type equal to
* some flag rather than whether the flag is set, and some code looks for
* particular combinations of flags -- such code must be done before
* pcmk__create_graph().)
*/
static void
add_action_to_graph(gpointer data, gpointer user_data)
{
pe_action_t *action = (pe_action_t *) data;
pe_working_set_t *data_set = (pe_working_set_t *) user_data;
xmlNode *syn = NULL;
xmlNode *set = NULL;
xmlNode *in = NULL;
/* If we haven't already, de-duplicate inputs (even if we won't be adding
* the action to the graph, so that crm_simulate's dot graphs don't have
* duplicates).
*/
if (!pcmk_is_set(action->flags, pe_action_dedup)) {
pcmk__deduplicate_action_inputs(action);
pe__set_action_flags(action, pe_action_dedup);
}
if (pcmk_is_set(action->flags, pe_action_dumped) // Already added, or
|| !should_add_action_to_graph(action)) { // shouldn't be added
return;
}
pe__set_action_flags(action, pe_action_dumped);
crm_trace("Adding action %d (%s%s%s) to graph",
action->id, action->uuid,
((action->node == NULL)? "" : " on "),
((action->node == NULL)? "" : action->node->details->uname));
syn = create_graph_synapse(action, data_set);
set = create_xml_node(syn, "action_set");
in = create_xml_node(syn, "inputs");
create_graph_action(set, action, false, data_set);
for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) lpc->data;
if (should_add_input_to_graph(action, input)) {
xmlNode *input_xml = create_xml_node(in, "trigger");
input->state = pe_link_dumped;
create_graph_action(input_xml, input->action, true, data_set);
}
}
}
static int transition_id = -1;
/*!
* \internal
* \brief Log a message after calculating a transition
*
* \param[in] filename Where transition input is stored
*/
void
pcmk__log_transition_summary(const char *filename)
{
if (was_processing_error) {
crm_err("Calculated transition %d (with errors)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else if (was_processing_warning) {
crm_warn("Calculated transition %d (with warnings)%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
} else {
crm_notice("Calculated transition %d%s%s",
transition_id,
(filename == NULL)? "" : ", saving inputs in ",
(filename == NULL)? "" : filename);
}
if (crm_config_error) {
crm_notice("Configuration errors found during scheduler processing,"
" please run \"crm_verify -L\" to identify issues");
}
}
/*!
* \internal
* \brief Add a resource's actions to the transition graph
*
* \param[in,out] rsc Resource whose actions should be added
*/
void
pcmk__add_rsc_actions_to_graph(pe_resource_t *rsc)
{
GList *iter = NULL;
CRM_ASSERT(rsc != NULL);
pe_rsc_trace(rsc, "Adding actions for %s to graph", rsc->id);
// First add the resource's own actions
g_list_foreach(rsc->actions, add_action_to_graph, rsc->cluster);
// Then recursively add its children's actions (appropriate to variant)
for (iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child_rsc = (pe_resource_t *) iter->data;
child_rsc->cmds->add_actions_to_graph(child_rsc);
}
}
/*!
* \internal
* \brief Create a transition graph with all cluster actions needed
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__create_graph(pe_working_set_t *data_set)
{
GList *iter = NULL;
const char *value = NULL;
long long limit = 0LL;
transition_id++;
crm_trace("Creating transition graph %d", transition_id);
data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
value = pe_pref(data_set->config_hash, "cluster-delay");
crm_xml_add(data_set->graph, "cluster-delay", value);
value = pe_pref(data_set->config_hash, "stonith-timeout");
crm_xml_add(data_set->graph, "stonith-timeout", value);
crm_xml_add(data_set->graph, "failed-stop-offset", "INFINITY");
if (pcmk_is_set(data_set->flags, 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,
pe_action_optional|pe_action_runnable)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* Eventually we should just ignore the 'fence' case, but for now
* it's the best way to detect (in CTS) when CIB resource updates
* are being lost.
*/
if (pcmk_is_set(data_set->flags, 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_actions.c b/lib/pacemaker/pcmk_sched_actions.c
index 34db9a7482..39f1a560b4 100644
--- a/lib/pacemaker/pcmk_sched_actions.c
+++ b/lib/pacemaker/pcmk_sched_actions.c
@@ -1,1930 +1,1930 @@
/*
* 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 <stdio.h>
#include <sys/param.h>
#include <glib.h>
#include <crm/lrmd_internal.h>
#include <crm/common/scheduler_internal.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Get the action flags relevant to ordering constraints
*
* \param[in,out] action Action to check
* \param[in] node Node that *other* action in the ordering is on
* (used only for clone resource actions)
*
* \return Action flags that should be used for orderings
*/
static uint32_t
action_flags_for_ordering(pe_action_t *action, const pe_node_t *node)
{
bool runnable = false;
uint32_t flags;
// For non-resource actions, return the action flags
if (action->rsc == NULL) {
return action->flags;
}
/* For non-clone resources, or a clone action not assigned to a node,
* return the flags as determined by the resource method without a node
* specified.
*/
flags = action->rsc->cmds->action_flags(action, NULL);
if ((node == NULL) || !pe_rsc_is_clone(action->rsc)) {
return flags;
}
/* Otherwise (i.e., for clone resource actions on a specific node), first
* remember whether the non-node-specific action is runnable.
*/
runnable = pcmk_is_set(flags, pe_action_runnable);
// Then recheck the resource method with the node
flags = action->rsc->cmds->action_flags(action, node);
/* For clones in ordering constraints, the node-specific "runnable" doesn't
* matter, just the non-node-specific setting (i.e., is the action runnable
* anywhere).
*
* This applies only to runnable, and only for ordering constraints. This
* function shouldn't be used for other types of constraints without
* changes. Not very satisfying, but it's logical and appears to work well.
*/
if (runnable && !pcmk_is_set(flags, pe_action_runnable)) {
pe__set_raw_action_flags(flags, action->rsc->id,
pe_action_runnable);
}
return flags;
}
/*!
* \internal
* \brief Get action UUID that should be used with a resource ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the UUID and resource of the first action in an
* ordering, this returns the UUID of the action that should actually be used
* for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0").
*
* \param[in] first_uuid UUID of first action in ordering
* \param[in] first_rsc Resource of first action in ordering
*
* \return Newly allocated copy of UUID to use with ordering
* \note It is the caller's responsibility to free the return value.
*/
static char *
action_uuid_for_ordering(const char *first_uuid, const pe_resource_t *first_rsc)
{
guint interval_ms = 0;
char *uuid = NULL;
char *rid = NULL;
char *first_task_str = NULL;
enum action_tasks first_task = pcmk_action_unspecified;
enum action_tasks remapped_task = pcmk_action_unspecified;
// Only non-notify actions for collective resources need remapping
if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL)
|| (first_rsc->variant < pcmk_rsc_variant_group)) {
goto done;
}
// Only non-recurring actions need remapping
CRM_ASSERT(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms));
if (interval_ms > 0) {
goto done;
}
first_task = text2task(first_task_str);
switch (first_task) {
case pcmk_action_stop:
case pcmk_action_start:
case pcmk_action_notify:
case pcmk_action_promote:
case pcmk_action_demote:
remapped_task = first_task + 1;
break;
case pcmk_action_stopped:
case pcmk_action_started:
case pcmk_action_notified:
case pcmk_action_promoted:
case pcmk_action_demoted:
remapped_task = first_task;
break;
case pcmk_action_monitor:
case pcmk_action_shutdown:
case pcmk_action_fence:
break;
default:
crm_err("Unknown action '%s' in ordering", first_task_str);
break;
}
if (remapped_task != pcmk_action_unspecified) {
/* If a clone or bundle has notifications enabled, the ordering will be
* relative to when notifications have been sent for the remapped task.
*/
if (pcmk_is_set(first_rsc->flags, pcmk_rsc_notify)
&& (pe_rsc_is_clone(first_rsc) || pe_rsc_is_bundled(first_rsc))) {
uuid = pcmk__notify_key(rid, "confirmed-post",
task2text(remapped_task));
} else {
uuid = pcmk__op_key(rid, task2text(remapped_task), 0);
}
pe_rsc_trace(first_rsc,
"Remapped action UUID %s to %s for ordering purposes",
first_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(first_uuid);
CRM_ASSERT(uuid != NULL);
}
free(first_task_str);
free(rid);
return uuid;
}
/*!
* \internal
* \brief Get actual action that should be used with an ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the first action in an ordering, this returns the
* the action that should actually be used for ordering (for example, the
* started action instead of the start action).
*
* \param[in] action First action in an ordering
*
* \return Actual action that should be used for the ordering
*/
static pe_action_t *
action_for_ordering(pe_action_t *action)
{
pe_action_t *result = action;
pe_resource_t *rsc = action->rsc;
if ((rsc != NULL) && (rsc->variant >= pcmk_rsc_variant_group)
&& (action->uuid != NULL)) {
char *uuid = action_uuid_for_ordering(action->uuid, rsc);
result = find_first_action(rsc->actions, uuid, NULL, NULL);
if (result == NULL) {
crm_warn("Not remapping %s to %s because %s does not have "
"remapped action", action->uuid, uuid, rsc->id);
result = action;
}
free(uuid);
}
return result;
}
/*!
* \internal
* \brief Wrapper for update_ordered_actions() method for readability
*
* \param[in,out] rsc Resource to call method for
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this
* node (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates
* (may include pe_action_optional to affect only
* mandatory actions, and pe_action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static inline uint32_t
update(pe_resource_t *rsc, pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
return rsc->cmds->update_ordered_actions(first, then, node, flags, filter,
type, data_set);
}
/*!
* \internal
* \brief Update flags for ordering's actions appropriately for ordering's flags
*
* \param[in,out] first First action in an ordering
* \param[in,out] then Then action in an ordering
* \param[in] first_flags Action flags for \p first for ordering purposes
* \param[in] then_flags Action flags for \p then for ordering purposes
* \param[in,out] order Action wrapper for \p first in ordering
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags
*/
static uint32_t
update_action_for_ordering_flags(pe_action_t *first, pe_action_t *then,
uint32_t first_flags, uint32_t then_flags,
pe_action_wrapper_t *order,
pe_working_set_t *data_set)
{
uint32_t changed = pcmk__updated_none;
/* The node will only be used for clones. If interleaved, node will be NULL,
* otherwise the ordering scope will be limited to the node. Normally, the
* whole 'then' clone should restart if 'first' is restarted, so then->node
* is needed.
*/
pe_node_t *node = then->node;
if (pcmk_is_set(order->type, pe_order_implies_then_on_node)) {
/* For unfencing, only instances of 'then' on the same node as 'first'
* (the unfencing operation) should restart, so reset node to
* first->node, at which point this case is handled like a normal
* pe_order_implies_then.
*/
pe__clear_order_flags(order->type, pe_order_implies_then_on_node);
pe__set_order_flags(order->type, pe_order_implies_then);
node = first->node;
pe_rsc_trace(then->rsc,
"%s then %s: mapped pe_order_implies_then_on_node to "
"pe_order_implies_then on %s",
first->uuid, then->uuid, pe__node_name(node));
}
if (pcmk_is_set(order->type, pe_order_implies_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pe_action_optional,
pe_action_optional, pe_order_implies_then,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_optional)
&& pcmk_is_set(then->flags, pe_action_optional)) {
pe__clear_action_flags(then, pe_action_optional);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_implies_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_restart) && (then->rsc != NULL)) {
enum pe_action_flags restart = pe_action_optional|pe_action_runnable;
changed |= update(then->rsc, first, then, node, first_flags, restart,
pe_order_restart, data_set);
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_restart",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_implies_first)) {
if (first->rsc != NULL) {
changed |= update(first->rsc, first, then, node, first_flags,
pe_action_optional, pe_order_implies_first,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_optional)
&& pcmk_is_set(first->flags, pe_action_runnable)) {
pe__clear_action_flags(first, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_promoted_implies_first)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pe_action_optional,
pe_action_optional,
pe_order_promoted_implies_first, data_set);
}
pe_rsc_trace(then->rsc,
"%s then %s: %s after pe_order_promoted_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_one_or_more)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_one_or_more,
data_set);
} else if (pcmk_is_set(first_flags, pe_action_runnable)) {
// We have another runnable instance of "first"
then->runnable_before++;
/* Mark "then" as runnable if it requires a certain number of
* "before" instances to be runnable, and they now are.
*/
if ((then->runnable_before >= then->required_runnable_before)
&& !pcmk_is_set(then->flags, pe_action_runnable)) {
pe__set_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_one_or_more",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_probe) && (then->rsc != NULL)) {
if (!pcmk_is_set(first_flags, pe_action_runnable)
&& (first->rsc->running_on != NULL)) {
pe_rsc_trace(then->rsc,
"%s then %s: ignoring because first is stopping",
first->uuid, then->uuid);
order->type = pe_order_none;
} else {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_runnable_left,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_probe",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_runnable_left)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_runnable_left,
data_set);
} else if (!pcmk_is_set(first_flags, pe_action_runnable)
&& pcmk_is_set(then->flags, pe_action_runnable)) {
pe__clear_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_runnable_left",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_implies_first_migratable)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_optional,
pe_order_implies_first_migratable, data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after "
"pe_order_implies_first_migratable",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_pseudo_left)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_optional, pe_order_pseudo_left,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_pseudo_left",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_optional)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_optional, data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_optional",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pe_order_asymmetrical)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pe_action_runnable, pe_order_asymmetrical,
data_set);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after pe_order_asymmetrical",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(first->flags, pe_action_runnable)
&& pcmk_is_set(order->type, pe_order_implies_then_printed)
&& !pcmk_is_set(first_flags, pe_action_optional)) {
pe_rsc_trace(then->rsc, "%s will be in graph because %s is required",
then->uuid, first->uuid);
pe__set_action_flags(then, pe_action_print_always);
// Don't bother marking 'then' as changed just for this
}
if (pcmk_is_set(order->type, pe_order_implies_first_printed)
&& !pcmk_is_set(then_flags, pe_action_optional)) {
pe_rsc_trace(then->rsc, "%s will be in graph because %s is required",
first->uuid, then->uuid);
pe__set_action_flags(first, pe_action_print_always);
// Don't bother marking 'first' as changed just for this
}
if (pcmk_any_flags_set(order->type, pe_order_implies_then
|pe_order_implies_first
|pe_order_restart)
&& (first->rsc != NULL)
&& !pcmk_is_set(first->rsc->flags, pcmk_rsc_managed)
&& pcmk_is_set(first->rsc->flags, pcmk_rsc_blocked)
&& !pcmk_is_set(first->flags, pe_action_runnable)
&& pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) {
if (pcmk_is_set(then->flags, pe_action_runnable)) {
pe__clear_action_flags(then, pe_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pe_rsc_trace(then->rsc, "%s then %s: %s after checking whether first "
"is blocked, unmanaged, unrunnable stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
return changed;
}
// Convenience macros for logging action properties
#define action_type_str(flags) \
- (pcmk_is_set((flags), pe_action_pseudo)? "pseudo-action" : "action")
+ (pcmk_is_set((flags), pcmk_action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pe_action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pe_action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->details->uname)
/*!
* \internal
* \brief Update an action's flags for all orderings where it is "then"
*
* \param[in,out] then Action to update
* \param[in,out] data_set Cluster working set
*/
void
pcmk__update_action_for_orderings(pe_action_t *then, pe_working_set_t *data_set)
{
GList *lpc = NULL;
uint32_t changed = pcmk__updated_none;
int last_flags = then->flags;
pe_rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s",
action_type_str(then->flags), then->uuid,
action_optional_str(then->flags),
action_runnable_str(then->flags), action_node_str(then));
if (pcmk_is_set(then->flags, pe_action_requires_any)) {
/* Initialize current known "runnable before" actions. As
* update_action_for_ordering_flags() is called for each of then's
* before actions, this number will increment as runnable 'first'
* actions are encountered.
*/
then->runnable_before = 0;
if (then->required_runnable_before == 0) {
/* @COMPAT This ordering constraint uses the deprecated
* "require-all=false" attribute. Treat it like "clone-min=1".
*/
then->required_runnable_before = 1;
}
/* The pe_order_one_or_more clause of update_action_for_ordering_flags()
* (called below) will reset runnable if appropriate.
*/
pe__clear_action_flags(then, pe_action_runnable);
}
for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc->data;
pe_action_t *first = other->action;
pe_node_t *then_node = then->node;
pe_node_t *first_node = first->node;
if ((first->rsc != NULL)
&& (first->rsc->variant == pcmk_rsc_variant_group)
&& pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) {
first_node = first->rsc->fns->location(first->rsc, NULL, FALSE);
if (first_node != NULL) {
pe_rsc_trace(first->rsc, "Found %s for 'first' %s",
pe__node_name(first_node), first->uuid);
}
}
if ((then->rsc != NULL)
&& (then->rsc->variant == pcmk_rsc_variant_group)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) {
then_node = then->rsc->fns->location(then->rsc, NULL, FALSE);
if (then_node != NULL) {
pe_rsc_trace(then->rsc, "Found %s for 'then' %s",
pe__node_name(then_node), then->uuid);
}
}
// Disable constraint if it only applies when on same node, but isn't
if (pcmk_is_set(other->type, pe_order_same_node)
&& (first_node != NULL) && (then_node != NULL)
&& !pe__same_node(first_node, then_node)) {
pe_rsc_trace(then->rsc,
"Disabled ordering %s on %s then %s on %s: "
"not same node",
other->action->uuid, pe__node_name(first_node),
then->uuid, pe__node_name(then_node));
other->type = pe_order_none;
continue;
}
pcmk__clear_updated_flags(changed, then, pcmk__updated_first);
if ((first->rsc != NULL)
&& pcmk_is_set(other->type, pe_order_then_cancels_first)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
/* 'then' is required, so we must abandon 'first'
* (e.g. a required stop cancels any agent reload).
*/
pe__set_action_flags(other->action, pe_action_optional);
if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) {
pe__clear_resource_flags(first->rsc, pcmk_rsc_reload);
}
}
if ((first->rsc != NULL) && (then->rsc != NULL)
&& (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) {
first = action_for_ordering(first);
}
if (first != other->action) {
pe_rsc_trace(then->rsc, "Ordering %s after %s instead of %s",
then->uuid, first->uuid, other->action->uuid);
}
pe_rsc_trace(then->rsc,
"%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s",
first->uuid, first->flags, then->uuid, then->flags,
other->type, action_node_str(first));
if (first == other->action) {
/* 'first' was not remapped (e.g. from 'start' to 'running'), which
* could mean it is a non-resource action, a primitive resource
* action, or already expanded.
*/
uint32_t first_flags, then_flags;
first_flags = action_flags_for_ordering(first, then_node);
then_flags = action_flags_for_ordering(then, first_node);
changed |= update_action_for_ordering_flags(first, then,
first_flags, then_flags,
other, data_set);
/* 'first' was for a complex resource (clone, group, etc),
* create a new dependency if necessary
*/
} else if (order_actions(first, then, other->type)) {
/* This was the first time 'first' and 'then' were associated,
* start again to get the new actions_before list
*/
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
pe_rsc_trace(then->rsc,
"Disabled ordering %s then %s in favor of %s then %s",
other->action->uuid, then->uuid, first->uuid,
then->uuid);
other->type = pe_order_none;
}
if (pcmk_is_set(changed, pcmk__updated_first)) {
crm_trace("Re-processing %s and its 'after' actions "
"because it changed", first->uuid);
for (GList *lpc2 = first->actions_after; lpc2 != NULL;
lpc2 = lpc2->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc2->data;
pcmk__update_action_for_orderings(other->action, data_set);
}
pcmk__update_action_for_orderings(first, data_set);
}
}
if (pcmk_is_set(then->flags, pe_action_requires_any)) {
if (last_flags == then->flags) {
pcmk__clear_updated_flags(changed, then, pcmk__updated_then);
} else {
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
}
}
if (pcmk_is_set(changed, pcmk__updated_then)) {
crm_trace("Re-processing %s and its 'after' actions because it changed",
then->uuid);
if (pcmk_is_set(last_flags, pe_action_runnable)
&& !pcmk_is_set(then->flags, pe_action_runnable)) {
pcmk__block_colocation_dependents(then);
}
pcmk__update_action_for_orderings(then, data_set);
for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) {
pe_action_wrapper_t *other = (pe_action_wrapper_t *) lpc->data;
pcmk__update_action_for_orderings(other->action, data_set);
}
}
}
static inline bool
is_primitive_action(const pe_action_t *action)
{
return (action != NULL) && (action->rsc != NULL)
&& (action->rsc->variant == pcmk_rsc_variant_primitive);
}
/*!
* \internal
* \brief Clear a single action flag and set reason text
*
* \param[in,out] action Action whose flag should be cleared
* \param[in] flag Action flag that should be cleared
* \param[in] reason Action that is the reason why flag is being cleared
*/
#define clear_action_flag_because(action, flag, reason) do { \
if (pcmk_is_set((action)->flags, (flag))) { \
pe__clear_action_flags(action, flag); \
if ((action)->rsc != (reason)->rsc) { \
char *reason_text = pe__action2reason((reason), (flag)); \
pe_action_set_reason((action), reason_text, false); \
free(reason_text); \
} \
} \
} while (0)
/*!
* \internal
* \brief Update actions in an asymmetric ordering
*
* If the "first" action in an asymmetric ordering is unrunnable, make the
* "second" action unrunnable as well, if appropriate.
*
* \param[in] first 'First' action in an asymmetric ordering
* \param[in,out] then 'Then' action in an asymmetric ordering
*/
static void
handle_asymmetric_ordering(const pe_action_t *first, pe_action_t *then)
{
/* Only resource actions after an unrunnable 'first' action need updates for
* asymmetric ordering.
*/
if ((then->rsc == NULL) || pcmk_is_set(first->flags, pe_action_runnable)) {
return;
}
// Certain optional 'then' actions are unaffected by unrunnable 'first'
if (pcmk_is_set(then->flags, pe_action_optional)) {
enum rsc_role_e then_rsc_role = then->rsc->fns->state(then->rsc, TRUE);
if ((then_rsc_role == pcmk_role_stopped)
&& pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* If 'then' should stop after 'first' but is already stopped, the
* ordering is irrelevant.
*/
return;
} else if ((then_rsc_role >= pcmk_role_started)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)
&& pe__rsc_running_on_only(then->rsc, then->node)) {
/* Similarly if 'then' should start after 'first' but is already
* started on a single node.
*/
return;
}
}
// 'First' can't run, so 'then' can't either
clear_action_flag_because(then, pe_action_optional, first);
clear_action_flag_because(then, pe_action_runnable, first);
}
/*!
* \internal
* \brief Set action bits appropriately when pe_restart_order is used
*
* \param[in,out] first 'First' action in an ordering with pe_restart_order
* \param[in,out] then 'Then' action in an ordering with pe_restart_order
* \param[in] filter What action flags to care about
*
* \note pe_restart_order is set for "stop resource before starting it" and
* "stop later group member before stopping earlier group member"
*/
static void
handle_restart_ordering(pe_action_t *first, pe_action_t *then, uint32_t filter)
{
const char *reason = NULL;
CRM_ASSERT(is_primitive_action(first));
CRM_ASSERT(is_primitive_action(then));
// We need to update the action in two cases:
// ... if 'then' is required
if (pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
reason = "restart";
}
/* ... if 'then' is unrunnable action on same resource (if a resource
* should restart but can't start, we still want to stop)
*/
if (pcmk_is_set(filter, pe_action_runnable)
&& !pcmk_is_set(then->flags, pe_action_runnable)
&& pcmk_is_set(then->rsc->flags, pcmk_rsc_managed)
&& (first->rsc == then->rsc)) {
reason = "stop";
}
if (reason == NULL) {
return;
}
pe_rsc_trace(first->rsc, "Handling %s -> %s for %s",
first->uuid, then->uuid, reason);
// Make 'first' required if it is runnable
if (pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(first, pe_action_optional, then);
}
// Make 'first' required if 'then' is required
if (!pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
// Make 'first' unmigratable if 'then' is unmigratable
if (!pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable, then);
}
// Make 'then' unrunnable if 'first' is required but unrunnable
if (!pcmk_is_set(first->flags, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_runnable, first);
}
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two actions, update the actions' flags
* (and runnable_before members if appropriate) as appropriate for the ordering.
* Effects may cascade to other orderings involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (ignored)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__update_ordered_actions(pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
uint32_t changed = pcmk__updated_none;
uint32_t then_flags = 0U;
uint32_t first_flags = 0U;
CRM_ASSERT((first != NULL) && (then != NULL) && (data_set != NULL));
then_flags = then->flags;
first_flags = first->flags;
if (pcmk_is_set(type, pe_order_asymmetrical)) {
handle_asymmetric_ordering(first, then);
}
if (pcmk_is_set(type, pe_order_implies_first)
&& !pcmk_is_set(then_flags, pe_action_optional)) {
// Then is required, and implies first should be, too
if (pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(flags, pe_action_optional)
&& pcmk_is_set(first_flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
if (pcmk_is_set(flags, pe_action_migrate_runnable)
&& !pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable, then);
}
}
if (pcmk_is_set(type, pe_order_promoted_implies_first)
&& (then->rsc != NULL) && (then->rsc->role == pcmk_role_promoted)
&& pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
if (pcmk_is_set(first->flags, pe_action_migrate_runnable)
&& !pcmk_is_set(then->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(first, pe_action_migrate_runnable,
then);
}
}
if (pcmk_is_set(type, pe_order_implies_first_migratable)
&& pcmk_is_set(filter, pe_action_optional)) {
if (!pcmk_all_flags_set(then->flags, pe_action_migrate_runnable
|pe_action_runnable)) {
clear_action_flag_because(first, pe_action_runnable, then);
}
if (!pcmk_is_set(then->flags, pe_action_optional)) {
clear_action_flag_because(first, pe_action_optional, then);
}
}
if (pcmk_is_set(type, pe_order_pseudo_left)
&& pcmk_is_set(filter, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_migrate_runnable, first);
- pe__clear_action_flags(then, pe_action_pseudo);
+ pe__clear_action_flags(then, pcmk_action_pseudo);
}
if (pcmk_is_set(type, pe_order_runnable_left)
&& pcmk_is_set(filter, pe_action_runnable)
&& pcmk_is_set(then->flags, pe_action_runnable)
&& !pcmk_is_set(flags, pe_action_runnable)) {
clear_action_flag_because(then, pe_action_runnable, first);
clear_action_flag_because(then, pe_action_migrate_runnable, first);
}
if (pcmk_is_set(type, pe_order_implies_then)
&& pcmk_is_set(filter, pe_action_optional)
&& pcmk_is_set(then->flags, pe_action_optional)
&& !pcmk_is_set(flags, pe_action_optional)
&& !pcmk_is_set(first->flags, pe_action_migrate_runnable)) {
clear_action_flag_because(then, pe_action_optional, first);
}
if (pcmk_is_set(type, pe_order_restart)) {
handle_restart_ordering(first, then, filter);
}
if (then_flags != then->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
pe_rsc_trace(then->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'first' %s (%#.6x)",
then->uuid, pe__node_name(then->node),
then->flags, then_flags, first->uuid, first->flags);
if ((then->rsc != NULL) && (then->rsc->parent != NULL)) {
// Required to handle "X_stop then X_start" for cloned groups
pcmk__update_action_for_orderings(then, data_set);
}
}
if (first_flags != first->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
pe_rsc_trace(first->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'then' %s (%#.6x)",
first->uuid, pe__node_name(first->node),
first->flags, first_flags, then->uuid, then->flags);
}
return changed;
}
/*!
* \internal
* \brief Trace-log an action (optionally with its dependent actions)
*
* \param[in] pre_text If not NULL, prefix the log with this plus ": "
* \param[in] action Action to log
* \param[in] details If true, recursively log dependent actions
*/
void
pcmk__log_action(const char *pre_text, const pe_action_t *action, bool details)
{
const char *node_uname = NULL;
const char *node_uuid = NULL;
const char *desc = NULL;
CRM_CHECK(action != NULL, return);
- if (!pcmk_is_set(action->flags, pe_action_pseudo)) {
+ if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) {
if (action->node != NULL) {
node_uname = action->node->details->uname;
node_uuid = action->node->details->id;
} else {
node_uname = "<none>";
}
}
switch (text2task(action->task)) {
case pcmk_action_fence:
case pcmk_action_shutdown:
- if (pcmk_is_set(action->flags, pe_action_pseudo)) {
+ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
desc = "Pseudo ";
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
desc = "Optional ";
} else if (!pcmk_is_set(action->flags, pe_action_runnable)) {
desc = "!!Non-Startable!! ";
} else if (pcmk_is_set(action->flags, pe_action_processed)) {
desc = "";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
default:
if (pcmk_is_set(action->flags, pe_action_optional)) {
desc = "Optional ";
- } else if (pcmk_is_set(action->flags, pe_action_pseudo)) {
+ } else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
desc = "Pseudo ";
} else if (!pcmk_is_set(action->flags, pe_action_runnable)) {
desc = "!!Non-Startable!! ";
} else if (pcmk_is_set(action->flags, pe_action_processed)) {
desc = "";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(action->rsc? action->rsc->id : "<none>"),
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
}
if (details) {
const GList *iter = NULL;
const pe_action_wrapper_t *other = NULL;
crm_trace("\t\t====== Preceding Actions");
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
other = (const pe_action_wrapper_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== Subsequent Actions");
for (iter = action->actions_after; iter != NULL; iter = iter->next) {
other = (const pe_action_wrapper_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== End");
} else {
crm_trace("\t\t(before=%d, after=%d)",
g_list_length(action->actions_before),
g_list_length(action->actions_after));
}
}
/*!
* \internal
* \brief Create a new shutdown action for a node
*
* \param[in,out] node Node being shut down
*
* \return Newly created shutdown action for \p node
*/
pe_action_t *
pcmk__new_shutdown_action(pe_node_t *node)
{
char *shutdown_id = NULL;
pe_action_t *shutdown_op = NULL;
CRM_ASSERT(node != NULL);
shutdown_id = crm_strdup_printf("%s-%s", PCMK_ACTION_DO_SHUTDOWN,
node->details->uname);
shutdown_op = custom_action(NULL, shutdown_id, PCMK_ACTION_DO_SHUTDOWN,
node, FALSE, TRUE, node->details->data_set);
pcmk__order_stops_before_shutdown(node, shutdown_op);
add_hash_param(shutdown_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE);
return shutdown_op;
}
/*!
* \internal
* \brief Calculate and add an operation digest to XML
*
* Calculate an operation digest, which enables us to later determine when a
* restart is needed due to the resource's parameters being changed, and add it
* to given XML.
*
* \param[in] op Operation result from executor
* \param[in,out] update XML to add digest to
*/
static void
add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update)
{
char *digest = NULL;
xmlNode *args_xml = NULL;
if (op->params == NULL) {
return;
}
args_xml = create_xml_node(NULL, XML_TAG_PARAMS);
g_hash_table_foreach(op->params, hash2field, args_xml);
pcmk__filter_op_for_digest(args_xml);
digest = calculate_operation_digest(args_xml, NULL);
crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest);
free_xml(args_xml);
free(digest);
}
#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
/*!
* \internal
* \brief Create XML for resource operation history update
*
* \param[in,out] parent Parent XML node to add to
* \param[in,out] op Operation event data
* \param[in] caller_version DC feature set
* \param[in] target_rc Expected result of operation
* \param[in] node Name of node on which operation was performed
* \param[in] origin Arbitrary description of update source
*
* \return Newly created XML node for history update
*/
xmlNode *
pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op,
const char *caller_version, int target_rc,
const char *node, const char *origin)
{
char *key = NULL;
char *magic = NULL;
char *op_id = NULL;
char *op_id_additional = NULL;
char *local_user_data = NULL;
const char *exit_reason = NULL;
xmlNode *xml_op = NULL;
const char *task = NULL;
CRM_CHECK(op != NULL, return NULL);
crm_trace("Creating history XML for %s-interval %s action for %s on %s "
"(DC version: %s, origin: %s)",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
((node == NULL)? "no node" : node), caller_version, origin);
task = op->op_type;
/* Record a successful agent reload as a start, and a failed one as a
* monitor, to make life easier for the scheduler when determining the
* current state.
*
* @COMPAT We should check "reload" here only if the operation was for a
* pre-OCF-1.1 resource agent, but we don't know that here, and we should
* only ever get results for actions scheduled by us, so we can reasonably
* assume any "reload" is actually a pre-1.1 agent reload.
*/
if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
NULL)) {
if (op->op_status == PCMK_EXEC_DONE) {
task = PCMK_ACTION_START;
} else {
task = PCMK_ACTION_MONITOR;
}
}
key = pcmk__op_key(op->rsc_id, task, op->interval_ms);
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = crm_meta_value(op->params, "notify_type");
const char *n_task = crm_meta_value(op->params, "notify_operation");
CRM_LOG_ASSERT(n_type != NULL);
CRM_LOG_ASSERT(n_task != NULL);
op_id = pcmk__notify_key(op->rsc_id, n_type, n_task);
if (op->op_status != PCMK_EXEC_PENDING) {
/* Ignore notify errors.
*
* @TODO It might be better to keep the correct result here, and
* ignore it in process_graph_event().
*/
lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
}
/* Migration history is preserved separately, which usually matters for
* multiple nodes and is important for future cluster transitions.
*/
} else if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
op_id = strdup(key);
} else if (did_rsc_op_fail(op, target_rc)) {
op_id = pcmk__op_key(op->rsc_id, "last_failure", 0);
if (op->interval_ms == 0) {
// Ensure 'last' gets updated, in case record-pending is true
op_id_additional = pcmk__op_key(op->rsc_id, "last", 0);
}
exit_reason = op->exit_reason;
} else if (op->interval_ms > 0) {
op_id = strdup(key);
} else {
op_id = pcmk__op_key(op->rsc_id, "last", 0);
}
again:
xml_op = pcmk__xe_match(parent, XML_LRM_TAG_RSC_OP, XML_ATTR_ID, op_id);
if (xml_op == NULL) {
xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP);
}
if (op->user_data == NULL) {
crm_debug("Generating fake transition key for: " PCMK__OP_FMT
" %d from %s", op->rsc_id, op->op_type, op->interval_ms,
op->call_id, origin);
local_user_data = pcmk__transition_key(-1, op->call_id, target_rc,
FAKE_TE_ID);
op->user_data = local_user_data;
}
if (magic == NULL) {
magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc,
(const char *) op->user_data);
}
crm_xml_add(xml_op, XML_ATTR_ID, op_id);
crm_xml_add(xml_op, XML_LRM_ATTR_TASK_KEY, key);
crm_xml_add(xml_op, XML_LRM_ATTR_TASK, task);
crm_xml_add(xml_op, XML_ATTR_ORIGIN, origin);
crm_xml_add(xml_op, XML_ATTR_CRM_VERSION, caller_version);
crm_xml_add(xml_op, XML_ATTR_TRANSITION_KEY, op->user_data);
crm_xml_add(xml_op, XML_ATTR_TRANSITION_MAGIC, magic);
crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, pcmk__s(exit_reason, ""));
crm_xml_add(xml_op, XML_LRM_ATTR_TARGET, node); // For context during triage
crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id);
crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc);
crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status);
crm_xml_add_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, op->interval_ms);
if (compare_version("2.1", caller_version) <= 0) {
if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) {
crm_trace("Timing data (" PCMK__OP_FMT
"): last=%u change=%u exec=%u queue=%u",
op->rsc_id, op->op_type, op->interval_ms,
op->t_run, op->t_rcchange, op->exec_time, op->queue_time);
if ((op->interval_ms != 0) && (op->t_rcchange != 0)) {
// Recurring ops may have changed rc after initial run
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_rcchange);
} else {
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_run);
}
crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time);
crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time);
}
}
if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/*
* Record migrate_source and migrate_target always for migrate ops.
*/
const char *name = XML_LRM_ATTR_MIGRATE_SOURCE;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
name = XML_LRM_ATTR_MIGRATE_TARGET;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
}
add_op_digest_to_xml(op, xml_op);
if (op_id_additional) {
free(op_id);
op_id = op_id_additional;
op_id_additional = NULL;
goto again;
}
if (local_user_data) {
free(local_user_data);
op->user_data = NULL;
}
free(magic);
free(op_id);
free(key);
return xml_op;
}
/*!
* \internal
* \brief Check whether an action shutdown-locks a resource to a node
*
* If the shutdown-lock cluster property is set, resources will not be recovered
* on a different node if cleanly stopped, and may start only on that same node.
* This function checks whether that applies to a given action, so that the
* transition graph can be marked appropriately.
*
* \param[in] action Action to check
*
* \return true if \p action locks its resource to the action's node,
* otherwise false
*/
bool
pcmk__action_locks_rsc_to_node(const pe_action_t *action)
{
// Only resource actions taking place on resource's lock node are locked
if ((action == NULL) || (action->rsc == NULL)
|| !pe__same_node(action->node, action->rsc->lock_node)) {
return false;
}
/* During shutdown, only stops are locked (otherwise, another action such as
* a demote would cause the controller to clear the lock)
*/
if (action->node->details->shutdown && (action->task != NULL)
&& (strcmp(action->task, PCMK_ACTION_STOP) != 0)) {
return false;
}
return true;
}
/* lowest to highest */
static gint
sort_action_id(gconstpointer a, gconstpointer b)
{
const pe_action_wrapper_t *action_wrapper2 = (const pe_action_wrapper_t *)a;
const pe_action_wrapper_t *action_wrapper1 = (const pe_action_wrapper_t *)b;
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (action_wrapper1->action->id < action_wrapper2->action->id) {
return 1;
}
if (action_wrapper1->action->id > action_wrapper2->action->id) {
return -1;
}
return 0;
}
/*!
* \internal
* \brief Remove any duplicate action inputs, merging action flags
*
* \param[in,out] action Action whose inputs should be checked
*/
void
pcmk__deduplicate_action_inputs(pe_action_t *action)
{
GList *item = NULL;
GList *next = NULL;
pe_action_wrapper_t *last_input = NULL;
action->actions_before = g_list_sort(action->actions_before,
sort_action_id);
for (item = action->actions_before; item != NULL; item = next) {
pe_action_wrapper_t *input = (pe_action_wrapper_t *) item->data;
next = item->next;
if ((last_input != NULL)
&& (input->action->id == last_input->action->id)) {
crm_trace("Input %s (%d) duplicate skipped for action %s (%d)",
input->action->uuid, input->action->id,
action->uuid, action->id);
/* For the purposes of scheduling, the ordering flags no longer
* matter, but crm_simulate looks at certain ones when creating a
* dot graph. Combining the flags is sufficient for that purpose.
*/
last_input->type |= input->type;
if (input->state == pe_link_dumped) {
last_input->state = pe_link_dumped;
}
free(item->data);
action->actions_before = g_list_delete_link(action->actions_before,
item);
} else {
last_input = input;
input->state = pe_link_not_dumped;
}
}
}
/*!
* \internal
* \brief Output all scheduled actions
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__output_actions(pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
// Output node (non-resource) actions
for (GList *iter = data_set->actions; iter != NULL; iter = iter->next) {
char *node_name = NULL;
char *task = NULL;
pe_action_t *action = (pe_action_t *) iter->data;
if (action->rsc != NULL) {
continue; // Resource actions will be output later
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
continue; // This action was not scheduled
}
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
task = strdup("Shutdown");
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta,
"stonith_action");
task = crm_strdup_printf("Fence (%s)", op);
} else {
continue; // Don't display other node action types
}
if (pe__is_guest_node(action->node)) {
const pe_resource_t *remote = action->node->details->remote_rsc;
node_name = crm_strdup_printf("%s (resource: %s)",
pe__node_name(action->node),
remote->container->id);
} else if (action->node != NULL) {
node_name = crm_strdup_printf("%s", pe__node_name(action->node));
}
out->message(out, "node-action", task, node_name, action->reason);
free(node_name);
free(task);
}
// Output resource actions
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->output_actions(rsc);
}
}
/*!
* \internal
* \brief Check whether action from resource history is still in configuration
*
* \param[in] rsc Resource that action is for
* \param[in] task Action's name
* \param[in] interval_ms Action's interval (in milliseconds)
*
* \return true if action is still in resource configuration, otherwise false
*/
static bool
action_in_config(const pe_resource_t *rsc, const char *task, guint interval_ms)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
bool config = (find_rsc_op_entry(rsc, key) != NULL);
free(key);
return config;
}
/*!
* \internal
* \brief Get action name needed to compare digest for configuration changes
*
* \param[in] task Action name from history
* \param[in] interval_ms Action interval (in milliseconds)
*
* \return Action name whose digest should be compared
*/
static const char *
task_for_digest(const char *task, guint interval_ms)
{
/* Certain actions need to be compared against the parameters used to start
* the resource.
*/
if ((interval_ms == 0)
&& pcmk__str_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_PROMOTE, NULL)) {
task = PCMK_ACTION_START;
}
return task;
}
/*!
* \internal
* \brief Check whether only sanitized parameters to an action changed
*
* When collecting CIB files for troubleshooting, crm_report will mask
* sensitive resource parameters. If simulations were run using that, affected
* resources would appear to need a restart, which would complicate
* troubleshooting. To avoid that, we save a "secure digest" of non-sensitive
* parameters. This function used that digest to check whether only masked
* parameters are different.
*
* \param[in] xml_op Resource history entry with secure digest
* \param[in] digest_data Operation digest information being compared
* \param[in] data_set Cluster working set
*
* \return true if only sanitized parameters changed, otherwise false
*/
static bool
only_sanitized_changed(const xmlNode *xml_op,
const op_digest_cache_t *digest_data,
const pe_working_set_t *data_set)
{
const char *digest_secure = NULL;
if (!pcmk_is_set(data_set->flags, pcmk_sched_sanitized)) {
// The scheduler is not being run as a simulation
return false;
}
digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST);
return (digest_data->rc != RSC_DIGEST_MATCH) && (digest_secure != NULL)
&& (digest_data->digest_secure_calc != NULL)
&& (strcmp(digest_data->digest_secure_calc, digest_secure) == 0);
}
/*!
* \internal
* \brief Force a restart due to a configuration change
*
* \param[in,out] rsc Resource that action is for
* \param[in] task Name of action whose configuration changed
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in,out] node Node where resource should be restarted
*/
static void
force_restart(pe_resource_t *rsc, const char *task, guint interval_ms,
pe_node_t *node)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
pe_action_t *required = custom_action(rsc, key, task, NULL, FALSE, TRUE,
rsc->cluster);
pe_action_set_reason(required, "resource definition change", true);
trigger_unfencing(rsc, node, "Device parameters changed", NULL,
rsc->cluster);
}
/*!
* \internal
* \brief Schedule a reload of a resource on a node
*
* \param[in,out] data Resource to reload
* \param[in] user_data Where resource should be reloaded
*/
static void
schedule_reload(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
pe_action_t *reload = NULL;
// For collective resources, just call recursively for children
if (rsc->variant > pcmk_rsc_variant_primitive) {
g_list_foreach(rsc->children, schedule_reload, user_data);
return;
}
// Skip the reload in certain situations
if ((node == NULL)
|| !pcmk_is_set(rsc->flags, pcmk_rsc_managed)
|| pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pe_rsc_trace(rsc, "Skip reload of %s:%s%s %s",
rsc->id,
pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : " unmanaged",
pcmk_is_set(rsc->flags, pcmk_rsc_failed)? " failed" : "",
(node == NULL)? "inactive" : node->details->uname);
return;
}
/* If a resource's configuration changed while a start was pending,
* force a full restart instead of a reload.
*/
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
pe_rsc_trace(rsc, "%s: preventing agent reload because start pending",
rsc->id);
custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, TRUE,
rsc->cluster);
return;
}
// Schedule the reload
pe__set_resource_flags(rsc, pcmk_rsc_reload);
reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node,
FALSE, TRUE, rsc->cluster);
pe_action_set_reason(reload, "resource definition change", FALSE);
// Set orderings so that a required stop or demote cancels the reload
pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
rsc->cluster);
pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL,
pe_order_optional|pe_order_then_cancels_first,
rsc->cluster);
}
/*!
* \internal
* \brief Handle any configuration change for an action
*
* Given an action from resource history, if the resource's configuration
* changed since the action was done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, etc.).
*
* \param[in,out] rsc Resource that action is for
* \param[in,out] node Node that action was on
* \param[in] xml_op Action XML from resource history
*
* \return true if action configuration changed, otherwise false
*/
bool
pcmk__check_action_config(pe_resource_t *rsc, pe_node_t *node,
const xmlNode *xml_op)
{
guint interval_ms = 0;
const char *task = NULL;
const op_digest_cache_t *digest_data = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL),
return false);
task = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
CRM_CHECK(task != NULL, return false);
crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
// If this is a recurring action, check whether it has been orphaned
if (interval_ms > 0) {
if (action_in_config(rsc, task, interval_ms)) {
pe_rsc_trace(rsc, "%s-interval %s for %s on %s is in configuration",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_cancel_removed_actions)) {
pcmk__schedule_cancel(rsc,
crm_element_value(xml_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "orphan");
return true;
} else {
pe_rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
return true;
}
}
crm_trace("Checking %s-interval %s for %s on %s for configuration changes",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
task = task_for_digest(task, interval_ms);
digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->cluster);
if (only_sanitized_changed(xml_op, digest_data, rsc->cluster)) {
if (!pcmk__is_daemon && (rsc->cluster->priv != NULL)) {
pcmk__output_t *out = rsc->cluster->priv;
out->info(out,
"Only 'private' parameters to %s-interval %s for %s "
"on %s changed: %s",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node),
crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC));
}
return false;
}
switch (digest_data->rc) {
case RSC_DIGEST_RESTART:
crm_log_xml_debug(digest_data->params_restart, "params:restart");
force_restart(rsc, task, interval_ms, node);
return true;
case RSC_DIGEST_ALL:
case RSC_DIGEST_UNKNOWN:
// Changes that can potentially be handled by an agent reload
if (interval_ms > 0) {
/* Recurring actions aren't reloaded per se, they are just
* re-scheduled so the next run uses the new parameters.
* The old instance will be cancelled automatically.
*/
crm_log_xml_debug(digest_data->params_all, "params:reschedule");
pcmk__reschedule_recurring(rsc, task, interval_ms, node);
} else if (crm_element_value(xml_op,
XML_LRM_ATTR_RESTART_DIGEST) != NULL) {
// Agent supports reload, so use it
trigger_unfencing(rsc, node,
"Device parameters changed (reload)", NULL,
rsc->cluster);
crm_log_xml_debug(digest_data->params_all, "params:reload");
schedule_reload((gpointer) rsc, (gpointer) node);
} else {
pe_rsc_trace(rsc,
"Restarting %s "
"because agent doesn't support reload", rsc->id);
crm_log_xml_debug(digest_data->params_restart,
"params:restart");
force_restart(rsc, task, interval_ms, node);
}
return true;
default:
break;
}
return false;
}
/*!
* \internal
* \brief Create a list of resource's action history entries, sorted by call ID
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[out] start_index Where to store index of start-like action, if any
* \param[out] stop_index Where to store index of stop action, if any
*/
static GList *
rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index)
{
GList *ops = NULL;
for (xmlNode *rsc_op = first_named_child(rsc_entry, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
ops = g_list_prepend(ops, rsc_op);
}
ops = g_list_sort(ops, sort_op_by_callid);
calculate_active_ops(ops, start_index, stop_index);
return ops;
}
/*!
* \internal
* \brief Process a resource's action history from the CIB status
*
* Given a resource's action history, if the resource's configuration
* changed since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[in,out] rsc Resource whose history is being processed
* \param[in,out] node Node whose history is being processed
*/
static void
process_rsc_history(const xmlNode *rsc_entry, pe_resource_t *rsc,
pe_node_t *node)
{
int offset = -1;
int stop_index = 0;
int start_index = 0;
GList *sorted_op_list = NULL;
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
if (pe_rsc_is_anon_clone(pe__const_top_resource(rsc, false))) {
pe_rsc_trace(rsc,
"Skipping configuration check "
"for orphaned clone instance %s",
rsc->id);
} else {
pe_rsc_trace(rsc,
"Skipping configuration check and scheduling clean-up "
"for orphaned resource %s", rsc->id);
pcmk__schedule_cleanup(rsc, node, false);
}
return;
}
if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) {
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) {
pcmk__schedule_cleanup(rsc, node, false);
}
pe_rsc_trace(rsc,
"Skipping configuration check for %s "
"because no longer active on %s",
rsc->id, pe__node_name(node));
return;
}
pe_rsc_trace(rsc, "Checking for configuration changes for %s on %s",
rsc->id, pe__node_name(node));
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) {
pcmk__schedule_cleanup(rsc, node, false);
}
sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index);
if (start_index < stop_index) {
return; // Resource is stopped
}
for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
xmlNode *rsc_op = (xmlNode *) iter->data;
const char *task = NULL;
guint interval_ms = 0;
if (++offset < start_index) {
// Skip actions that happened before a start
continue;
}
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms);
if ((interval_ms > 0)
&& (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)
|| node->details->maintenance)) {
// Maintenance mode cancels recurring operations
pcmk__schedule_cancel(rsc,
crm_element_value(rsc_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "maintenance mode");
} else if ((interval_ms > 0)
|| pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START,
PCMK_ACTION_PROMOTE,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* If a resource operation failed, and the operation's definition
* has changed, clear any fail count so they can be retried fresh.
*/
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't assigned resources to nodes yet, so if the
* REMOTE_CONTAINER_HACK is used, we may calculate the digest
* based on the literal "#uname" value rather than the properly
* substituted value. That would mistakenly make the action
* definition appear to have been changed. Defer the check until
* later in this case.
*/
pe__add_param_check(rsc_op, rsc, node, pcmk__check_active,
rsc->cluster);
} else if (pcmk__check_action_config(rsc, node, rsc_op)
&& (pe_get_failcount(node, rsc, NULL, pe_fc_effective,
NULL) != 0)) {
pe__clear_failcount(rsc, node, "action definition changed",
rsc->cluster);
}
}
}
g_list_free(sorted_op_list);
}
/*!
* \internal
* \brief Process a node's action history from the CIB status
*
* Given a node's resource history, if the resource's configuration changed
* since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] node Node whose history is being processed
* \param[in] lrm_rscs Node's <lrm_resources> from CIB status XML
*/
static void
process_node_history(pe_node_t *node, const xmlNode *lrm_rscs)
{
crm_trace("Processing node history for %s", pe__node_name(node));
for (const xmlNode *rsc_entry = first_named_child(lrm_rscs,
XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
if (rsc_entry->children != NULL) {
GList *result = pcmk__rscs_matching_id(ID(rsc_entry),
node->details->data_set);
for (GList *iter = result; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (rsc->variant == pcmk_rsc_variant_primitive) {
process_rsc_history(rsc_entry, rsc, node);
}
}
g_list_free(result);
}
}
}
// XPath to find a node's resource history
#define XPATH_NODE_HISTORY "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \
"/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES
/*!
* \internal
* \brief Process any resource configuration changes in the CIB status
*
* Go through all nodes' resource history, and if a resource's configuration
* changed since its actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__handle_rsc_config_changes(pe_working_set_t *data_set)
{
crm_trace("Check resource and action configuration for changes");
/* Rather than iterate through the status section, iterate through the nodes
* and search for the appropriate status subsection for each. This skips
* orphaned nodes and lets us eliminate some cases before searching the XML.
*/
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
/* Don't bother checking actions for a node that can't run actions ...
* unless it's in maintenance mode, in which case we still need to
* cancel any existing recurring monitors.
*/
if (node->details->maintenance
|| pcmk__node_available(node, false, false)) {
char *xpath = NULL;
xmlNode *history = NULL;
xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->details->uname);
history = get_xpath_object(xpath, data_set->input, LOG_NEVER);
free(xpath);
process_node_history(node, history);
}
}
}
diff --git a/lib/pacemaker/pcmk_sched_fencing.c b/lib/pacemaker/pcmk_sched_fencing.c
index e3819dd138..e711d4a586 100644
--- a/lib/pacemaker/pcmk_sched_fencing.c
+++ b/lib/pacemaker/pcmk_sched_fencing.c
@@ -1,496 +1,496 @@
/*
* 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 Check whether a resource is known on a particular node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return TRUE if resource (or parent if an anonymous clone) is known
*/
static bool
rsc_is_known_on(const pe_resource_t *rsc, const pe_node_t *node)
{
if (g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) {
return TRUE;
} else if ((rsc->variant == pcmk_rsc_variant_primitive)
&& pe_rsc_is_anon_clone(rsc->parent)
&& (g_hash_table_lookup(rsc->parent->known_on,
node->details->id) != NULL)) {
/* We check only the parent, not the uber-parent, because we cannot
* assume that the resource is known if it is in an anonymously cloned
* group (which may be only partially known).
*/
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Order a resource's start and promote actions relative to fencing
*
* \param[in,out] rsc Resource to be ordered
* \param[in,out] stonith_op Fence action
*/
static void
order_start_vs_fencing(pe_resource_t *rsc, pe_action_t *stonith_op)
{
pe_node_t *target;
CRM_CHECK(stonith_op && stonith_op->node, return);
target = stonith_op->node;
for (GList *iter = rsc->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = iter->data;
switch (action->needs) {
case pcmk_requires_nothing:
// Anything other than start or promote requires nothing
break;
case pcmk_requires_fencing:
order_actions(stonith_op, action, pe_order_optional);
break;
case pcmk_requires_quorum:
if (pcmk__str_eq(action->task, PCMK_ACTION_START,
pcmk__str_none)
&& (g_hash_table_lookup(rsc->allowed_nodes,
target->details->id) != NULL)
&& !rsc_is_known_on(rsc, target)) {
/* If we don't know the status of the resource on the node
* we're about to shoot, we have to assume it may be active
* there. Order the resource start after the fencing. This
* is analogous to waiting for all the probes for a resource
* to complete before starting it.
*
* The most likely explanation is that the DC died and took
* its status with it.
*/
pe_rsc_debug(rsc, "Ordering %s after %s recovery",
action->uuid, pe__node_name(target));
order_actions(stonith_op, action,
pe_order_optional | pe_order_runnable_left);
}
break;
}
}
}
/*!
* \internal
* \brief Order a resource's stop and demote actions relative to fencing
*
* \param[in,out] rsc Resource to be ordered
* \param[in,out] stonith_op Fence action
*/
static void
order_stop_vs_fencing(pe_resource_t *rsc, pe_action_t *stonith_op)
{
GList *iter = NULL;
GList *action_list = NULL;
bool order_implicit = false;
pe_resource_t *top = uber_parent(rsc);
pe_action_t *parent_stop = NULL;
pe_node_t *target;
CRM_CHECK(stonith_op && stonith_op->node, return);
target = stonith_op->node;
/* Get a list of stop actions potentially implied by the fencing */
action_list = pe__resource_actions(rsc, target, PCMK_ACTION_STOP, FALSE);
/* If resource requires fencing, implicit actions must occur after fencing.
*
* Implied stops and demotes of resources running on guest nodes are always
* ordered after fencing, even if the resource does not require fencing,
* because guest node "fencing" is actually just a resource stop.
*/
if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)
|| pe__is_guest_node(target)) {
order_implicit = true;
}
if (action_list && order_implicit) {
parent_stop = find_first_action(top->actions, NULL, PCMK_ACTION_STOP,
NULL);
}
for (iter = action_list; iter != NULL; iter = iter->next) {
pe_action_t *action = iter->data;
// The stop would never complete, so convert it into a pseudo-action.
- pe__set_action_flags(action, pe_action_pseudo|pe_action_runnable);
+ pe__set_action_flags(action, pcmk_action_pseudo|pe_action_runnable);
if (order_implicit) {
pe__set_action_flags(action, pe_action_implied_by_stonith);
/* Order the stonith before the parent stop (if any).
*
* Also order the stonith before the resource stop, unless the
* resource is inside a bundle -- that would cause a graph loop.
* We can rely on the parent stop's ordering instead.
*
* User constraints must not order a resource in a guest node
* relative to the guest node container resource. The
* pe_order_preserve flag marks constraints as generated by the
* cluster and thus immune to that check (and is irrelevant if
* target is not a guest).
*/
if (!pe_rsc_is_bundled(rsc)) {
order_actions(stonith_op, action, pe_order_preserve);
}
order_actions(stonith_op, parent_stop, pe_order_preserve);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
crm_notice("Stop of failed resource %s is implicit %s %s is fenced",
rsc->id, (order_implicit? "after" : "because"),
pe__node_name(target));
} else {
crm_info("%s is implicit %s %s is fenced",
action->uuid, (order_implicit? "after" : "because"),
pe__node_name(target));
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_notify)) {
pe__order_notifs_after_fencing(action, rsc, stonith_op);
}
#if 0
/* It might be a good idea to stop healthy resources on a node about to
* be fenced, when possible.
*
* However, fencing must be done before a failed resource's
* (pseudo-)stop action, so that could create a loop. For example, given
* a group of A and B running on node N with a failed stop of B:
*
* fence N -> stop B (pseudo-op) -> stop A -> fence N
*
* The block below creates the stop A -> fence N ordering and therefore
* must (at least for now) be disabled. Instead, run the block above and
* treat all resources on N as B would be (i.e., as a pseudo-op after
* the fencing).
*
* @TODO Maybe break the "A requires B" dependency in
* pcmk__update_action_for_orderings() and use this block for healthy
* resources instead of the above.
*/
crm_info("Moving healthy resource %s off %s before fencing",
rsc->id, pe__node_name(node));
pcmk__new_ordering(rsc, stop_key(rsc), NULL, NULL,
strdup(PCMK_ACTION_STONITH), stonith_op,
pe_order_optional, rsc->cluster);
#endif
}
g_list_free(action_list);
/* Get a list of demote actions potentially implied by the fencing */
action_list = pe__resource_actions(rsc, target, PCMK_ACTION_DEMOTE, FALSE);
for (iter = action_list; iter != NULL; iter = iter->next) {
pe_action_t *action = iter->data;
if (!(action->node->details->online) || action->node->details->unclean
|| pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pe_rsc_info(rsc,
"Demote of failed resource %s is implicit "
"after %s is fenced",
rsc->id, pe__node_name(target));
} else {
pe_rsc_info(rsc, "%s is implicit after %s is fenced",
action->uuid, pe__node_name(target));
}
/* The demote would never complete and is now implied by the
* fencing, so convert it into a pseudo-action.
*/
- pe__set_action_flags(action, pe_action_pseudo|pe_action_runnable);
+ pe__set_action_flags(action, pcmk_action_pseudo|pe_action_runnable);
if (pe_rsc_is_bundled(rsc)) {
// Recovery will be ordered as usual after parent's implied stop
} else if (order_implicit) {
order_actions(stonith_op, action,
pe_order_preserve|pe_order_optional);
}
}
}
g_list_free(action_list);
}
/*!
* \internal
* \brief Order resource actions properly relative to fencing
*
* \param[in,out] rsc Resource whose actions should be ordered
* \param[in,out] stonith_op Fencing operation to be ordered against
*/
static void
rsc_stonith_ordering(pe_resource_t *rsc, pe_action_t *stonith_op)
{
if (rsc->children) {
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child_rsc = iter->data;
rsc_stonith_ordering(child_rsc, stonith_op);
}
} else if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pe_rsc_trace(rsc,
"Skipping fencing constraints for unmanaged resource: %s",
rsc->id);
} else {
order_start_vs_fencing(rsc, stonith_op);
order_stop_vs_fencing(rsc, stonith_op);
}
}
/*!
* \internal
* \brief Order all actions appropriately relative to a fencing operation
*
* Ensure start operations of affected resources are ordered after fencing,
* imply stop and demote operations of affected resources by marking them as
* pseudo-actions, etc.
*
* \param[in,out] stonith_op Fencing operation
* \param[in,out] data_set Working set of cluster
*/
void
pcmk__order_vs_fence(pe_action_t *stonith_op, pe_working_set_t *data_set)
{
CRM_CHECK(stonith_op && data_set, return);
for (GList *r = data_set->resources; r != NULL; r = r->next) {
rsc_stonith_ordering((pe_resource_t *) r->data, stonith_op);
}
}
/*!
* \internal
* \brief Order an action after unfencing
*
* \param[in] rsc Resource that action is for
* \param[in,out] node Node that action is on
* \param[in,out] action Action to be ordered after unfencing
* \param[in] order Ordering flags
*/
void
pcmk__order_vs_unfence(const pe_resource_t *rsc, pe_node_t *node,
pe_action_t *action, enum pe_ordering order)
{
/* When unfencing is in use, we order unfence actions before any probe or
* start of resources that require unfencing, and also of fence devices.
*
* This might seem to violate the principle that fence devices require
* only quorum. However, fence agents that unfence often don't have enough
* information to even probe or start unless the node is first unfenced.
*/
if ((pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags, pcmk_sched_enable_unfencing))
|| pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) {
/* Start with an optional ordering. Requiring unfencing would result in
* the node being unfenced, and all its resources being stopped,
* whenever a new resource is added -- which would be highly suboptimal.
*/
pe_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, TRUE, NULL,
FALSE, node->details->data_set);
order_actions(unfence, action, order);
if (!pcmk__node_unfenced(node)) {
// But unfencing is required if it has never been done
char *reason = crm_strdup_printf("required by %s %s",
rsc->id, action->task);
trigger_unfencing(NULL, node, reason, NULL,
node->details->data_set);
free(reason);
}
}
}
/*!
* \internal
* \brief Create pseudo-op for guest node fence, and order relative to it
*
* \param[in,out] node Guest node to fence
*/
void
pcmk__fence_guest(pe_node_t *node)
{
pe_resource_t *container = NULL;
pe_action_t *stop = NULL;
pe_action_t *stonith_op = NULL;
/* The fence action is just a label; we don't do anything differently for
* off vs. reboot. We specify it explicitly, rather than let it default to
* cluster's default action, because we are not _initiating_ fencing -- we
* are creating a pseudo-event to describe fencing that is already occurring
* by other means (container recovery).
*/
const char *fence_action = PCMK_ACTION_OFF;
CRM_ASSERT(node != NULL);
/* Check whether guest's container resource has any explicit stop or
* start (the stop may be implied by fencing of the guest's host).
*/
container = node->details->remote_rsc->container;
if (container) {
stop = find_first_action(container->actions, NULL, PCMK_ACTION_STOP,
NULL);
if (find_first_action(container->actions, NULL, PCMK_ACTION_START,
NULL)) {
fence_action = PCMK_ACTION_REBOOT;
}
}
/* Create a fence pseudo-event, so we have an event to order actions
* against, and the controller can always detect it.
*/
stonith_op = pe_fence_op(node, fence_action, FALSE, "guest is unclean",
FALSE, node->details->data_set);
- pe__set_action_flags(stonith_op, pe_action_pseudo|pe_action_runnable);
+ pe__set_action_flags(stonith_op, pcmk_action_pseudo|pe_action_runnable);
/* We want to imply stops/demotes after the guest is stopped, not wait until
* it is restarted, so we always order pseudo-fencing after stop, not start
* (even though start might be closer to what is done for a real reboot).
*/
- if ((stop != NULL) && pcmk_is_set(stop->flags, pe_action_pseudo)) {
+ if ((stop != NULL) && pcmk_is_set(stop->flags, pcmk_action_pseudo)) {
pe_action_t *parent_stonith_op = pe_fence_op(stop->node, NULL, FALSE,
NULL, FALSE,
node->details->data_set);
crm_info("Implying guest %s is down (action %d) after %s fencing",
pe__node_name(node), stonith_op->id,
pe__node_name(stop->node));
order_actions(parent_stonith_op, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
} else if (stop) {
order_actions(stop, stonith_op,
pe_order_runnable_left|pe_order_implies_then);
crm_info("Implying guest %s is down (action %d) "
"after container %s is stopped (action %d)",
pe__node_name(node), stonith_op->id,
container->id, stop->id);
} else {
/* If we're fencing the guest node but there's no stop for the guest
* resource, we must think the guest is already stopped. However, we may
* think so because its resource history was just cleaned. To avoid
* unnecessarily considering the guest node down if it's really up,
* order the pseudo-fencing after any stop of the connection resource,
* which will be ordered after any container (re-)probe.
*/
stop = find_first_action(node->details->remote_rsc->actions, NULL,
PCMK_ACTION_STOP, NULL);
if (stop) {
order_actions(stop, stonith_op, pe_order_optional);
crm_info("Implying guest %s is down (action %d) "
"after connection is stopped (action %d)",
pe__node_name(node), stonith_op->id, stop->id);
} else {
/* Not sure why we're fencing, but everything must already be
* cleanly stopped.
*/
crm_info("Implying guest %s is down (action %d) ",
pe__node_name(node), stonith_op->id);
}
}
// Order/imply other actions relative to pseudo-fence as with real fence
pcmk__order_vs_fence(stonith_op, node->details->data_set);
}
/*!
* \internal
* \brief Check whether node has already been unfenced
*
* \param[in] node Node to check
*
* \return true if node has a nonzero #node-unfenced attribute (or none),
* otherwise false
*/
bool
pcmk__node_unfenced(const pe_node_t *node)
{
const char *unfenced = pe_node_attribute_raw(node, CRM_ATTR_UNFENCED);
return !pcmk__str_eq(unfenced, "0", pcmk__str_null_matches);
}
/*!
* \internal
* \brief Order a resource's start and stop relative to unfencing of a node
*
* \param[in,out] data Node that could be unfenced
* \param[in,out] user_data Resource to order
*/
void
pcmk__order_restart_vs_unfence(gpointer data, gpointer user_data)
{
pe_node_t *node = (pe_node_t *) data;
pe_resource_t *rsc = (pe_resource_t *) user_data;
pe_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, true, NULL, false,
rsc->cluster);
crm_debug("Ordering any stops of %s before %s, and any starts after",
rsc->id, unfence->uuid);
/*
* It would be more efficient to order clone resources once,
* rather than order each instance, but ordering the instance
* allows us to avoid unnecessary dependencies that might conflict
* with user constraints.
*
* @TODO: This constraint can still produce a transition loop if the
* resource has a stop scheduled on the node being unfenced, and
* there is a user ordering constraint to start some other resource
* (which will be ordered after the unfence) before stopping this
* resource. An example is "start some slow-starting cloned service
* before stopping an associated virtual IP that may be moving to
* it":
* stop this -> unfencing -> start that -> stop this
*/
pcmk__new_ordering(rsc, stop_key(rsc), NULL,
NULL, strdup(unfence->uuid), unfence,
pe_order_optional|pe_order_same_node,
rsc->cluster);
pcmk__new_ordering(NULL, strdup(unfence->uuid), unfence,
rsc, start_key(rsc), NULL,
pe_order_implies_then_on_node|pe_order_same_node,
rsc->cluster);
}
diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c
index 0dc307d6c5..98c6f0c2a6 100644
--- a/lib/pacemaker/pcmk_sched_group.c
+++ b/lib/pacemaker/pcmk_sched_group.c
@@ -1,948 +1,948 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Assign a group resource to a node
*
* \param[in,out] rsc Group 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 a child of \p rsc can't be
* assigned to a node, set the child's next role to
* stopped and update existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pe_node_t *
pcmk__group_assign(pe_resource_t *rsc, const pe_node_t *prefer,
bool stop_if_fail)
{
pe_node_t *first_assigned_node = NULL;
pe_resource_t *first_member = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return rsc->allocated_to; // Assignment already done
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) {
pe_rsc_debug(rsc, "Assignment dependency loop detected involving %s",
rsc->id);
return NULL;
}
if (rsc->children == NULL) {
// No members to assign
pe__clear_resource_flags(rsc, pcmk_rsc_unassigned);
return NULL;
}
pe__set_resource_flags(rsc, pcmk_rsc_assigning);
first_member = (pe_resource_t *) rsc->children->data;
rsc->role = first_member->role;
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->allowed_nodes, rsc->cluster);
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
pe_node_t *node = NULL;
pe_rsc_trace(rsc, "Assigning group %s member %s",
rsc->id, member->id);
node = member->cmds->assign(member, prefer, stop_if_fail);
if (first_assigned_node == NULL) {
first_assigned_node = node;
}
}
pe__set_next_role(rsc, first_member->next_role, "first group member");
pe__clear_resource_flags(rsc, pcmk_rsc_assigning|pcmk_rsc_unassigned);
if (!pe__group_flag_is_set(rsc, pe__group_colocated)) {
return NULL;
}
return first_assigned_node;
}
/*!
* \internal
* \brief Create a pseudo-operation for a group as an ordering point
*
* \param[in,out] group Group resource to create action for
* \param[in] action Action name
*
* \return Newly created pseudo-operation
*/
static pe_action_t *
create_group_pseudo_op(pe_resource_t *group, const char *action)
{
pe_action_t *op = custom_action(group, pcmk__op_key(group->id, action, 0),
action, NULL, TRUE, TRUE, group->cluster);
- pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable);
+ pe__set_action_flags(op, pcmk_action_pseudo|pe_action_runnable);
return op;
}
/*!
* \internal
* \brief Create all actions needed for a given group resource
*
* \param[in,out] rsc Group resource to create actions for
*/
void
pcmk__group_create_actions(pe_resource_t *rsc)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group));
pe_rsc_trace(rsc, "Creating actions for group %s", rsc->id);
// Create actions for individual group members
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
member->cmds->create_actions(member);
}
// Create pseudo-actions for group itself to serve as ordering points
create_group_pseudo_op(rsc, PCMK_ACTION_START);
create_group_pseudo_op(rsc, PCMK_ACTION_RUNNING);
create_group_pseudo_op(rsc, PCMK_ACTION_STOP);
create_group_pseudo_op(rsc, PCMK_ACTION_STOPPED);
if (crm_is_true(g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_PROMOTABLE))) {
create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTE);
create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTED);
create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTE);
create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTED);
}
}
// User data for member_internal_constraints()
struct member_data {
// These could be derived from member but this avoids some function calls
bool ordered;
bool colocated;
bool promotable;
pe_resource_t *last_active;
pe_resource_t *previous_member;
};
/*!
* \internal
* \brief Create implicit constraints needed for a group member
*
* \param[in,out] data Group member to create implicit constraints for
* \param[in,out] user_data Member data (struct member_data *)
*/
static void
member_internal_constraints(gpointer data, gpointer user_data)
{
pe_resource_t *member = (pe_resource_t *) data;
struct member_data *member_data = (struct member_data *) user_data;
// For ordering demote vs demote or stop vs stop
uint32_t down_flags = pe_order_implies_first_printed;
// For ordering demote vs demoted or stop vs stopped
uint32_t post_down_flags = pe_order_implies_then_printed;
// Create the individual member's implicit constraints
member->cmds->internal_constraints(member);
if (member_data->previous_member == NULL) {
// This is first member
if (member_data->ordered) {
pe__set_order_flags(down_flags, pe_order_optional);
post_down_flags = pe_order_implies_then;
}
} else if (member_data->colocated) {
uint32_t flags = pcmk__coloc_none;
if (pcmk_is_set(member->flags, pcmk_rsc_critical)) {
flags |= pcmk__coloc_influence;
}
// Colocate this member with the previous one
pcmk__new_colocation("#group-members", NULL, INFINITY, member,
member_data->previous_member, NULL, NULL, flags);
}
if (member_data->promotable) {
// Demote group -> demote member -> group is demoted
pcmk__order_resource_actions(member->parent, PCMK_ACTION_DEMOTE,
member, PCMK_ACTION_DEMOTE, down_flags);
pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE,
member->parent, PCMK_ACTION_DEMOTED,
post_down_flags);
// Promote group -> promote member -> group is promoted
pcmk__order_resource_actions(member, PCMK_ACTION_PROMOTE,
member->parent, PCMK_ACTION_PROMOTED,
pe_order_runnable_left
|pe_order_implies_then
|pe_order_implies_then_printed);
pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE,
member, PCMK_ACTION_PROMOTE,
pe_order_implies_first_printed);
}
// Stop group -> stop member -> group is stopped
pcmk__order_stops(member->parent, member, down_flags);
pcmk__order_resource_actions(member, PCMK_ACTION_STOP,
member->parent, PCMK_ACTION_STOPPED,
post_down_flags);
// Start group -> start member -> group is started
pcmk__order_starts(member->parent, member, pe_order_implies_first_printed);
pcmk__order_resource_actions(member, PCMK_ACTION_START,
member->parent, PCMK_ACTION_RUNNING,
pe_order_runnable_left
|pe_order_implies_then
|pe_order_implies_then_printed);
if (!member_data->ordered) {
pcmk__order_starts(member->parent, member,
pe_order_implies_then
|pe_order_runnable_left
|pe_order_implies_first_printed);
if (member_data->promotable) {
pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE,
member, PCMK_ACTION_PROMOTE,
pe_order_implies_then
|pe_order_runnable_left
|pe_order_implies_first_printed);
}
} else if (member_data->previous_member == NULL) {
pcmk__order_starts(member->parent, member, pe_order_none);
if (member_data->promotable) {
pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE,
member, PCMK_ACTION_PROMOTE,
pe_order_none);
}
} else {
// Order this member relative to the previous one
pcmk__order_starts(member_data->previous_member, member,
pe_order_implies_then|pe_order_runnable_left);
pcmk__order_stops(member, member_data->previous_member,
pe_order_optional|pe_order_restart);
/* In unusual circumstances (such as adding a new member to the middle
* of a group with unmanaged later members), this member may be active
* while the previous (new) member is inactive. In this situation, the
* usual restart orderings will be irrelevant, so we need to order this
* member's stop before the previous member's start.
*/
if ((member->running_on != NULL)
&& (member_data->previous_member->running_on == NULL)) {
pcmk__order_resource_actions(member, PCMK_ACTION_STOP,
member_data->previous_member,
PCMK_ACTION_START,
pe_order_implies_first
|pe_order_runnable_left);
}
if (member_data->promotable) {
pcmk__order_resource_actions(member_data->previous_member,
PCMK_ACTION_PROMOTE, member,
PCMK_ACTION_PROMOTE,
pe_order_implies_then
|pe_order_runnable_left);
pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE,
member_data->previous_member,
PCMK_ACTION_DEMOTE,
pe_order_optional);
}
}
// Make sure partially active groups shut down in sequence
if (member->running_on != NULL) {
if (member_data->ordered && (member_data->previous_member != NULL)
&& (member_data->previous_member->running_on == NULL)
&& (member_data->last_active != NULL)
&& (member_data->last_active->running_on != NULL)) {
pcmk__order_stops(member, member_data->last_active,
pe_order_optional);
}
member_data->last_active = member;
}
member_data->previous_member = member;
}
/*!
* \internal
* \brief Create implicit constraints needed for a group resource
*
* \param[in,out] rsc Group resource to create implicit constraints for
*/
void
pcmk__group_internal_constraints(pe_resource_t *rsc)
{
struct member_data member_data = { false, };
const pe_resource_t *top = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group));
/* Order group pseudo-actions relative to each other for restarting:
* stop group -> group is stopped -> start group -> group is started
*/
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_STOPPED,
pe_order_runnable_left);
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED,
rsc, PCMK_ACTION_START,
pe_order_optional);
pcmk__order_resource_actions(rsc, PCMK_ACTION_START,
rsc, PCMK_ACTION_RUNNING,
pe_order_runnable_left);
top = pe__const_top_resource(rsc, false);
member_data.ordered = pe__group_flag_is_set(rsc, pe__group_ordered);
member_data.colocated = pe__group_flag_is_set(rsc, pe__group_colocated);
member_data.promotable = pcmk_is_set(top->flags, pcmk_rsc_promotable);
g_list_foreach(rsc->children, member_internal_constraints, &member_data);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint for a group with some other resource, apply the
* 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 group resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
*/
static void
colocate_group_with(pe_resource_t *dependent, const pe_resource_t *primary,
const pcmk__colocation_t *colocation)
{
pe_resource_t *member = NULL;
if (dependent->children == NULL) {
return;
}
pe_rsc_trace(primary, "Processing %s (group %s with %s) for dependent",
colocation->id, dependent->id, primary->id);
if (pe__group_flag_is_set(dependent, pe__group_colocated)) {
// Colocate first member (internal colocations will handle the rest)
member = (pe_resource_t *) dependent->children->data;
member->cmds->apply_coloc_score(member, primary, colocation, true);
return;
}
if (colocation->score >= INFINITY) {
pcmk__config_err("%s: Cannot perform mandatory colocation between "
"non-colocated group and %s",
dependent->id, primary->id);
return;
}
// Colocate each member individually
for (GList *iter = dependent->children; iter != NULL; iter = iter->next) {
member = (pe_resource_t *) iter->data;
member->cmds->apply_coloc_score(member, primary, colocation, true);
}
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint for some other resource with a group, apply the
* 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 group resource in colocation
* \param[in] colocation Colocation constraint to apply
*/
static void
colocate_with_group(pe_resource_t *dependent, const pe_resource_t *primary,
const pcmk__colocation_t *colocation)
{
const pe_resource_t *member = NULL;
pe_rsc_trace(primary,
"Processing colocation %s (%s with group %s) for primary",
colocation->id, dependent->id, primary->id);
if (pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) {
return;
}
if (pe__group_flag_is_set(primary, pe__group_colocated)) {
if (colocation->score >= INFINITY) {
/* For mandatory colocations, the entire group must be assignable
* (and in the specified role if any), so apply the colocation based
* on the last member.
*/
member = pe__last_group_member(primary);
} else if (primary->children != NULL) {
/* For optional colocations, whether the group is partially or fully
* up doesn't matter, so apply the colocation based on the first
* member.
*/
member = (pe_resource_t *) primary->children->data;
}
if (member == NULL) {
return; // Nothing to colocate with
}
member->cmds->apply_coloc_score(dependent, member, colocation, false);
return;
}
if (colocation->score >= INFINITY) {
pcmk__config_err("%s: Cannot perform mandatory colocation with"
" non-colocated group %s",
dependent->id, primary->id);
return;
}
// Colocate dependent with each member individually
for (const GList *iter = primary->children; iter != NULL;
iter = iter->next) {
member = iter->data;
member->cmds->apply_coloc_score(dependent, member, colocation, false);
}
}
/*!
* \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__group_apply_coloc_score(pe_resource_t *dependent,
const pe_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
CRM_ASSERT((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
colocate_group_with(dependent, primary, colocation);
} else {
// Method should only be called for primitive dependents
CRM_ASSERT(dependent->variant == pcmk_rsc_variant_primitive);
colocate_with_group(dependent, primary, colocation);
}
}
/*!
* \internal
* \brief Return action flags for a given group resource action
*
* \param[in,out] action Group action to get flags for
* \param[in] node If not NULL, limit effects to this node
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__group_action_flags(pe_action_t *action, const pe_node_t *node)
{
// Default flags for a group action
- uint32_t flags = pe_action_optional|pe_action_runnable|pe_action_pseudo;
+ uint32_t flags = pe_action_optional|pe_action_runnable|pcmk_action_pseudo;
CRM_ASSERT(action != NULL);
// Update flags considering each member's own flags for same action
for (GList *iter = action->rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
// Check whether member has the same action
enum action_tasks task = get_complex_task(member, action->task);
const char *task_s = task2text(task);
pe_action_t *member_action = find_first_action(member->actions, NULL,
task_s, node);
if (member_action != NULL) {
uint32_t member_flags = member->cmds->action_flags(member_action,
node);
// Group action is mandatory if any member action is
if (pcmk_is_set(flags, pe_action_optional)
&& !pcmk_is_set(member_flags, pe_action_optional)) {
pe_rsc_trace(action->rsc, "%s is mandatory because %s is",
action->uuid, member_action->uuid);
pe__clear_raw_action_flags(flags, "group action",
pe_action_optional);
pe__clear_action_flags(action, pe_action_optional);
}
// Group action is unrunnable if any member action is
if (!pcmk__str_eq(task_s, action->task, pcmk__str_none)
&& pcmk_is_set(flags, pe_action_runnable)
&& !pcmk_is_set(member_flags, pe_action_runnable)) {
pe_rsc_trace(action->rsc, "%s is unrunnable because %s is",
action->uuid, member_action->uuid);
pe__clear_raw_action_flags(flags, "group action",
pe_action_runnable);
pe__clear_action_flags(action, pe_action_runnable);
}
/* Group (pseudo-)actions other than stop or demote are unrunnable
* unless every member will do it.
*/
} else if ((task != pcmk_action_stop) && (task != pcmk_action_demote)) {
pe_rsc_trace(action->rsc,
"%s is not runnable because %s will not %s",
action->uuid, member->id, task_s);
pe__clear_raw_action_flags(flags, "group action",
pe_action_runnable);
}
}
return flags;
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two actions, update the actions' flags
* (and runnable_before members if appropriate) as appropriate for the ordering.
* Effects may cascade to other orderings involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__group_update_ordered_actions(pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
uint32_t changed = pcmk__updated_none;
// Group method can be called only on behalf of "then" action
CRM_ASSERT((first != NULL) && (then != NULL) && (then->rsc != NULL)
&& (data_set != NULL));
// Update the actions for the group itself
changed |= pcmk__update_ordered_actions(first, then, node, flags, filter,
type, data_set);
// Update the actions for each group member
for (GList *iter = then->rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
pe_action_t *member_action = find_first_action(member->actions, NULL,
then->task, node);
if (member_action != NULL) {
changed |= member->cmds->update_ordered_actions(first,
member_action, node,
flags, filter, type,
data_set);
}
}
return changed;
}
/*!
* \internal
* \brief Apply a location constraint to a group's allowed node scores
*
* \param[in,out] rsc Group resource to apply constraint to
* \param[in,out] location Location constraint to apply
*/
void
pcmk__group_apply_location(pe_resource_t *rsc, pe__location_t *location)
{
GList *node_list_orig = NULL;
GList *node_list_copy = NULL;
bool reset_scores = true;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)
&& (location != NULL));
node_list_orig = location->node_list_rh;
node_list_copy = pcmk__copy_node_list(node_list_orig, true);
reset_scores = pe__group_flag_is_set(rsc, pe__group_colocated);
// Apply the constraint for the group itself (updates node scores)
pcmk__apply_location(rsc, location);
// Apply the constraint for each member
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
member->cmds->apply_location(member, location);
if (reset_scores) {
/* The first member of colocated groups needs to use the original
* node scores, but subsequent members should work on a copy, since
* the first member's scores already incorporate theirs.
*/
reset_scores = false;
location->node_list_rh = node_list_copy;
}
}
location->node_list_rh = node_list_orig;
g_list_free_full(node_list_copy, free);
}
// Group implementation of resource_alloc_functions_t:colocated_resources()
GList *
pcmk__group_colocated_resources(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc,
GList *colocated_rscs)
{
const pe_resource_t *member = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group));
if (orig_rsc == NULL) {
orig_rsc = rsc;
}
if (pe__group_flag_is_set(rsc, pe__group_colocated)
|| pe_rsc_is_clone(rsc->parent)) {
/* This group has colocated members and/or is cloned -- either way,
* add every child's colocated resources to the list. The first and last
* members will include the group's own colocations.
*/
colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc);
for (const GList *iter = rsc->children;
iter != NULL; iter = iter->next) {
member = (const pe_resource_t *) iter->data;
colocated_rscs = member->cmds->colocated_resources(member, orig_rsc,
colocated_rscs);
}
} else if (rsc->children != NULL) {
/* This group's members are not colocated, and the group is not cloned,
* so just add the group's own colocations to the list.
*/
colocated_rscs = pcmk__colocated_resources(rsc, orig_rsc,
colocated_rscs);
}
return colocated_rscs;
}
// Group implementation of resource_alloc_functions_t:with_this_colocations()
void
pcmk__with_group_colocations(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)
&& (orig_rsc != NULL) && (list != NULL));
// Ignore empty groups
if (rsc->children == NULL) {
return;
}
/* "With this" colocations are needed only for the group itself and for its
* last member. (Previous members will chain via the group internal
* colocations.)
*/
if ((orig_rsc != rsc) && (orig_rsc != pe__last_group_member(rsc))) {
return;
}
pe_rsc_trace(rsc, "Adding 'with %s' colocations to list for %s",
rsc->id, orig_rsc->id);
// Add the group's own colocations
pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc);
// If cloned, add any relevant colocations with the clone
if (rsc->parent != NULL) {
rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc,
list);
}
if (!pe__group_flag_is_set(rsc, pe__group_colocated)) {
// @COMPAT Non-colocated groups are deprecated
return;
}
// Add explicit colocations with the group's (other) children
for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) {
const pe_resource_t *member = iter->data;
if (member != orig_rsc) {
member->cmds->with_this_colocations(member, orig_rsc, list);
}
}
}
// Group implementation of resource_alloc_functions_t:this_with_colocations()
void
pcmk__group_with_colocations(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList **list)
{
const pe_resource_t *member = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)
&& (orig_rsc != NULL) && (list != NULL));
// Ignore empty groups
if (rsc->children == NULL) {
return;
}
/* "This with" colocations are normally needed only for the group itself and
* for its first member.
*/
if ((rsc == orig_rsc)
|| (orig_rsc == (const pe_resource_t *) rsc->children->data)) {
pe_rsc_trace(rsc, "Adding '%s with' colocations to list for %s",
rsc->id, orig_rsc->id);
// Add the group's own colocations
pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc);
// If cloned, add any relevant colocations involving the clone
if (rsc->parent != NULL) {
rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc,
list);
}
if (!pe__group_flag_is_set(rsc, pe__group_colocated)) {
// @COMPAT Non-colocated groups are deprecated
return;
}
// Add explicit colocations involving the group's (other) children
for (const GList *iter = rsc->children;
iter != NULL; iter = iter->next) {
member = iter->data;
if (member != orig_rsc) {
member->cmds->this_with_colocations(member, orig_rsc, list);
}
}
return;
}
/* Later group members honor the group's colocations indirectly, due to the
* internal group colocations that chain everything from the first member.
* However, if an earlier group member is unmanaged, this chaining will not
* happen, so the group's mandatory colocations must be explicitly added.
*/
for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) {
member = iter->data;
if (orig_rsc == member) {
break; // We've seen all earlier members, and none are unmanaged
}
if (!pcmk_is_set(member->flags, pcmk_rsc_managed)) {
crm_trace("Adding mandatory '%s with' colocations to list for "
"member %s because earlier member %s is unmanaged",
rsc->id, orig_rsc->id, member->id);
for (const GList *cons_iter = rsc->rsc_cons; cons_iter != NULL;
cons_iter = cons_iter->next) {
const pcmk__colocation_t *colocation = NULL;
colocation = (const pcmk__colocation_t *) cons_iter->data;
if (colocation->score == INFINITY) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
// @TODO Add mandatory (or all?) clone constraints if cloned
break;
}
}
}
/*!
* \internal
* \brief Update nodes with scores of colocated resources' nodes
*
* Given a table of nodes and a resource, update the nodes' scores with the
* scores of the best nodes matching the attribute used for each of the
* resource's relevant colocations.
*
* \param[in,out] source_rsc Group resource whose node scores to add
* \param[in] target_rsc Resource on whose behalf to update \p *nodes
* \param[in] log_id Resource ID for logs (if \c NULL, use
* \p source_rsc ID)
* \param[in,out] nodes Nodes to update (set initial contents to \c NULL
* to copy allowed nodes from \p source_rsc)
* \param[in] colocation Original colocation constraint (used to get
* configured primary resource's stickiness, and
* to get colocation node attribute; if \c NULL,
* <tt>source_rsc</tt>'s own matching node scores will
* not be added, and \p *nodes must be \c NULL as
* well)
* \param[in] factor Incorporate scores multiplied by this factor
* \param[in] flags Bitmask of enum pcmk__coloc_select values
*
* \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, and
* the \c pcmk__coloc_select_this_with flag are used together (and only by
* \c cmp_resources()).
* \note The caller remains responsible for freeing \p *nodes.
* \note This is the group implementation of
* \c resource_alloc_functions_t:add_colocated_node_scores().
*/
void
pcmk__group_add_colocated_node_scores(pe_resource_t *source_rsc,
const pe_resource_t *target_rsc,
const char *log_id, GHashTable **nodes,
const pcmk__colocation_t *colocation,
float factor, uint32_t flags)
{
pe_resource_t *member = NULL;
CRM_ASSERT((source_rsc != NULL)
&& (source_rsc->variant == pcmk_rsc_variant_group)
&& (nodes != NULL)
&& ((colocation != NULL)
|| ((target_rsc == NULL) && (*nodes == NULL))));
if (log_id == NULL) {
log_id = source_rsc->id;
}
// Avoid infinite recursion
if (pcmk_is_set(source_rsc->flags, pcmk_rsc_updating_nodes)) {
pe_rsc_info(source_rsc, "%s: Breaking dependency loop at %s",
log_id, source_rsc->id);
return;
}
pe__set_resource_flags(source_rsc, pcmk_rsc_updating_nodes);
// Ignore empty groups (only possible with schema validation disabled)
if (source_rsc->children == NULL) {
return;
}
/* Refer the operation to the first or last member as appropriate.
*
* cmp_resources() is the only caller that passes a NULL nodes table,
* and is also the only caller using pcmk__coloc_select_this_with.
* For "this with" colocations, the last member will recursively incorporate
* all the other members' "this with" colocations via the internal group
* colocations (and via the first member, the group's own colocations).
*
* For "with this" colocations, the first member works similarly.
*/
if (*nodes == NULL) {
member = pe__last_group_member(source_rsc);
} else {
member = source_rsc->children->data;
}
pe_rsc_trace(source_rsc, "%s: Merging scores from group %s using member %s "
"(at %.6f)", log_id, source_rsc->id, member->id, factor);
member->cmds->add_colocated_node_scores(member, target_rsc, log_id, nodes,
colocation, factor, flags);
pe__clear_resource_flags(source_rsc, pcmk_rsc_updating_nodes);
}
// Group implementation of resource_alloc_functions_t:add_utilization()
void
pcmk__group_add_utilization(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList *all_rscs,
GHashTable *utilization)
{
pe_resource_t *member = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)
&& (orig_rsc != NULL) && (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return;
}
pe_rsc_trace(orig_rsc, "%s: Adding group %s as colocated utilization",
orig_rsc->id, rsc->id);
if (pe__group_flag_is_set(rsc, pe__group_colocated)
|| pe_rsc_is_clone(rsc->parent)) {
// Every group member will be on same node, so sum all members
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
member = (pe_resource_t *) iter->data;
if (pcmk_is_set(member->flags, pcmk_rsc_unassigned)
&& (g_list_find(all_rscs, member) == NULL)) {
member->cmds->add_utilization(member, orig_rsc, all_rscs,
utilization);
}
}
} else if (rsc->children != NULL) {
// Just add first member's utilization
member = (pe_resource_t *) rsc->children->data;
if ((member != NULL)
&& pcmk_is_set(member->flags, pcmk_rsc_unassigned)
&& (g_list_find(all_rscs, member) == NULL)) {
member->cmds->add_utilization(member, orig_rsc, all_rscs,
utilization);
}
}
}
// Group implementation of resource_alloc_functions_t:shutdown_lock()
void
pcmk__group_shutdown_lock(pe_resource_t *rsc)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group));
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *member = (pe_resource_t *) iter->data;
member->cmds->shutdown_lock(member);
}
}
diff --git a/lib/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c
index fedb63320d..2cf7272b36 100644
--- a/lib/pacemaker/pcmk_sched_instances.c
+++ b/lib/pacemaker/pcmk_sched_instances.c
@@ -1,1670 +1,1670 @@
/*
* 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.
*/
/* This file is intended for code usable with both clone instances and bundle
* replica containers.
*/
#include <crm_internal.h>
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Check whether a node is allowed to run an instance
*
* \param[in] instance Clone instance or bundle container to check
* \param[in] node Node to check
* \param[in] max_per_node Maximum number of instances allowed to run on a node
*
* \return true if \p node is allowed to run \p instance, otherwise false
*/
static bool
can_run_instance(const pe_resource_t *instance, const pe_node_t *node,
int max_per_node)
{
pe_node_t *allowed_node = NULL;
if (pcmk_is_set(instance->flags, pcmk_rsc_removed)) {
pe_rsc_trace(instance, "%s cannot run on %s: orphaned",
instance->id, pe__node_name(node));
return false;
}
if (!pcmk__node_available(node, false, false)) {
pe_rsc_trace(instance,
"%s cannot run on %s: node cannot run resources",
instance->id, pe__node_name(node));
return false;
}
allowed_node = pcmk__top_allowed_node(instance, node);
if (allowed_node == NULL) {
crm_warn("%s cannot run on %s: node not allowed",
instance->id, pe__node_name(node));
return false;
}
if (allowed_node->weight < 0) {
pe_rsc_trace(instance, "%s cannot run on %s: parent score is %s there",
instance->id, pe__node_name(node),
pcmk_readable_score(allowed_node->weight));
return false;
}
if (allowed_node->count >= max_per_node) {
pe_rsc_trace(instance,
"%s cannot run on %s: node already has %d instance%s",
instance->id, pe__node_name(node), max_per_node,
pcmk__plural_s(max_per_node));
return false;
}
pe_rsc_trace(instance, "%s can run on %s (%d already running)",
instance->id, pe__node_name(node), allowed_node->count);
return true;
}
/*!
* \internal
* \brief Ban a clone instance or bundle replica from unavailable allowed nodes
*
* \param[in,out] instance Clone instance or bundle replica to ban
* \param[in] max_per_node Maximum instances allowed to run on a node
*/
static void
ban_unavailable_allowed_nodes(pe_resource_t *instance, int max_per_node)
{
if (instance->allowed_nodes != NULL) {
GHashTableIter iter;
pe_node_t *node = NULL;
g_hash_table_iter_init(&iter, instance->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (!can_run_instance(instance, node, max_per_node)) {
pe_rsc_trace(instance, "Banning %s from unavailable node %s",
instance->id, pe__node_name(node));
node->weight = -INFINITY;
for (GList *child_iter = instance->children;
child_iter != NULL; child_iter = child_iter->next) {
pe_resource_t *child = (pe_resource_t *) child_iter->data;
pe_node_t *child_node = NULL;
child_node = g_hash_table_lookup(child->allowed_nodes,
node->details->id);
if (child_node != NULL) {
pe_rsc_trace(instance,
"Banning %s child %s "
"from unavailable node %s",
instance->id, child->id,
pe__node_name(node));
child_node->weight = -INFINITY;
}
}
}
}
}
}
/*!
* \internal
* \brief Create a hash table with a single node in it
*
* \param[in] node Node to copy into new table
*
* \return Newly created hash table containing a copy of \p node
* \note The caller is responsible for freeing the result with
* g_hash_table_destroy().
*/
static GHashTable *
new_node_table(pe_node_t *node)
{
GHashTable *table = pcmk__strkey_table(NULL, free);
node = pe__copy_node(node);
g_hash_table_insert(table, (gpointer) node->details->id, node);
return table;
}
/*!
* \internal
* \brief Apply a resource's parent's colocation scores to a node table
*
* \param[in] rsc Resource whose colocations should be applied
* \param[in,out] nodes Node table to apply colocations to
*/
static void
apply_parent_colocations(const pe_resource_t *rsc, GHashTable **nodes)
{
GList *colocations = pcmk__this_with_colocations(rsc);
for (const GList *iter = colocations; iter != NULL; iter = iter->next) {
const pcmk__colocation_t *colocation = iter->data;
pe_resource_t *other = colocation->primary;
float factor = colocation->score / (float) INFINITY;
other->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes,
colocation, factor,
pcmk__coloc_select_default);
}
g_list_free(colocations);
colocations = pcmk__with_this_colocations(rsc);
for (const GList *iter = colocations; iter != NULL; iter = iter->next) {
const pcmk__colocation_t *colocation = iter->data;
pe_resource_t *other = colocation->dependent;
float factor = colocation->score / (float) INFINITY;
if (!pcmk__colocation_has_influence(colocation, rsc)) {
continue;
}
other->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes,
colocation, factor,
pcmk__coloc_select_nonnegative);
}
g_list_free(colocations);
}
/*!
* \internal
* \brief Compare clone or bundle instances based on colocation scores
*
* Determine the relative order in which two clone or bundle instances should be
* assigned to nodes, considering the scores of colocation constraints directly
* or indirectly involving them.
*
* \param[in] instance1 First instance to compare
* \param[in] instance2 Second instance to compare
*
* \return A negative number if \p instance1 should be assigned first,
* a positive number if \p instance2 should be assigned first,
* or 0 if assignment order doesn't matter
*/
static int
cmp_instance_by_colocation(const pe_resource_t *instance1,
const pe_resource_t *instance2)
{
int rc = 0;
pe_node_t *node1 = NULL;
pe_node_t *node2 = NULL;
pe_node_t *current_node1 = pe__current_node(instance1);
pe_node_t *current_node2 = pe__current_node(instance2);
GHashTable *colocated_scores1 = NULL;
GHashTable *colocated_scores2 = NULL;
CRM_ASSERT((instance1 != NULL) && (instance1->parent != NULL)
&& (instance2 != NULL) && (instance2->parent != NULL)
&& (current_node1 != NULL) && (current_node2 != NULL));
// Create node tables initialized with each node
colocated_scores1 = new_node_table(current_node1);
colocated_scores2 = new_node_table(current_node2);
// Apply parental colocations
apply_parent_colocations(instance1, &colocated_scores1);
apply_parent_colocations(instance2, &colocated_scores2);
// Find original nodes again, with scores updated for colocations
node1 = g_hash_table_lookup(colocated_scores1, current_node1->details->id);
node2 = g_hash_table_lookup(colocated_scores2, current_node2->details->id);
// Compare nodes by updated scores
if (node1->weight < node2->weight) {
crm_trace("Assign %s (%d on %s) after %s (%d on %s)",
instance1->id, node1->weight, pe__node_name(node1),
instance2->id, node2->weight, pe__node_name(node2));
rc = 1;
} else if (node1->weight > node2->weight) {
crm_trace("Assign %s (%d on %s) before %s (%d on %s)",
instance1->id, node1->weight, pe__node_name(node1),
instance2->id, node2->weight, pe__node_name(node2));
rc = -1;
}
g_hash_table_destroy(colocated_scores1);
g_hash_table_destroy(colocated_scores2);
return rc;
}
/*!
* \internal
* \brief Check whether a resource or any of its children are failed
*
* \param[in] rsc Resource to check
*
* \return true if \p rsc or any of its children are failed, otherwise false
*/
static bool
did_fail(const pe_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
return true;
}
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
if (did_fail((const pe_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a node is allowed to run a resource
*
* \param[in] rsc Resource to check
* \param[in,out] node Node to check (will be set NULL if not allowed)
*
* \return true if *node is either NULL or allowed for \p rsc, otherwise false
*/
static bool
node_is_allowed(const pe_resource_t *rsc, pe_node_t **node)
{
if (*node != NULL) {
pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes,
(*node)->details->id);
if ((allowed == NULL) || (allowed->weight < 0)) {
pe_rsc_trace(rsc, "%s: current location (%s) is unavailable",
rsc->id, pe__node_name(*node));
*node = NULL;
return false;
}
}
return true;
}
/*!
* \internal
* \brief Compare two clone or bundle instances' instance numbers
*
* \param[in] a First instance to compare
* \param[in] b Second instance to compare
*
* \return A negative number if \p a's instance number is lower,
* a positive number if \p b's instance number is lower,
* or 0 if their instance numbers are the same
*/
gint
pcmk__cmp_instance_number(gconstpointer a, gconstpointer b)
{
const pe_resource_t *instance1 = (const pe_resource_t *) a;
const pe_resource_t *instance2 = (const pe_resource_t *) b;
char *div1 = NULL;
char *div2 = NULL;
CRM_ASSERT((instance1 != NULL) && (instance2 != NULL));
// Clone numbers are after a colon, bundle numbers after a dash
div1 = strrchr(instance1->id, ':');
if (div1 == NULL) {
div1 = strrchr(instance1->id, '-');
}
div2 = strrchr(instance2->id, ':');
if (div2 == NULL) {
div2 = strrchr(instance2->id, '-');
}
CRM_ASSERT((div1 != NULL) && (div2 != NULL));
return (gint) (strtol(div1 + 1, NULL, 10) - strtol(div2 + 1, NULL, 10));
}
/*!
* \internal
* \brief Compare clone or bundle instances according to assignment order
*
* Compare two clone or bundle instances according to the order they should be
* assigned to nodes, preferring (in order):
*
* - Active instance that is less multiply active
* - Instance that is not active on a disallowed node
* - Instance with higher configured priority
* - Active instance whose current node can run resources
* - Active instance whose parent is allowed on current node
* - Active instance whose current node has fewer other instances
* - Active instance
* - Instance that isn't failed
* - Instance whose colocations result in higher score on current node
* - Instance with lower ID in lexicographic order
*
* \param[in] a First instance to compare
* \param[in] b Second instance to compare
*
* \return A negative number if \p a should be assigned first,
* a positive number if \p b should be assigned first,
* or 0 if assignment order doesn't matter
*/
gint
pcmk__cmp_instance(gconstpointer a, gconstpointer b)
{
int rc = 0;
pe_node_t *node1 = NULL;
pe_node_t *node2 = NULL;
unsigned int nnodes1 = 0;
unsigned int nnodes2 = 0;
bool can1 = true;
bool can2 = true;
const pe_resource_t *instance1 = (const pe_resource_t *) a;
const pe_resource_t *instance2 = (const pe_resource_t *) b;
CRM_ASSERT((instance1 != NULL) && (instance2 != NULL));
node1 = instance1->fns->active_node(instance1, &nnodes1, NULL);
node2 = instance2->fns->active_node(instance2, &nnodes2, NULL);
/* If both instances are running and at least one is multiply
* active, prefer instance that's running on fewer nodes.
*/
if ((nnodes1 > 0) && (nnodes2 > 0)) {
if (nnodes1 < nnodes2) {
crm_trace("Assign %s (active on %d) before %s (active on %d): "
"less multiply active",
instance1->id, nnodes1, instance2->id, nnodes2);
return -1;
} else if (nnodes1 > nnodes2) {
crm_trace("Assign %s (active on %d) after %s (active on %d): "
"more multiply active",
instance1->id, nnodes1, instance2->id, nnodes2);
return 1;
}
}
/* An instance that is either inactive or active on an allowed node is
* preferred over an instance that is active on a no-longer-allowed node.
*/
can1 = node_is_allowed(instance1, &node1);
can2 = node_is_allowed(instance2, &node2);
if (can1 && !can2) {
crm_trace("Assign %s before %s: not active on a disallowed node",
instance1->id, instance2->id);
return -1;
} else if (!can1 && can2) {
crm_trace("Assign %s after %s: active on a disallowed node",
instance1->id, instance2->id);
return 1;
}
// Prefer instance with higher configured priority
if (instance1->priority > instance2->priority) {
crm_trace("Assign %s before %s: priority (%d > %d)",
instance1->id, instance2->id,
instance1->priority, instance2->priority);
return -1;
} else if (instance1->priority < instance2->priority) {
crm_trace("Assign %s after %s: priority (%d < %d)",
instance1->id, instance2->id,
instance1->priority, instance2->priority);
return 1;
}
// Prefer active instance
if ((node1 == NULL) && (node2 == NULL)) {
crm_trace("No assignment preference for %s vs. %s: inactive",
instance1->id, instance2->id);
return 0;
} else if (node1 == NULL) {
crm_trace("Assign %s after %s: active", instance1->id, instance2->id);
return 1;
} else if (node2 == NULL) {
crm_trace("Assign %s before %s: active", instance1->id, instance2->id);
return -1;
}
// Prefer instance whose current node can run resources
can1 = pcmk__node_available(node1, false, false);
can2 = pcmk__node_available(node2, false, false);
if (can1 && !can2) {
crm_trace("Assign %s before %s: current node can run resources",
instance1->id, instance2->id);
return -1;
} else if (!can1 && can2) {
crm_trace("Assign %s after %s: current node can't run resources",
instance1->id, instance2->id);
return 1;
}
// Prefer instance whose parent is allowed to run on instance's current node
node1 = pcmk__top_allowed_node(instance1, node1);
node2 = pcmk__top_allowed_node(instance2, node2);
if ((node1 == NULL) && (node2 == NULL)) {
crm_trace("No assignment preference for %s vs. %s: "
"parent not allowed on either instance's current node",
instance1->id, instance2->id);
return 0;
} else if (node1 == NULL) {
crm_trace("Assign %s after %s: parent not allowed on current node",
instance1->id, instance2->id);
return 1;
} else if (node2 == NULL) {
crm_trace("Assign %s before %s: parent allowed on current node",
instance1->id, instance2->id);
return -1;
}
// Prefer instance whose current node is running fewer other instances
if (node1->count < node2->count) {
crm_trace("Assign %s before %s: fewer active instances on current node",
instance1->id, instance2->id);
return -1;
} else if (node1->count > node2->count) {
crm_trace("Assign %s after %s: more active instances on current node",
instance1->id, instance2->id);
return 1;
}
// Prefer instance that isn't failed
can1 = did_fail(instance1);
can2 = did_fail(instance2);
if (!can1 && can2) {
crm_trace("Assign %s before %s: not failed",
instance1->id, instance2->id);
return -1;
} else if (can1 && !can2) {
crm_trace("Assign %s after %s: failed",
instance1->id, instance2->id);
return 1;
}
// Prefer instance with higher cumulative colocation score on current node
rc = cmp_instance_by_colocation(instance1, instance2);
if (rc != 0) {
return rc;
}
// Prefer instance with lower instance number
rc = pcmk__cmp_instance_number(instance1, instance2);
if (rc < 0) {
crm_trace("Assign %s before %s: instance number",
instance1->id, instance2->id);
} else if (rc > 0) {
crm_trace("Assign %s after %s: instance number",
instance1->id, instance2->id);
} else {
crm_trace("No assignment preference for %s vs. %s",
instance1->id, instance2->id);
}
return rc;
}
/*!
* \internal
* \brief Increment the parent's instance count after assigning an instance
*
* An instance's parent tracks how many instances have been assigned to each
* node via its pe_node_t:count member. After assigning an instance to a node,
* find the corresponding node in the parent's allowed table and increment it.
*
* \param[in,out] instance Instance whose parent to update
* \param[in] assigned_to Node to which the instance was assigned
*/
static void
increment_parent_count(pe_resource_t *instance, const pe_node_t *assigned_to)
{
pe_node_t *allowed = NULL;
if (assigned_to == NULL) {
return;
}
allowed = pcmk__top_allowed_node(instance, assigned_to);
if (allowed == NULL) {
/* The instance is allowed on the node, but its parent isn't. This
* shouldn't be possible if the resource is managed, and we won't be
* able to limit the number of instances assigned to the node.
*/
CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pcmk_rsc_managed));
} else {
allowed->count++;
}
}
/*!
* \internal
* \brief Assign an instance to a node
*
* \param[in,out] instance Clone instance or bundle replica container
* \param[in] prefer If not NULL, attempt early assignment to this
* node, if still the best choice; otherwise,
* perform final assignment
* \param[in] max_per_node Assign at most this many instances to one node
*
* \return Node to which \p instance is assigned
*/
static const pe_node_t *
assign_instance(pe_resource_t *instance, const pe_node_t *prefer,
int max_per_node)
{
pe_node_t *chosen = NULL;
pe_rsc_trace(instance, "Assigning %s (preferring %s)", instance->id,
((prefer == NULL)? "no node" : prefer->details->uname));
if (pcmk_is_set(instance->flags, pcmk_rsc_assigning)) {
pe_rsc_debug(instance,
"Assignment loop detected involving %s colocations",
instance->id);
return NULL;
}
ban_unavailable_allowed_nodes(instance, max_per_node);
// Failed early assignments are reversible (stop_if_fail=false)
chosen = instance->cmds->assign(instance, prefer, (prefer == NULL));
increment_parent_count(instance, chosen);
return chosen;
}
/*!
* \internal
* \brief Try to assign an instance to its current node early
*
* \param[in] rsc Clone or bundle being assigned (for logs only)
* \param[in] instance Clone instance or bundle replica container
* \param[in] current Instance's current node
* \param[in] max_per_node Maximum number of instances per node
* \param[in] available Number of instances still available for assignment
*
* \return \c true if \p instance was successfully assigned to its current node,
* or \c false otherwise
*/
static bool
assign_instance_early(const pe_resource_t *rsc, pe_resource_t *instance,
const pe_node_t *current, int max_per_node, int available)
{
const pe_node_t *chosen = NULL;
int reserved = 0;
pe_resource_t *parent = instance->parent;
GHashTable *allowed_orig = NULL;
GHashTable *allowed_orig_parent = parent->allowed_nodes;
const pe_node_t *allowed_node = g_hash_table_lookup(instance->allowed_nodes,
current->details->id);
pe_rsc_trace(instance, "Trying to assign %s to its current node %s",
instance->id, pe__node_name(current));
if (!pcmk__node_available(allowed_node, true, false)) {
pe_rsc_info(instance,
"Not assigning %s to current node %s: unavailable",
instance->id, pe__node_name(current));
return false;
}
/* On each iteration, if instance gets assigned to a node other than its
* current one, we reserve one instance for the chosen node, unassign
* instance, restore instance's original node tables, and try again. This
* way, instances are proportionally assigned to nodes based on preferences,
* but shuffling of specific instances is minimized. If a node will be
* assigned instances at all, it preferentially receives instances that are
* currently active there.
*
* parent->allowed_nodes tracks the number of instances assigned to each
* node. If a node already has max_per_node instances assigned,
* ban_unavailable_allowed_nodes() marks it as unavailable.
*
* In the end, we restore the original parent->allowed_nodes to undo the
* changes to counts during tentative assignments. If we successfully
* assigned instance to its current node, we increment that node's counter.
*/
// Back up the allowed node tables of instance and its children recursively
pcmk__copy_node_tables(instance, &allowed_orig);
// Update instances-per-node counts in a scratch table
parent->allowed_nodes = pcmk__copy_node_table(parent->allowed_nodes);
while (reserved < available) {
chosen = assign_instance(instance, current, max_per_node);
if (pe__same_node(chosen, current)) {
// Successfully assigned to current node
break;
}
// Assignment updates scores, so restore to original state
pe_rsc_debug(instance, "Rolling back node scores for %s", instance->id);
pcmk__restore_node_tables(instance, allowed_orig);
if (chosen == NULL) {
// Assignment failed, so give up
pe_rsc_info(instance,
"Not assigning %s to current node %s: unavailable",
instance->id, pe__node_name(current));
pe__set_resource_flags(instance, pcmk_rsc_unassigned);
break;
}
// We prefer more strongly to assign an instance to the chosen node
pe_rsc_debug(instance,
"Not assigning %s to current node %s: %s is better",
instance->id, pe__node_name(current),
pe__node_name(chosen));
// Reserve one instance for the chosen node and try again
if (++reserved >= available) {
pe_rsc_info(instance,
"Not assigning %s to current node %s: "
"other assignments are more important",
instance->id, pe__node_name(current));
} else {
pe_rsc_debug(instance,
"Reserved an instance of %s for %s. Retrying "
"assignment of %s to %s",
rsc->id, pe__node_name(chosen), instance->id,
pe__node_name(current));
}
// Clear this assignment (frees chosen); leave instance counts in parent
pcmk__unassign_resource(instance);
chosen = NULL;
}
g_hash_table_destroy(allowed_orig);
// Restore original instances-per-node counts
g_hash_table_destroy(parent->allowed_nodes);
parent->allowed_nodes = allowed_orig_parent;
if (chosen == NULL) {
// Couldn't assign instance to current node
return false;
}
pe_rsc_trace(instance, "Assigned %s to current node %s",
instance->id, pe__node_name(current));
increment_parent_count(instance, chosen);
return true;
}
/*!
* \internal
* \brief Reset the node counts of a resource's allowed nodes to zero
*
* \param[in,out] rsc Resource to reset
*
* \return Number of nodes that are available to run resources
*/
static unsigned int
reset_allowed_node_counts(pe_resource_t *rsc)
{
unsigned int available_nodes = 0;
pe_node_t *node = NULL;
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
node->count = 0;
if (pcmk__node_available(node, false, false)) {
available_nodes++;
}
}
return available_nodes;
}
/*!
* \internal
* \brief Check whether an instance has a preferred node
*
* \param[in] instance Clone instance or bundle replica container
* \param[in] optimal_per_node Optimal number of instances per node
*
* \return Instance's current node if still available, otherwise NULL
*/
static const pe_node_t *
preferred_node(const pe_resource_t *instance, int optimal_per_node)
{
const pe_node_t *node = NULL;
const pe_node_t *parent_node = NULL;
// Check whether instance is active, healthy, and not yet assigned
if ((instance->running_on == NULL)
|| !pcmk_is_set(instance->flags, pcmk_rsc_unassigned)
|| pcmk_is_set(instance->flags, pcmk_rsc_failed)) {
return NULL;
}
// Check whether instance's current node can run resources
node = pe__current_node(instance);
if (!pcmk__node_available(node, true, false)) {
pe_rsc_trace(instance, "Not assigning %s to %s early (unavailable)",
instance->id, pe__node_name(node));
return NULL;
}
// Check whether node already has optimal number of instances assigned
parent_node = pcmk__top_allowed_node(instance, node);
if ((parent_node != NULL) && (parent_node->count >= optimal_per_node)) {
pe_rsc_trace(instance,
"Not assigning %s to %s early "
"(optimal instances already assigned)",
instance->id, pe__node_name(node));
return NULL;
}
return node;
}
/*!
* \internal
* \brief Assign collective instances to nodes
*
* \param[in,out] collective Clone or bundle resource being assigned
* \param[in,out] instances List of clone instances or bundle containers
* \param[in] max_total Maximum instances to assign in total
* \param[in] max_per_node Maximum instances to assign to any one node
*/
void
pcmk__assign_instances(pe_resource_t *collective, GList *instances,
int max_total, int max_per_node)
{
// Reuse node count to track number of assigned instances
unsigned int available_nodes = reset_allowed_node_counts(collective);
int optimal_per_node = 0;
int assigned = 0;
GList *iter = NULL;
pe_resource_t *instance = NULL;
const pe_node_t *current = NULL;
if (available_nodes > 0) {
optimal_per_node = max_total / available_nodes;
}
if (optimal_per_node < 1) {
optimal_per_node = 1;
}
pe_rsc_debug(collective,
"Assigning up to %d %s instance%s to up to %u node%s "
"(at most %d per host, %d optimal)",
max_total, collective->id, pcmk__plural_s(max_total),
available_nodes, pcmk__plural_s(available_nodes),
max_per_node, optimal_per_node);
// Assign as many instances as possible to their current location
for (iter = instances; (iter != NULL) && (assigned < max_total);
iter = iter->next) {
int available = max_total - assigned;
instance = iter->data;
if (!pcmk_is_set(instance->flags, pcmk_rsc_unassigned)) {
continue; // Already assigned
}
current = preferred_node(instance, optimal_per_node);
if ((current != NULL)
&& assign_instance_early(collective, instance, current,
max_per_node, available)) {
assigned++;
}
}
pe_rsc_trace(collective, "Assigned %d of %d instance%s to current node",
assigned, max_total, pcmk__plural_s(max_total));
for (iter = instances; iter != NULL; iter = iter->next) {
instance = (pe_resource_t *) iter->data;
if (!pcmk_is_set(instance->flags, pcmk_rsc_unassigned)) {
continue; // Already assigned
}
if (instance->running_on != NULL) {
current = pe__current_node(instance);
if (pcmk__top_allowed_node(instance, current) == NULL) {
const char *unmanaged = "";
if (!pcmk_is_set(instance->flags, pcmk_rsc_managed)) {
unmanaged = "Unmanaged resource ";
}
crm_notice("%s%s is running on %s which is no longer allowed",
unmanaged, instance->id, pe__node_name(current));
}
}
if (assigned >= max_total) {
pe_rsc_debug(collective,
"Not assigning %s because maximum %d instances "
"already assigned",
instance->id, max_total);
resource_location(instance, NULL, -INFINITY,
"collective_limit_reached", collective->cluster);
} else if (assign_instance(instance, NULL, max_per_node) != NULL) {
assigned++;
}
}
pe_rsc_debug(collective, "Assigned %d of %d possible instance%s of %s",
assigned, max_total, pcmk__plural_s(max_total),
collective->id);
}
enum instance_state {
instance_starting = (1 << 0),
instance_stopping = (1 << 1),
/* This indicates that some instance is restarting. It's not the same as
* instance_starting|instance_stopping, which would indicate that some
* instance is starting, and some instance (not necessarily the same one) is
* stopping.
*/
instance_restarting = (1 << 2),
instance_active = (1 << 3),
instance_all = instance_starting|instance_stopping
|instance_restarting|instance_active,
};
/*!
* \internal
* \brief Check whether an instance is active, starting, and/or stopping
*
* \param[in] instance Clone instance or bundle replica container
* \param[in,out] state Whether any instance is starting, stopping, etc.
*/
static void
check_instance_state(const pe_resource_t *instance, uint32_t *state)
{
const GList *iter = NULL;
uint32_t instance_state = 0; // State of just this instance
// No need to check further if all conditions have already been detected
if (pcmk_all_flags_set(*state, instance_all)) {
return;
}
// If instance is a collective (a cloned group), check its children instead
if (instance->variant > pcmk_rsc_variant_primitive) {
for (iter = instance->children;
(iter != NULL) && !pcmk_all_flags_set(*state, instance_all);
iter = iter->next) {
check_instance_state((const pe_resource_t *) iter->data, state);
}
return;
}
// If we get here, instance is a primitive
if (instance->running_on != NULL) {
instance_state |= instance_active;
}
// Check each of the instance's actions for runnable start or stop
for (iter = instance->actions;
(iter != NULL) && !pcmk_all_flags_set(instance_state,
instance_starting
|instance_stopping);
iter = iter->next) {
const pe_action_t *action = (const pe_action_t *) iter->data;
const bool optional = pcmk_is_set(action->flags, pe_action_optional);
if (pcmk__str_eq(PCMK_ACTION_START, action->task, pcmk__str_none)) {
if (!optional && pcmk_is_set(action->flags, pe_action_runnable)) {
pe_rsc_trace(instance, "Instance is starting due to %s",
action->uuid);
instance_state |= instance_starting;
} else {
pe_rsc_trace(instance, "%s doesn't affect %s state (%s)",
action->uuid, instance->id,
(optional? "optional" : "unrunnable"));
}
} else if (pcmk__str_eq(PCMK_ACTION_STOP, action->task,
pcmk__str_none)) {
/* Only stop actions can be pseudo-actions for primitives. That
* indicates that the node they are on is being fenced, so the stop
* is implied rather than actually executed.
*/
if (!optional
- && pcmk_any_flags_set(action->flags,
- pe_action_pseudo|pe_action_runnable)) {
+ && pcmk_any_flags_set(action->flags, pcmk_action_pseudo
+ |pe_action_runnable)) {
pe_rsc_trace(instance, "Instance is stopping due to %s",
action->uuid);
instance_state |= instance_stopping;
} else {
pe_rsc_trace(instance, "%s doesn't affect %s state (%s)",
action->uuid, instance->id,
(optional? "optional" : "unrunnable"));
}
}
}
if (pcmk_all_flags_set(instance_state,
instance_starting|instance_stopping)) {
instance_state |= instance_restarting;
}
*state |= instance_state;
}
/*!
* \internal
* \brief Create actions for collective resource instances
*
* \param[in,out] collective Clone or bundle resource to create actions for
* \param[in,out] instances List of clone instances or bundle containers
*/
void
pcmk__create_instance_actions(pe_resource_t *collective, GList *instances)
{
uint32_t state = 0;
pe_action_t *stop = NULL;
pe_action_t *stopped = NULL;
pe_action_t *start = NULL;
pe_action_t *started = NULL;
pe_rsc_trace(collective, "Creating collective instance actions for %s",
collective->id);
// Create actions for each instance appropriate to its variant
for (GList *iter = instances; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
instance->cmds->create_actions(instance);
check_instance_state(instance, &state);
}
// Create pseudo-actions for rsc start and started
start = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_START,
!pcmk_is_set(state, instance_starting),
true);
started = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_RUNNING,
!pcmk_is_set(state, instance_starting),
false);
started->priority = INFINITY;
if (pcmk_any_flags_set(state, instance_active|instance_starting)) {
pe__set_action_flags(started, pe_action_runnable);
}
// Create pseudo-actions for rsc stop and stopped
stop = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOP,
!pcmk_is_set(state, instance_stopping),
true);
stopped = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOPPED,
!pcmk_is_set(state, instance_stopping),
true);
stopped->priority = INFINITY;
if (!pcmk_is_set(state, instance_restarting)) {
pe__set_action_flags(stop, pe_action_migrate_runnable);
}
if (collective->variant == pcmk_rsc_variant_clone) {
pe__create_clone_notif_pseudo_ops(collective, start, started, stop,
stopped);
}
}
/*!
* \internal
* \brief Get a list of clone instances or bundle replica containers
*
* \param[in] rsc Clone or bundle resource
*
* \return Clone instances if \p rsc is a clone, or a newly created list of
* \p rsc's replica containers if \p rsc is a bundle
* \note The caller must call free_instance_list() on the result when the list
* is no longer needed.
*/
static inline GList *
get_instance_list(const pe_resource_t *rsc)
{
if (rsc->variant == pcmk_rsc_variant_bundle) {
return pe__bundle_containers(rsc);
} else {
return rsc->children;
}
}
/*!
* \internal
* \brief Free any memory created by get_instance_list()
*
* \param[in] rsc Clone or bundle resource passed to get_instance_list()
* \param[in,out] list Return value of get_instance_list() for \p rsc
*/
static inline void
free_instance_list(const pe_resource_t *rsc, GList *list)
{
if (list != rsc->children) {
g_list_free(list);
}
}
/*!
* \internal
* \brief Check whether an instance is compatible with a role and node
*
* \param[in] instance Clone instance or bundle replica container
* \param[in] node Instance must match this node
* \param[in] role If not pcmk_role_unknown, instance must match this role
* \param[in] current If true, compare instance's original node and role,
* otherwise compare assigned next node and role
*
* \return true if \p instance is compatible with \p node and \p role,
* otherwise false
*/
bool
pcmk__instance_matches(const pe_resource_t *instance, const pe_node_t *node,
enum rsc_role_e role, bool current)
{
pe_node_t *instance_node = NULL;
CRM_CHECK((instance != NULL) && (node != NULL), return false);
if ((role != pcmk_role_unknown)
&& (role != instance->fns->state(instance, current))) {
pe_rsc_trace(instance,
"%s is not a compatible instance (role is not %s)",
instance->id, role2text(role));
return false;
}
if (!is_set_recursive(instance, pcmk_rsc_blocked, true)) {
// We only want instances that haven't failed
instance_node = instance->fns->location(instance, NULL, current);
}
if (instance_node == NULL) {
pe_rsc_trace(instance,
"%s is not a compatible instance (not assigned to a node)",
instance->id);
return false;
}
if (!pe__same_node(instance_node, node)) {
pe_rsc_trace(instance,
"%s is not a compatible instance (assigned to %s not %s)",
instance->id, pe__node_name(instance_node),
pe__node_name(node));
return false;
}
return true;
}
/*!
* \internal
* \brief Find an instance that matches a given resource by node and role
*
* \param[in] match_rsc Resource that instance must match (for logging only)
* \param[in] rsc Clone or bundle resource to check for matching instance
* \param[in] node Instance must match this node
* \param[in] role If not pcmk_role_unknown, instance must match this role
* \param[in] current If true, compare instance's original node and role,
* otherwise compare assigned next node and role
*
* \return \p rsc instance matching \p node and \p role if any, otherwise NULL
*/
static pe_resource_t *
find_compatible_instance_on_node(const pe_resource_t *match_rsc,
const pe_resource_t *rsc,
const pe_node_t *node, enum rsc_role_e role,
bool current)
{
GList *instances = NULL;
instances = get_instance_list(rsc);
for (GList *iter = instances; iter != NULL; iter = iter->next) {
pe_resource_t *instance = (pe_resource_t *) iter->data;
if (pcmk__instance_matches(instance, node, role, current)) {
pe_rsc_trace(match_rsc,
"Found %s %s instance %s compatible with %s on %s",
role == pcmk_role_unknown? "matching" : role2text(role),
rsc->id, instance->id, match_rsc->id,
pe__node_name(node));
free_instance_list(rsc, instances); // Only frees list, not contents
return instance;
}
}
free_instance_list(rsc, instances);
pe_rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s",
((role == pcmk_role_unknown)? "matching" : role2text(role)),
rsc->id, match_rsc->id, pe__node_name(node));
return NULL;
}
/*!
* \internal
* \brief Find a clone instance or bundle container compatible with a resource
*
* \param[in] match_rsc Resource that instance must match
* \param[in] rsc Clone or bundle resource to check for matching instance
* \param[in] role If not pcmk_role_unknown, instance must match this role
* \param[in] current If true, compare instance's original node and role,
* otherwise compare assigned next node and role
*
* \return Compatible (by \p role and \p match_rsc location) instance of \p rsc
* if any, otherwise NULL
*/
pe_resource_t *
pcmk__find_compatible_instance(const pe_resource_t *match_rsc,
const pe_resource_t *rsc, enum rsc_role_e role,
bool current)
{
pe_resource_t *instance = NULL;
GList *nodes = NULL;
const pe_node_t *node = match_rsc->fns->location(match_rsc, NULL, current);
// If match_rsc has a node, check only that node
if (node != NULL) {
return find_compatible_instance_on_node(match_rsc, rsc, node, role,
current);
}
// Otherwise check for an instance matching any of match_rsc's allowed nodes
nodes = pcmk__sort_nodes(g_hash_table_get_values(match_rsc->allowed_nodes),
NULL);
for (GList *iter = nodes; (iter != NULL) && (instance == NULL);
iter = iter->next) {
instance = find_compatible_instance_on_node(match_rsc, rsc,
(pe_node_t *) iter->data,
role, current);
}
if (instance == NULL) {
pe_rsc_debug(rsc, "No %s instance found compatible with %s",
rsc->id, match_rsc->id);
}
g_list_free(nodes);
return instance;
}
/*!
* \internal
* \brief Unassign an instance if mandatory ordering has no interleave match
*
* \param[in] first 'First' action in an ordering
* \param[in] then 'Then' action in an ordering
* \param[in,out] then_instance 'Then' instance that has no interleave match
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in] current If true, "then" action is stopped or demoted
*
* \return true if \p then_instance was unassigned, otherwise false
*/
static bool
unassign_if_mandatory(const pe_action_t *first, const pe_action_t *then,
pe_resource_t *then_instance, uint32_t type, bool current)
{
// Allow "then" instance to go down even without an interleave match
if (current) {
pe_rsc_trace(then->rsc,
"%s has no instance to order before stopping "
"or demoting %s",
first->rsc->id, then_instance->id);
/* If the "first" action must be runnable, but there is no "first"
* instance, the "then" instance must not be allowed to come up.
*/
} else if (pcmk_any_flags_set(type, pe_order_runnable_left
|pe_order_implies_then)) {
pe_rsc_info(then->rsc,
"Inhibiting %s from being active "
"because there is no %s instance to interleave",
then_instance->id, first->rsc->id);
return pcmk__assign_resource(then_instance, NULL, true, true);
}
return false;
}
/*!
* \internal
* \brief Find first matching action for a clone instance or bundle container
*
* \param[in] action Action in an interleaved ordering
* \param[in] instance Clone instance or bundle container being interleaved
* \param[in] action_name Action to look for
* \param[in] node If not NULL, require action to be on this node
* \param[in] for_first If true, \p instance is the 'first' resource in the
* ordering, otherwise it is the 'then' resource
*
* \return First action for \p instance (or in some cases if \p instance is a
* bundle container, its containerized resource) that matches
* \p action_name and \p node if any, otherwise NULL
*/
static pe_action_t *
find_instance_action(const pe_action_t *action, const pe_resource_t *instance,
const char *action_name, const pe_node_t *node,
bool for_first)
{
const pe_resource_t *rsc = NULL;
pe_action_t *matching_action = NULL;
/* If instance is a bundle container, sometimes we should interleave the
* action for the container itself, and sometimes for the containerized
* resource.
*
* For example, given "start bundle A then bundle B", B likely requires the
* service inside A's container to be active, rather than just the
* container, so we should interleave the action for A's containerized
* resource. On the other hand, it's possible B's container itself requires
* something from A, so we should interleave the action for B's container.
*
* Essentially, for 'first', we should use the containerized resource for
* everything except stop, and for 'then', we should use the container for
* everything except promote and demote (which can only be performed on the
* containerized resource).
*/
if ((for_first && !pcmk__str_any_of(action->task, PCMK_ACTION_STOP,
PCMK_ACTION_STOPPED, NULL))
|| (!for_first && pcmk__str_any_of(action->task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_PROMOTED,
PCMK_ACTION_DEMOTE,
PCMK_ACTION_DEMOTED, NULL))) {
rsc = pe__get_rsc_in_container(instance);
}
if (rsc == NULL) {
rsc = instance; // No containerized resource, use instance itself
} else {
node = NULL; // Containerized actions are on bundle-created guest
}
matching_action = find_first_action(rsc->actions, NULL, action_name, node);
if (matching_action != NULL) {
return matching_action;
}
if (pcmk_is_set(instance->flags, pcmk_rsc_removed)
|| pcmk__str_any_of(action_name, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE,
NULL)) {
crm_trace("No %s action found for %s%s",
action_name,
pcmk_is_set(instance->flags, pcmk_rsc_removed)? "orphan " : "",
instance->id);
} else {
crm_err("No %s action found for %s to interleave (bug?)",
action_name, instance->id);
}
return NULL;
}
/*!
* \internal
* \brief Get the original action name of a bundle or clone action
*
* Given an action for a bundle or clone, get the original action name,
* mapping notify to the action being notified, and if the instances are
* primitives, mapping completion actions to the action that was completed
* (for example, stopped to stop).
*
* \param[in] action Clone or bundle action to check
*
* \return Original action name for \p action
*/
static const char *
orig_action_name(const pe_action_t *action)
{
const pe_resource_t *instance = action->rsc->children->data; // Any instance
char *action_type = NULL;
const char *action_name = action->task;
enum action_tasks orig_task = pcmk_action_unspecified;
if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY,
PCMK_ACTION_NOTIFIED, NULL)) {
// action->uuid is RSC_(confirmed-){pre,post}_notify_ACTION_INTERVAL
CRM_CHECK(parse_op_key(action->uuid, NULL, &action_type, NULL),
return task2text(pcmk_action_unspecified));
action_name = strstr(action_type, "_notify_");
CRM_CHECK(action_name != NULL,
return task2text(pcmk_action_unspecified));
action_name += strlen("_notify_");
}
orig_task = get_complex_task(instance, action_name);
free(action_type);
return task2text(orig_task);
}
/*!
* \internal
* \brief Update two interleaved actions according to an ordering between them
*
* Given information about an ordering of two interleaved actions, update the
* actions' flags (and runnable_before members if appropriate) as appropriate
* for the ordering. Effects may cascade to other orderings involving the
* actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static uint32_t
update_interleaved_actions(pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t filter,
uint32_t type)
{
GList *instances = NULL;
uint32_t changed = pcmk__updated_none;
const char *orig_first_task = orig_action_name(first);
// Stops and demotes must be interleaved with instance on current node
bool current = pcmk__ends_with(first->uuid, "_" PCMK_ACTION_STOPPED "_0")
|| pcmk__ends_with(first->uuid,
"_" PCMK_ACTION_DEMOTED "_0");
// Update the specified actions for each "then" instance individually
instances = get_instance_list(then->rsc);
for (GList *iter = instances; iter != NULL; iter = iter->next) {
pe_resource_t *first_instance = NULL;
pe_resource_t *then_instance = iter->data;
pe_action_t *first_action = NULL;
pe_action_t *then_action = NULL;
// Find a "first" instance to interleave with this "then" instance
first_instance = pcmk__find_compatible_instance(then_instance,
first->rsc,
pcmk_role_unknown,
current);
if (first_instance == NULL) { // No instance can be interleaved
if (unassign_if_mandatory(first, then, then_instance, type,
current)) {
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
continue;
}
first_action = find_instance_action(first, first_instance,
orig_first_task, node, true);
if (first_action == NULL) {
continue;
}
then_action = find_instance_action(then, then_instance, then->task,
node, false);
if (then_action == NULL) {
continue;
}
if (order_actions(first_action, then_action, type)) {
pcmk__set_updated_flags(changed, first,
pcmk__updated_first|pcmk__updated_then);
}
changed |= then_instance->cmds->update_ordered_actions(
first_action, then_action, node,
first_instance->cmds->action_flags(first_action, node), filter,
type, then->rsc->cluster);
}
free_instance_list(then->rsc, instances);
return changed;
}
/*!
* \internal
* \brief Check whether two actions in an ordering can be interleaved
*
* \param[in] first 'First' action in the ordering
* \param[in] then 'Then' action in the ordering
*
* \return true if \p first and \p then can be interleaved, otherwise false
*/
static bool
can_interleave_actions(const pe_action_t *first, const pe_action_t *then)
{
bool interleave = false;
pe_resource_t *rsc = NULL;
if ((first->rsc == NULL) || (then->rsc == NULL)) {
crm_trace("Not interleaving %s with %s: not resource actions",
first->uuid, then->uuid);
return false;
}
if (first->rsc == then->rsc) {
crm_trace("Not interleaving %s with %s: same resource",
first->uuid, then->uuid);
return false;
}
if ((first->rsc->variant < pcmk_rsc_variant_clone)
|| (then->rsc->variant < pcmk_rsc_variant_clone)) {
crm_trace("Not interleaving %s with %s: not clones or bundles",
first->uuid, then->uuid);
return false;
}
if (pcmk__ends_with(then->uuid, "_stop_0")
|| pcmk__ends_with(then->uuid, "_demote_0")) {
rsc = first->rsc;
} else {
rsc = then->rsc;
}
interleave = crm_is_true(g_hash_table_lookup(rsc->meta,
XML_RSC_ATTR_INTERLEAVE));
pe_rsc_trace(rsc, "'%s then %s' will %sbe interleaved (based on %s)",
first->uuid, then->uuid, (interleave? "" : "not "), rsc->id);
return interleave;
}
/*!
* \internal
* \brief Update non-interleaved instance actions according to an ordering
*
* Given information about an ordering of two non-interleaved actions, update
* the actions' flags (and runnable_before members if appropriate) as
* appropriate for the ordering. Effects may cascade to other orderings
* involving the actions as well.
*
* \param[in,out] instance Clone instance or bundle container
* \param[in,out] first "First" action in ordering
* \param[in] then "Then" action in ordering (for \p instance's parent)
* \param[in] node If not NULL, limit scope of ordering to this node
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static uint32_t
update_noninterleaved_actions(pe_resource_t *instance, pe_action_t *first,
const pe_action_t *then, const pe_node_t *node,
uint32_t flags, uint32_t filter, uint32_t type)
{
pe_action_t *instance_action = NULL;
uint32_t instance_flags = 0;
uint32_t changed = pcmk__updated_none;
// Check whether instance has an equivalent of "then" action
instance_action = find_first_action(instance->actions, NULL, then->task,
node);
if (instance_action == NULL) {
return changed;
}
// Check whether action is runnable
instance_flags = instance->cmds->action_flags(instance_action, node);
if (!pcmk_is_set(instance_flags, pe_action_runnable)) {
return changed;
}
// If so, update actions for the instance
changed = instance->cmds->update_ordered_actions(first, instance_action,
node, flags, filter, type,
instance->cluster);
// Propagate any changes to later actions
if (pcmk_is_set(changed, pcmk__updated_then)) {
for (GList *after_iter = instance_action->actions_after;
after_iter != NULL; after_iter = after_iter->next) {
pe_action_wrapper_t *after = after_iter->data;
pcmk__update_action_for_orderings(after->action, instance->cluster);
}
}
return changed;
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two clone or bundle actions, update
* the actions' flags (and runnable_before members if appropriate) as
* appropriate for the ordering. Effects may cascade to other orderings
* involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pe_action_optional to affect only mandatory
* actions, and pe_action_runnable to affect only
* runnable actions)
* \param[in] type Group of enum pe_ordering flags to apply
* \param[in,out] data_set Cluster working set
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__instance_update_ordered_actions(pe_action_t *first, pe_action_t *then,
const pe_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pe_working_set_t *data_set)
{
CRM_ASSERT((first != NULL) && (then != NULL) && (data_set != NULL));
if (then->rsc == NULL) {
return pcmk__updated_none;
} else if (can_interleave_actions(first, then)) {
return update_interleaved_actions(first, then, node, filter, type);
} else {
uint32_t changed = pcmk__updated_none;
GList *instances = get_instance_list(then->rsc);
// Update actions for the clone or bundle resource itself
changed |= pcmk__update_ordered_actions(first, then, node, flags,
filter, type, data_set);
// Update the 'then' clone instances or bundle containers individually
for (GList *iter = instances; iter != NULL; iter = iter->next) {
pe_resource_t *instance = iter->data;
changed |= update_noninterleaved_actions(instance, first, then,
node, flags, filter, type);
}
free_instance_list(then->rsc, instances);
return changed;
}
}
#define pe__clear_action_summary_flags(flags, action, flag) do { \
flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \
"Action summary", action->rsc->id, \
flags, flag, #flag); \
} while (0)
/*!
* \internal
* \brief Return action flags for a given clone or bundle action
*
* \param[in,out] action Action for a clone or bundle
* \param[in] instances Clone instances or bundle containers
* \param[in] node If not NULL, limit effects to this node
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__collective_action_flags(pe_action_t *action, const GList *instances,
const pe_node_t *node)
{
bool any_runnable = false;
const char *action_name = orig_action_name(action);
// Set original assumptions (optional and runnable may be cleared below)
- uint32_t flags = pe_action_optional|pe_action_runnable|pe_action_pseudo;
+ uint32_t flags = pe_action_optional|pe_action_runnable|pcmk_action_pseudo;
for (const GList *iter = instances; iter != NULL; iter = iter->next) {
const pe_resource_t *instance = iter->data;
const pe_node_t *instance_node = NULL;
pe_action_t *instance_action = NULL;
uint32_t instance_flags;
// Node is relevant only to primitive instances
if (instance->variant == pcmk_rsc_variant_primitive) {
instance_node = node;
}
instance_action = find_first_action(instance->actions, NULL,
action_name, instance_node);
if (instance_action == NULL) {
pe_rsc_trace(action->rsc, "%s has no %s action on %s",
instance->id, action_name, pe__node_name(node));
continue;
}
pe_rsc_trace(action->rsc, "%s has %s for %s on %s",
instance->id, instance_action->uuid, action_name,
pe__node_name(node));
instance_flags = instance->cmds->action_flags(instance_action, node);
// If any instance action is mandatory, so is the collective action
if (pcmk_is_set(flags, pe_action_optional)
&& !pcmk_is_set(instance_flags, pe_action_optional)) {
pe_rsc_trace(instance, "%s is mandatory because %s is",
action->uuid, instance_action->uuid);
pe__clear_action_summary_flags(flags, action, pe_action_optional);
pe__clear_action_flags(action, pe_action_optional);
}
// If any instance action is runnable, so is the collective action
if (pcmk_is_set(instance_flags, pe_action_runnable)) {
any_runnable = true;
}
}
if (!any_runnable) {
pe_rsc_trace(action->rsc,
"%s is not runnable because no instance can run %s",
action->uuid, action_name);
pe__clear_action_summary_flags(flags, action, pe_action_runnable);
if (node == NULL) {
pe__clear_action_flags(action, pe_action_runnable);
}
}
return flags;
}
diff --git a/lib/pacemaker/pcmk_sched_migration.c b/lib/pacemaker/pcmk_sched_migration.c
index 21381f9ab7..75d0cde1de 100644
--- a/lib/pacemaker/pcmk_sched_migration.c
+++ b/lib/pacemaker/pcmk_sched_migration.c
@@ -1,420 +1,420 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Add migration source and target meta-attributes to an action
*
* \param[in,out] action Action to add meta-attributes to
* \param[in] source Node to add as migration source
* \param[in] target Node to add as migration target
*/
static void
add_migration_meta(pe_action_t *action, const pe_node_t *source,
const pe_node_t *target)
{
add_hash_param(action->meta, XML_LRM_ATTR_MIGRATE_SOURCE,
source->details->uname);
add_hash_param(action->meta, XML_LRM_ATTR_MIGRATE_TARGET,
target->details->uname);
}
/*!
* \internal
* \brief Create internal migration actions for a migrateable resource
*
* \param[in,out] rsc Resource to create migration actions for
* \param[in] current Node that resource is originally active on
*/
void
pcmk__create_migration_actions(pe_resource_t *rsc, const pe_node_t *current)
{
pe_action_t *migrate_to = NULL;
pe_action_t *migrate_from = NULL;
pe_action_t *start = NULL;
pe_action_t *stop = NULL;
pe_rsc_trace(rsc, "Creating actions to %smigrate %s from %s to %s",
((rsc->partial_migration_target == NULL)? "" : "partially "),
rsc->id, pe__node_name(current),
pe__node_name(rsc->allocated_to));
start = start_action(rsc, rsc->allocated_to, TRUE);
stop = stop_action(rsc, current, TRUE);
if (rsc->partial_migration_target == NULL) {
migrate_to = custom_action(rsc, pcmk__op_key(rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
PCMK_ACTION_MIGRATE_TO, current, TRUE, TRUE,
rsc->cluster);
}
migrate_from = custom_action(rsc, pcmk__op_key(rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
PCMK_ACTION_MIGRATE_FROM, rsc->allocated_to,
TRUE, TRUE,
rsc->cluster);
if ((migrate_from != NULL)
&& ((migrate_to != NULL) || (rsc->partial_migration_target != NULL))) {
pe__set_action_flags(start, pe_action_migrate_runnable);
pe__set_action_flags(stop, pe_action_migrate_runnable);
// This is easier than trying to delete it from the graph
- pe__set_action_flags(start, pe_action_pseudo);
+ pe__set_action_flags(start, pcmk_action_pseudo);
if (rsc->partial_migration_target == NULL) {
pe__set_action_flags(migrate_from, pe_action_migrate_runnable);
if (migrate_to != NULL) {
pe__set_action_flags(migrate_to, pe_action_migrate_runnable);
migrate_to->needs = start->needs;
}
// Probe -> migrate_to -> migrate_from
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0),
NULL, pe_order_optional, rsc->cluster);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0),
NULL, rsc,
pcmk__op_key(rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
NULL,
pe_order_optional
|pe_order_implies_first_migratable,
rsc->cluster);
} else {
pe__set_action_flags(migrate_from, pe_action_migrate_runnable);
migrate_from->needs = start->needs;
// Probe -> migrate_from (migrate_to already completed)
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL, rsc,
pcmk__op_key(rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
NULL, pe_order_optional, rsc->cluster);
}
// migrate_from before stop or start
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM,
0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pe_order_optional|pe_order_implies_first_migratable,
rsc->cluster);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM,
0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL, pe_order_optional
|pe_order_implies_first_migratable
|pe_order_pseudo_left,
rsc->cluster);
}
if (migrate_to != NULL) {
add_migration_meta(migrate_to, current, rsc->allocated_to);
if (!rsc->is_remote_node) {
/* migrate_to takes place on the source node, but can affect the
* target node depending on how the agent is written. Because of
* this, pending migrate_to actions must be recorded in the CIB,
* in case the source node loses membership while the migrate_to
* action is still in flight.
*
* However we know Pacemaker Remote connection resources don't
* require this, so we skip this for them. (Although it wouldn't
* hurt, and now that record-pending defaults to true, skipping it
* matters even less.)
*/
add_hash_param(migrate_to->meta, XML_OP_ATTR_PENDING, "true");
}
}
if (migrate_from != NULL) {
add_migration_meta(migrate_from, current, rsc->allocated_to);
}
}
/*!
* \internal
* \brief Abort a dangling migration by scheduling a stop (and possibly cleanup)
*
* \param[in] data Source node of dangling migration
* \param[in,out] user_data Resource involved in dangling migration
*/
void
pcmk__abort_dangling_migration(void *data, void *user_data)
{
const pe_node_t *dangling_source = (const pe_node_t *) data;
pe_resource_t *rsc = (pe_resource_t *) user_data;
pe_action_t *stop = NULL;
bool cleanup = pcmk_is_set(rsc->cluster->flags,
pcmk_sched_remove_after_stop);
pe_rsc_trace(rsc,
"Scheduling stop%s for %s on %s due to dangling migration",
(cleanup? " and cleanup" : ""), rsc->id,
pe__node_name(dangling_source));
stop = stop_action(rsc, dangling_source, FALSE);
pe__set_action_flags(stop, pe_action_dangle);
if (cleanup) {
pcmk__schedule_cleanup(rsc, dangling_source, false);
}
}
/*!
* \internal
* \brief Check whether a resource can migrate
*
* \param[in] rsc Resource to check
* \param[in] node Resource's current node
*
* \return true if \p rsc can migrate, otherwise false
*/
bool
pcmk__rsc_can_migrate(const pe_resource_t *rsc, const pe_node_t *current)
{
CRM_CHECK(rsc != NULL, return false);
if (!pcmk_is_set(rsc->flags, pcmk_rsc_migratable)) {
pe_rsc_trace(rsc, "%s cannot migrate because "
"the configuration does not allow it",
rsc->id);
return false;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pe_rsc_trace(rsc, "%s cannot migrate because it is not managed",
rsc->id);
return false;
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pe_rsc_trace(rsc, "%s cannot migrate because it is failed",
rsc->id);
return false;
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
pe_rsc_trace(rsc, "%s cannot migrate because it has a start pending",
rsc->id);
return false;
}
if ((current == NULL) || current->details->unclean) {
pe_rsc_trace(rsc, "%s cannot migrate because "
"its current node (%s) is unclean",
rsc->id, pe__node_name(current));
return false;
}
if ((rsc->allocated_to == NULL) || rsc->allocated_to->details->unclean) {
pe_rsc_trace(rsc, "%s cannot migrate because "
"its next node (%s) is unclean",
rsc->id, pe__node_name(rsc->allocated_to));
return false;
}
return true;
}
/*!
* \internal
* \brief Get an action name from an action or operation key
*
* \param[in] action If not NULL, get action name from here
* \param[in] key If not NULL, get action name from here
*
* \return Newly allocated copy of action name (or NULL if none available)
*/
static char *
task_from_action_or_key(const pe_action_t *action, const char *key)
{
char *res = NULL;
if (action != NULL) {
res = strdup(action->task);
CRM_ASSERT(res != NULL);
} else if (key != NULL) {
parse_op_key(key, NULL, &res, NULL);
}
return res;
}
/*!
* \internal
* \brief Order migration actions equivalent to a given ordering
*
* Orderings involving start, stop, demote, and promote actions must be honored
* during a migration as well, so duplicate any such ordering for the
* corresponding migration actions.
*
* \param[in,out] order Ordering constraint to check
*/
void
pcmk__order_migration_equivalents(pe__ordering_t *order)
{
char *first_task = NULL;
char *then_task = NULL;
bool then_migratable;
bool first_migratable;
// Only orderings between unrelated resources are relevant
if ((order->lh_rsc == NULL) || (order->rh_rsc == NULL)
|| (order->lh_rsc == order->rh_rsc)
|| is_parent(order->lh_rsc, order->rh_rsc)
|| is_parent(order->rh_rsc, order->lh_rsc)) {
return;
}
// Only orderings involving at least one migratable resource are relevant
first_migratable = pcmk_is_set(order->lh_rsc->flags, pcmk_rsc_migratable);
then_migratable = pcmk_is_set(order->rh_rsc->flags, pcmk_rsc_migratable);
if (!first_migratable && !then_migratable) {
return;
}
// Check which actions are involved
first_task = task_from_action_or_key(order->lh_action,
order->lh_action_task);
then_task = task_from_action_or_key(order->rh_action,
order->rh_action_task);
if (pcmk__str_eq(first_task, PCMK_ACTION_START, pcmk__str_none)
&& pcmk__str_eq(then_task, PCMK_ACTION_START, pcmk__str_none)) {
uint32_t flags = pe_order_optional;
if (first_migratable && then_migratable) {
/* A start then B start
* -> A migrate_from then B migrate_to */
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
NULL, flags, order->lh_rsc->cluster);
}
if (then_migratable) {
if (first_migratable) {
pe__set_order_flags(flags, pe_order_apply_first_non_migratable);
}
/* A start then B start
* -> A start then B migrate_to (if start is not part of a
* migration)
*/
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id,
PCMK_ACTION_START, 0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
NULL, flags, order->lh_rsc->cluster);
}
} else if (then_migratable
&& pcmk__str_eq(first_task, PCMK_ACTION_STOP, pcmk__str_none)
&& pcmk__str_eq(then_task, PCMK_ACTION_STOP, pcmk__str_none)) {
uint32_t flags = pe_order_optional;
if (first_migratable) {
pe__set_order_flags(flags, pe_order_apply_first_non_migratable);
}
/* For an ordering "stop A then stop B", if A is moving via restart, and
* B is migrating, enforce that B's migrate_to occurs after A's stop.
*/
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id, PCMK_ACTION_STOP, 0),
NULL,
order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
NULL, flags, order->lh_rsc->cluster);
// Also order B's migrate_from after A's stop during partial migrations
if (order->rh_rsc->partial_migration_target) {
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id, PCMK_ACTION_STOP,
0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
NULL, flags, order->lh_rsc->cluster);
}
} else if (pcmk__str_eq(first_task, PCMK_ACTION_PROMOTE, pcmk__str_none)
&& pcmk__str_eq(then_task, PCMK_ACTION_START, pcmk__str_none)) {
uint32_t flags = pe_order_optional;
if (then_migratable) {
/* A promote then B start
* -> A promote then B migrate_to */
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id,
PCMK_ACTION_PROMOTE, 0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
NULL, flags, order->lh_rsc->cluster);
}
} else if (pcmk__str_eq(first_task, PCMK_ACTION_DEMOTE, pcmk__str_none)
&& pcmk__str_eq(then_task, PCMK_ACTION_STOP, pcmk__str_none)) {
uint32_t flags = pe_order_optional;
if (then_migratable) {
/* A demote then B stop
* -> A demote then B migrate_to */
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id,
PCMK_ACTION_DEMOTE, 0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_TO, 0),
NULL, flags, order->lh_rsc->cluster);
// Order B migrate_from after A demote during partial migrations
if (order->rh_rsc->partial_migration_target) {
pcmk__new_ordering(order->lh_rsc,
pcmk__op_key(order->lh_rsc->id,
PCMK_ACTION_DEMOTE, 0),
NULL, order->rh_rsc,
pcmk__op_key(order->rh_rsc->id,
PCMK_ACTION_MIGRATE_FROM, 0),
NULL, flags, order->lh_rsc->cluster);
}
}
}
free(first_task);
free(then_task);
}
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index ff9c644736..0e1e884a07 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1652 +1,1652 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <stdint.h> // uint8_t, uint32_t
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
static void stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional);
static void promote_resource(pe_resource_t *rsc, pe_node_t *node,
bool optional);
static void assert_role_error(pe_resource_t *rsc, pe_node_t *node,
bool optional);
#define RSC_ROLE_MAX (pcmk_role_promoted + 1)
static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the immediate next role when transitioning from one role
* to a target role. For example, when going from Stopped to Promoted, the
* next role is Unpromoted, because the resource must be started before it
* can be promoted. The current state then becomes Started, which is fed
* into this array again, giving a next role of Promoted.
*
* Current role Immediate next role Final target role
* ------------ ------------------- -----------------
*/
/* Unknown */ { pcmk_role_unknown, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_stopped, /* Unpromoted */
pcmk_role_stopped, /* Promoted */
},
/* Stopped */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_unpromoted, /* Promoted */
},
/* Started */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Unpromoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Promoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_unpromoted, /* Stopped */
pcmk_role_unpromoted, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
};
/*!
* \internal
* \brief Function to schedule actions needed for a role change
*
* \param[in,out] rsc Resource whose role is changing
* \param[in,out] node Node where resource will be in its next role
* \param[in] optional Whether scheduled actions should be optional
*/
typedef void (*rsc_transition_fn)(pe_resource_t *rsc, pe_node_t *node,
bool optional);
static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the function needed to transition directly from one role
* to another. NULL indicates that nothing is needed.
*
* Current role Transition function Next role
* ------------ ------------------- ----------
*/
/* Unknown */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
assert_role_error, /* Started */
assert_role_error, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Stopped */ { assert_role_error, /* Unknown */
NULL, /* Stopped */
start_resource, /* Started */
start_resource, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Started */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
NULL, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Unpromoted */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
stop_resource, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Promoted */ { assert_role_error, /* Unknown */
demote_resource, /* Stopped */
demote_resource, /* Started */
demote_resource, /* Unpromoted */
NULL, /* Promoted */
},
};
/*!
* \internal
* \brief Get a list of a resource's allowed nodes sorted by node score
*
* \param[in] rsc Resource to check
*
* \return List of allowed nodes sorted by node score
*/
static GList *
sorted_allowed_nodes(const pe_resource_t *rsc)
{
if (rsc->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->allowed_nodes);
if (nodes != NULL) {
return pcmk__sort_nodes(nodes, pe__current_node(rsc));
}
}
return NULL;
}
/*!
* \internal
* \brief Assign a resource to its best allowed node, if possible
*
* \param[in,out] rsc Resource to choose a node for
* \param[in] prefer If not \c NULL, prefer this node when all else
* equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return true if \p rsc could be assigned to a node, otherwise false
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
static bool
assign_best_node(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail)
{
GList *nodes = NULL;
pe_node_t *chosen = NULL;
pe_node_t *best = NULL;
const pe_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// We've already finished assignment of resources to nodes
return rsc->allocated_to != NULL;
}
// Sort allowed nodes by score
nodes = sorted_allowed_nodes(rsc);
if (nodes != NULL) {
best = (pe_node_t *) nodes->data; // First node has best score
}
if ((prefer != NULL) && (nodes != NULL)) {
// Get the allowed node version of prefer
chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id);
if (chosen == NULL) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unknown",
pe__node_name(prefer), rsc->id);
/* Favor the preferred node as long as its score is at least as good as
* the best allowed node's.
*
* An alternative would be to favor the preferred node even if the best
* node is better, when the best node's score is less than INFINITY.
*/
} else if (chosen->weight < best->weight) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
pe__node_name(chosen), rsc->id);
chosen = NULL;
} else if (!pcmk__node_available(chosen, true, false)) {
pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pe__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pe_rsc_trace(rsc,
"Chose preferred node %s for %s "
"(ignoring %d candidates)",
pe__node_name(chosen), rsc->id, g_list_length(nodes));
}
}
if ((chosen == NULL) && (best != NULL)) {
/* Either there is no preferred node, or the preferred node is not
* suitable, but another node is allowed to run the resource.
*/
chosen = best;
if (!pe_rsc_is_unique_clone(rsc->parent)
&& (chosen->weight > 0) // Zero not acceptable
&& pcmk__node_available(chosen, false, false)) {
/* If the resource is already running on a node, prefer that node if
* it is just as good as the chosen node.
*
* We don't do this for unique clone instances, because
* pcmk__assign_instances() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unassigned instances to prefer a node that's already
* running another instance.
*/
pe_node_t *running = pe__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pe_rsc_trace(rsc,
"Current node for %s (%s) can't run resources",
rsc->id, pe__node_name(running));
} else {
int nodes_with_best_score = 1;
for (GList *iter = nodes->next; iter; iter = iter->next) {
pe_node_t *allowed = (pe_node_t *) iter->data;
if (allowed->weight != chosen->weight) {
// The nodes are sorted by score, so no more are equal
break;
}
if (pe__same_node(allowed, running)) {
// Scores are equal, so prefer the current node
chosen = allowed;
}
nodes_with_best_score++;
}
if (nodes_with_best_score > 1) {
uint8_t log_level = LOG_INFO;
if (chosen->weight >= INFINITY) {
log_level = LOG_WARNING;
}
do_crm_log(log_level,
"Chose %s for %s from %d nodes with score %s",
pe__node_name(chosen), rsc->id,
nodes_with_best_score,
pcmk_readable_score(chosen->weight));
}
}
}
pe_rsc_trace(rsc, "Chose %s for %s from %d candidates",
pe__node_name(chosen), rsc->id, g_list_length(nodes));
}
pcmk__assign_resource(rsc, chosen, false, stop_if_fail);
g_list_free(nodes);
return rsc->allocated_to != NULL;
}
/*!
* \internal
* \brief Apply a "this with" colocation to a node's allowed node scores
*
* \param[in,out] colocation Colocation to apply
* \param[in,out] rsc Resource being assigned
*/
static void
apply_this_with(pcmk__colocation_t *colocation, pe_resource_t *rsc)
{
GHashTable *archive = NULL;
pe_resource_t *other = colocation->primary;
// In certain cases, we will need to revert the node scores
if ((colocation->dependent_role >= pcmk_role_promoted)
|| ((colocation->score < 0) && (colocation->score > -INFINITY))) {
archive = pcmk__copy_node_table(rsc->allowed_nodes);
}
if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) {
pe_rsc_trace(rsc,
"%s: Assigning colocation %s primary %s first"
"(score=%d role=%s)",
rsc->id, colocation->id, other->id,
colocation->score, role2text(colocation->dependent_role));
other->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->allowed_nodes)) {
pe_rsc_info(rsc,
"%s: Reverting scores from colocation with %s "
"because no nodes allowed",
rsc->id, other->id);
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = archive;
archive = NULL;
}
if (archive != NULL) {
g_hash_table_destroy(archive);
}
}
/*!
* \internal
* \brief Update a Pacemaker Remote node once its connection has been assigned
*
* \param[in] connection Connection resource that has been assigned
*/
static void
remote_connection_assigned(const pe_resource_t *connection)
{
pe_node_t *remote_node = pe_find_node(connection->cluster->nodes,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->allocated_to != NULL)
&& (connection->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->details->id);
remote_node->details->online = TRUE;
if (remote_node->details->unseen) {
// Avoid unnecessary fence, since we will attempt connection
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Pacemaker Remote node %s will be shut down "
"(%sassigned connection's next role is %s)",
remote_node->details->id,
((connection->allocated_to == NULL)? "un" : ""),
role2text(connection->next_role));
remote_node->details->shutdown = TRUE;
}
}
/*!
* \internal
* \brief Assign a primitive resource to a node
*
* \param[in,out] rsc Resource to assign to a node
* \param[in] prefer Node to prefer, if all else is equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pe_node_t *
pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer,
bool stop_if_fail)
{
GList *this_with_colocations = NULL;
GList *with_this_colocations = NULL;
GList *iter = NULL;
pcmk__colocation_t *colocation = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
// Never assign a child without parent being assigned first
if ((rsc->parent != NULL)
&& !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) {
pe_rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, rsc->parent->id);
rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->allocated_to != NULL) {
node_name = pe__node_name(rsc->allocated_to);
}
pe_rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->allocated_to;
}
// Ensure we detect assignment loops
if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) {
pe_rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pe__set_resource_flags(rsc, pcmk_rsc_assigning);
pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes,
rsc->cluster);
this_with_colocations = pcmk__this_with_colocations(rsc);
with_this_colocations = pcmk__with_this_colocations(rsc);
// Apply mandatory colocations first, to satisfy as many as possible
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->allowed_nodes, rsc->cluster);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->next_role == pcmk_role_stopped) {
pe_rsc_trace(rsc,
"Banning %s from all nodes because it will be stopped",
rsc->id);
resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE,
rsc->cluster);
} else if ((rsc->next_role > rsc->role)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate)
&& (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Resource %s cannot be elevated from %s to %s due to "
"no-quorum-policy=freeze",
rsc->id, role2text(rsc->role), role2text(rsc->next_role));
pe__set_next_role(rsc, rsc->role, "no-quorum-policy=freeze");
}
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->allowed_nodes, rsc->cluster);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) {
pe__clear_resource_flags(rsc, pcmk_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pe_node_t *assign_to = NULL;
pe__set_next_role(rsc, rsc->role, "unmanaged");
assign_to = pe__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->role == pcmk_role_promoted) {
reason = "promoted";
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pe_rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id,
(assign_to? assign_to->details->uname : "no node"), reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
pe_rsc_debug(rsc, "Forcing %s to stop: stop-all-resources",
rsc->id);
}
pcmk__assign_resource(rsc, NULL, true, stop_if_fail);
} else if (!assign_best_node(rsc, prefer, stop_if_fail)) {
// Assignment failed
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->running_on != NULL) && stop_if_fail) {
pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
}
pe__clear_resource_flags(rsc, pcmk_rsc_assigning);
if (rsc->is_remote_node) {
remote_connection_assigned(rsc);
}
return rsc->allocated_to;
}
/*!
* \internal
* \brief Schedule actions to bring resource down and back to current role
*
* \param[in,out] rsc Resource to restart
* \param[in,out] current Node that resource should be brought down on
* \param[in] need_stop Whether the resource must be stopped
* \param[in] need_promote Whether the resource must be promoted
*
* \return Role that resource would have after scheduled actions are taken
*/
static void
schedule_restart_actions(pe_resource_t *rsc, pe_node_t *current,
bool need_stop, bool need_promote)
{
enum rsc_role_e role = rsc->role;
enum rsc_role_e next_role;
rsc_transition_fn fn = NULL;
pe__set_resource_flags(rsc, 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];
pe_rsc_trace(rsc, "Creating %s action to take %s down from %s to %s",
(need_stop? "required" : "optional"), rsc->id,
role2text(role), role2text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, current, !need_stop);
role = next_role;
}
// Bring resource up to its next role on its next node
while ((rsc->role <= rsc->next_role) && (role != rsc->role)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pe_rsc_trace(rsc, "Creating %s action to take %s up from %s to %s",
(required? "required" : "optional"), rsc->id,
role2text(role), role2text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, !required);
role = next_role;
}
pe__clear_resource_flags(rsc, 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(pe_resource_t *rsc)
{
if (rsc->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->allocated_to == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "assignment");
} else {
pe__set_next_role(rsc, pcmk_role_started, "assignment");
}
return "implicit";
}
/*!
* \internal
* \brief Create an action to represent an already pending start
*
* \param[in,out] rsc Resource to create start action for
*/
static void
create_pending_start(pe_resource_t *rsc)
{
pe_action_t *start = NULL;
pe_rsc_trace(rsc,
"Creating action for %s to represent already pending start",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
pe__set_action_flags(start, pe_action_print_always);
}
/*!
* \internal
* \brief Schedule actions needed to take a resource to its next role
*
* \param[in,out] rsc Resource to schedule actions for
*/
static void
schedule_role_transition_actions(pe_resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
while (role != rsc->next_role) {
enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role];
rsc_transition_fn fn = NULL;
pe_rsc_trace(rsc,
"Creating action to take %s from %s to %s (ending at %s)",
rsc->id, role2text(role), role2text(next_role),
role2text(rsc->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, false);
role = next_role;
}
}
/*!
* \internal
* \brief Create all actions needed for a given primitive resource
*
* \param[in,out] rsc Primitive resource to create actions for
*/
void
pcmk__primitive_create_actions(pe_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pe_node_t *current = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
next_role_source = set_default_next_role(rsc);
pe_rsc_trace(rsc,
"Creating all actions for %s transition from %s to %s "
"(%s) on %s",
rsc->id, role2text(rsc->role), role2text(rsc->next_role),
next_role_source, pe__node_name(rsc->allocated_to));
current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active);
g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration,
rsc);
if ((current != NULL) && (rsc->allocated_to != NULL)
&& !pe__same_node(current, rsc->allocated_to)
&& (rsc->next_role >= pcmk_role_started)) {
pe_rsc_trace(rsc, "Moving %s from %s to %s",
rsc->id, pe__node_name(current),
pe__node_name(rsc->allocated_to));
is_moving = true;
allow_migrate = pcmk__rsc_can_migrate(rsc, current);
// This is needed even if migrating (though I'm not sure why ...)
need_stop = true;
}
// Check whether resource is partially migrated and/or multiply active
if ((rsc->partial_migration_source != NULL)
&& (rsc->partial_migration_target != NULL)
&& allow_migrate && (num_all_active == 2)
&& pe__same_node(current, rsc->partial_migration_source)
&& pe__same_node(rsc->allocated_to, rsc->partial_migration_target)) {
/* A partial migration is in progress, and the migration target remains
* the same as when the migration began.
*/
pe_rsc_trace(rsc, "Partial migration of %s from %s to %s will continue",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
} else if ((rsc->partial_migration_source != NULL)
|| (rsc->partial_migration_target != NULL)) {
// A partial migration is in progress but can't be continued
if (num_all_active > 2) {
// The resource is migrating *and* multiply active!
crm_notice("Forcing recovery of %s because it is migrating "
"from %s to %s and possibly active elsewhere",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
} else {
// The migration source or target isn't available
crm_notice("Forcing recovery of %s because it can no longer "
"migrate from %s to %s",
rsc->id, pe__node_name(rsc->partial_migration_source),
pe__node_name(rsc->partial_migration_target));
}
need_stop = true;
rsc->partial_migration_source = rsc->partial_migration_target = NULL;
allow_migrate = false;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
multiply_active = (num_all_active > 1);
} else {
/* If a resource has "requires" set to nothing or quorum, don't consider
* it active on unclean nodes (similar to how all resources behave when
* stonith-enabled is false). We can start such resources elsewhere
* before fencing completes, and if we considered the resource active on
* the failed node, we would attempt recovery for being active on
* multiple nodes.
*/
multiply_active = (num_clean_active > 1);
}
if (multiply_active) {
const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
// Resource was (possibly) incorrectly multiply active
pe_proc_err("%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), rsc->id, num_all_active,
recovery2text(rsc->recovery_type));
crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ"
"#Resource_is_Too_Active for more information");
switch (rsc->recovery_type) {
case pcmk_multiply_active_restart:
need_stop = true;
break;
case pcmk_multiply_active_unexpected:
need_stop = true; // stop_resource() will skip expected node
pe__set_resource_flags(rsc, pcmk_rsc_stop_unexpected);
break;
default:
break;
}
} else {
pe__clear_resource_flags(rsc, pcmk_rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
create_pending_start(rsc);
}
if (is_moving) {
// Remaining tests are only for resources staying where they are
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) {
need_stop = true;
pe_rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pe_rsc_trace(rsc, "Recovering %s by demotion", rsc->id);
if (rsc->next_role == pcmk_role_promoted) {
need_promote = true;
}
}
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
pe_rsc_trace(rsc, "Blocking further actions on %s", rsc->id);
need_stop = true;
} else if ((rsc->role > pcmk_role_started) && (current != NULL)
&& (rsc->allocated_to != NULL)) {
pe_action_t *start = NULL;
pe_rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
if (!pcmk_is_set(start->flags, pe_action_optional)) {
// Recovery of a promoted resource
pe_rsc_trace(rsc, "%s restart is required for recovery", rsc->id);
need_stop = true;
}
}
// Create any actions needed to bring resource down and back up to same role
schedule_restart_actions(rsc, current, need_stop, need_promote);
// Create any actions needed to take resource from this role to the next
schedule_role_transition_actions(rsc);
pcmk__create_recurring_actions(rsc);
if (allow_migrate) {
pcmk__create_migration_actions(rsc, current);
}
}
/*!
* \internal
* \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes
*
* \param[in] rsc Resource to check
*/
static void
rsc_avoids_remote_nodes(const pe_resource_t *rsc)
{
GHashTableIter iter;
pe_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (node->details->remote_rsc != NULL) {
node->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Return allowed nodes as (possibly sorted) list
*
* Convert a resource's hash table of allowed nodes to a list. If printing to
* stdout, sort the list, to keep action ID numbers consistent for regression
* test output (while avoiding the performance hit on a live cluster).
*
* \param[in] rsc Resource to check for allowed nodes
*
* \return List of resource's allowed nodes
* \note Callers should take care not to rely on the list being sorted.
*/
static GList *
allowed_nodes_as_list(const pe_resource_t *rsc)
{
GList *allowed_nodes = NULL;
if (rsc->allowed_nodes) {
allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
if (!pcmk__is_daemon) {
allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name);
}
return allowed_nodes;
}
/*!
* \internal
* \brief Create implicit constraints needed for a primitive resource
*
* \param[in,out] rsc Primitive resource to create implicit constraints for
*/
void
pcmk__primitive_internal_constraints(pe_resource_t *rsc)
{
GList *allowed_nodes = NULL;
bool check_unfencing = false;
bool check_utilization = false;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pe_rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource %s",
rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags,
pcmk_sched_enable_unfencing)
&& pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& !pcmk__str_eq(rsc->cluster->placement_strategy,
"default", pcmk__str_casei);
// Order stops before starts (i.e. restart)
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL,
pe_order_optional|pe_order_implies_then|pe_order_restart,
rsc->cluster);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable)
|| (rsc->role > pcmk_role_unpromoted)) {
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pe_order_promoted_implies_first, rsc->cluster);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0),
NULL,
pe_order_runnable_left, rsc->cluster);
}
// Don't clear resource history if probing on same node
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL, pe_order_same_node|pe_order_then_cancels_first,
rsc->cluster);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization || (rsc->container != NULL)) {
allowed_nodes = allowed_nodes_as_list(rsc);
}
if (check_unfencing) {
g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc);
}
if (check_utilization) {
pcmk__create_utilization_constraints(rsc, allowed_nodes);
}
if (rsc->container != NULL) {
pe_resource_t *remote_rsc = NULL;
if (rsc->is_remote_node) {
// rsc is the implicit remote connection for a guest or bundle node
/* Guest resources are not allowed to run on Pacemaker Remote nodes,
* to avoid nesting remotes. However, bundles are allowed.
*/
if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
rsc_avoids_remote_nodes(rsc->container);
}
/* If someone cleans up a guest or bundle node's container, we will
* likely schedule a (re-)probe of the container and recovery of the
* connection. Order the connection stop after the container probe,
* so that if we detect the container running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pe_order_optional);
/* A user can specify that a resource must start on a Pacemaker Remote
* node by explicitly configuring it with the container=NODENAME
* meta-attribute. This is of questionable merit, since location
* constraints can accomplish the same thing. But we support it, so here
* we check whether a resource (that is not itself a remote connection)
* has container set to a remote node or guest node resource.
*/
} else if (rsc->container->is_remote_node) {
remote_rsc = rsc->container;
} else {
remote_rsc = pe__resource_contains_guest_node(rsc->cluster,
rsc->container);
}
if (remote_rsc != NULL) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the container resource.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pe_node_t *node = item->data;
if (node->details->remote_rsc != remote_rsc) {
node->weight = -INFINITY;
}
}
} else {
/* This resource is either a filler for a container that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its container %s",
rsc->id, rsc->container->id);
pcmk__new_ordering(rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_START, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
pe_order_implies_then|pe_order_runnable_left,
rsc->cluster);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_STOP, 0),
NULL, pe_order_implies_first, rsc->cluster);
if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->container, NULL, NULL,
pcmk__coloc_influence);
}
}
if (rsc->is_remote_node
|| pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
/* Remote connections and fencing devices are not allowed to run on
* Pacemaker Remote nodes
*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint, apply its score to the dependent's
* allowed node scores (if we are still placing resources) or priority (if
* we are choosing promotable clone instance roles).
*
* \param[in,out] dependent Dependent resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
* \param[in] for_dependent true if called on behalf of dependent
*/
void
pcmk__primitive_apply_coloc_score(pe_resource_t *dependent,
const pe_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
enum pcmk__coloc_affects filter_results;
CRM_ASSERT((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
// Always process on behalf of primary resource
primary->cmds->apply_coloc_score(dependent, primary, colocation, false);
return;
}
filter_results = pcmk__colocation_affects(dependent, primary, colocation,
false);
pe_rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)",
((colocation->score > 0)? "Colocating" : "Anti-colocating"),
dependent->id, primary->id, colocation->id, colocation->score,
filter_results);
switch (filter_results) {
case pcmk__coloc_affects_role:
pcmk__apply_coloc_to_priority(dependent, primary, colocation);
break;
case pcmk__coloc_affects_location:
pcmk__apply_coloc_to_scores(dependent, primary, colocation);
break;
default: // pcmk__coloc_affects_nothing
return;
}
}
/* Primitive implementation of
* resource_alloc_functions_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_with_this(list, colocation, orig_rsc);
}
}
}
}
/* Primitive implementation of
* resource_alloc_functions_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
}
}
/*!
* \internal
* \brief Return action flags for a given primitive resource action
*
* \param[in,out] action Action to get flags for
* \param[in] node If not NULL, limit effects to this node (ignored)
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node)
{
CRM_ASSERT(action != NULL);
return (uint32_t) action->flags;
}
/*!
* \internal
* \brief Check whether a node is a multiply active resource's expected node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p rsc is multiply active with multiple-active set to
* stop_unexpected, and \p node is the node where it will remain active
* \note This assumes that the resource's next role cannot be changed to stopped
* after this is called, which should be reasonable if status has already
* been unpacked and resources have been assigned to nodes.
*/
static bool
is_expected_node(const pe_resource_t *rsc, const pe_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pcmk_rsc_stop_unexpected|pcmk_rsc_restarting)
&& (rsc->next_role > pcmk_role_stopped)
&& pe__same_node(rsc->allocated_to, node);
}
/*!
* \internal
* \brief Schedule actions needed to stop a resource wherever it is active
*
* \param[in,out] rsc Resource being stopped
* \param[in] node Node where resource is being stopped (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *current = (pe_node_t *) iter->data;
pe_action_t *stop = NULL;
if (is_expected_node(rsc, current)) {
/* We are scheduling restart actions for a multiply active resource
* with multiple-active=stop_unexpected, and this is where it should
* not be stopped.
*/
pe_rsc_trace(rsc,
"Skipping stop of multiply active resource %s "
"on expected node %s",
rsc->id, pe__node_name(current));
continue;
}
if (rsc->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pe__same_node(current, rsc->partial_migration_target)
&& pe__same_node(current, rsc->allocated_to)) {
pe_rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pe__node_name(current));
continue;
} else {
pe_rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pe__node_name(current));
optional = false;
}
}
pe_rsc_trace(rsc, "Scheduling stop of %s on %s",
rsc->id, pe__node_name(current));
stop = stop_action(rsc, current, optional);
if (rsc->allocated_to == NULL) {
pe_action_set_reason(stop, "node availability", true);
} else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting
|pcmk_rsc_stop_unexpected)) {
/* We are stopping a multiply active resource on a node that is
* not its expected node, and we are still scheduling restart
* actions, so the stop is for being multiply active.
*/
pe_action_set_reason(stop, "being multiply active", true);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pe__clear_action_flags(stop, pe_action_runnable);
}
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) {
pcmk__schedule_cleanup(rsc, current, optional);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) {
pe_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false, rsc->cluster);
order_actions(stop, unfence, pe_order_implies_first);
if (!pcmk__node_unfenced(current)) {
pe_proc_err("Stopping %s until %s can be unfenced",
rsc->id, pe__node_name(current));
}
}
}
}
/*!
* \internal
* \brief Schedule actions needed to start a resource on a node
*
* \param[in,out] rsc Resource being started
* \param[in,out] node Node where resource should be started
* \param[in] optional Whether actions should be optional
*/
static void
start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
pe_action_t *start = NULL;
CRM_ASSERT(node != NULL);
pe_rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)",
(optional? "optional" : "required"), rsc->id,
pe__node_name(node), node->weight);
start = start_action(rsc, node, TRUE);
pcmk__order_vs_unfence(rsc, node, start, pe_order_implies_then);
if (pcmk_is_set(start->flags, pe_action_runnable) && !optional) {
pe__clear_action_flags(start, pe_action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pe_rsc_trace(rsc,
"Start of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pe__node_name(node));
- pe__set_action_flags(start, pe_action_pseudo);
+ pe__set_action_flags(start, pcmk_action_pseudo);
}
}
/*!
* \internal
* \brief Schedule actions needed to promote a resource on a node
*
* \param[in,out] rsc Resource being promoted
* \param[in] node Node where resource should be promoted
* \param[in] optional Whether actions should be optional
*/
static void
promote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
GList *iter = NULL;
GList *action_list = NULL;
bool runnable = true;
CRM_ASSERT(node != NULL);
// Any start must be runnable for promotion to be runnable
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pe_action_t *start = (pe_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pe_action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pe_action_t *promote = promote_action(rsc, node, optional);
pe_rsc_trace(rsc, "Scheduling %s promotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pe__node_name(node));
if (is_expected_node(rsc, node)) {
/* This could be a problem if the promote becomes necessary for
* other reasons later.
*/
pe_rsc_trace(rsc,
"Promotion of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pe__node_name(node));
- pe__set_action_flags(promote, pe_action_pseudo);
+ pe__set_action_flags(promote, pcmk_action_pseudo);
}
} else {
pe_rsc_trace(rsc, "Not promoting %s on %s: start unrunnable",
rsc->id, pe__node_name(node));
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE,
true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pe_action_t *promote = (pe_action_t *) iter->data;
pe__clear_action_flags(promote, pe_action_runnable);
}
g_list_free(action_list);
}
}
/*!
* \internal
* \brief Schedule actions needed to demote a resource wherever it is active
*
* \param[in,out] rsc Resource being demoted
* \param[in] node Node where resource should be demoted (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
/* Since this will only be called for a primitive (possibly as an instance
* of a collective resource), the resource is multiply active if it is
* running on more than one node, so we want to demote on all of them as
* part of recovery, regardless of which one is the desired node.
*/
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *current = (pe_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pe_rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pe__node_name(current));
} else {
pe_rsc_trace(rsc, "Scheduling %s demotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pe__node_name(current));
demote_action(rsc, current, optional);
}
}
}
static void
assert_role_error(pe_resource_t *rsc, pe_node_t *node, bool optional)
{
CRM_ASSERT(false);
}
/*!
* \internal
* \brief Schedule cleanup of a resource
*
* \param[in,out] rsc Resource to clean up
* \param[in] node Node to clean up on
* \param[in] optional Whether clean-up should be optional
*/
void
pcmk__schedule_cleanup(pe_resource_t *rsc, const pe_node_t *node, bool optional)
{
/* If the cleanup is required, its orderings are optional, because they're
* relevant only if both actions are required. Conversely, if the cleanup is
* optional, the orderings make the then action required if the first action
* becomes required.
*/
uint32_t flag = optional? pe_order_implies_then : pe_order_optional;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed",
rsc->id, pe__node_name(node));
return;
}
if (node->details->unclean || !node->details->online) {
pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable",
rsc->id, pe__node_name(node));
return;
}
crm_notice("Scheduling clean-up of %s on %s", rsc->id, pe__node_name(node));
delete_action(rsc, node, optional);
// stop -> clean-up -> start
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_DELETE, flag);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE,
rsc, PCMK_ACTION_START, flag);
}
/*!
* \internal
* \brief Add primitive meta-attributes relevant to graph actions to XML
*
* \param[in] rsc Primitive resource whose meta-attributes should be added
* \param[in,out] xml Transition graph action attributes XML to add to
*/
void
pcmk__primitive_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pe_resource_t *parent = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (xml != NULL));
/* Clone instance numbers get set internally as meta-attributes, and are
* needed in the transition graph (for example, to tell unique clone
* instances apart).
*/
value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION);
if (value != NULL) {
name = crm_meta_name(XML_RSC_ATTR_INCARNATION);
crm_xml_add(xml, name, value);
free(name);
}
// Not sure if this one is really needed ...
value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE);
if (value != NULL) {
name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE);
crm_xml_add(xml, name, value);
free(name);
}
/* The container meta-attribute can be set on the primitive itself or one of
* its parents (for example, a group inside a container resource), so check
* them all, and keep the highest one found.
*/
for (parent = rsc; parent != NULL; parent = parent->parent) {
if (parent->container != NULL) {
crm_xml_add(xml, CRM_META "_" XML_RSC_ATTR_CONTAINER,
parent->container->id);
}
}
/* Bundle replica children will get their external-ip set internally as a
* meta-attribute. The graph action needs it, but under a different naming
* convention than other meta-attributes.
*/
value = g_hash_table_lookup(rsc->meta, "external-ip");
if (value != NULL) {
crm_xml_add(xml, "pcmk_external_ip", value);
}
}
// Primitive implementation of resource_alloc_functions_t:add_utilization()
void
pcmk__primitive_add_utilization(const pe_resource_t *rsc,
const pe_resource_t *orig_rsc, GList *all_rscs,
GHashTable *utilization)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (orig_rsc != NULL) && (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return;
}
pe_rsc_trace(orig_rsc, "%s: Adding primitive %s as colocated utilization",
orig_rsc->id, rsc->id);
pcmk__release_node_capacity(utilization, rsc);
}
/*!
* \internal
* \brief Get epoch time of node's shutdown attribute (or now if none)
*
* \param[in,out] node Node to check
*
* \return Epoch time corresponding to shutdown attribute if set or now if not
*/
static time_t
shutdown_time(pe_node_t *node)
{
const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
time_t result = 0;
if (shutdown != NULL) {
long long result_ll;
if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) {
result = (time_t) result_ll;
}
}
return (result == 0)? get_effective_time(node->details->data_set) : result;
}
/*!
* \internal
* \brief Ban a resource from a node if it's not locked to the node
*
* \param[in] data Node to check
* \param[in,out] user_data Resource to check
*/
static void
ban_if_not_locked(gpointer data, gpointer user_data)
{
const pe_node_t *node = (const pe_node_t *) data;
pe_resource_t *rsc = (pe_resource_t *) user_data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
XML_CONFIG_ATTR_SHUTDOWN_LOCK, rsc->cluster);
}
}
// Primitive implementation of resource_alloc_functions_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pe_resource_t *rsc)
{
const char *class = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
// Fence devices and remote connections can't be locked
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches)
|| pe__resource_is_remote_conn(rsc)) {
return;
}
if (rsc->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->running_on != NULL) {
/* The resource was started elsewhere even though it is now
* considered locked. This shouldn't be possible, but as a
* failsafe, we don't want to disturb the resource now.
*/
pe_rsc_info(rsc,
"Cancelling shutdown lock because %s is already active",
rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node, rsc->cluster);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pe_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown",
rsc->id, pe__node_name(node));
} else {
rsc->lock_node = node;
rsc->lock_time = shutdown_time(node);
}
}
}
if (rsc->lock_node == NULL) {
// No lock needed
return;
}
if (rsc->cluster->shutdown_lock > 0) {
time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock;
pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pe__node_name(rsc->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, rsc->cluster);
} else {
pe_rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pe__node_name(rsc->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pacemaker/pcmk_sched_probes.c b/lib/pacemaker/pcmk_sched_probes.c
index 054bac072c..0ebfde3c82 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, pe_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, pe_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, pe_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)) {
return;
}
pe__set_action_flags(after, pe_action_tracking);
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, pe_action_pseudo)) {
+ 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);
}
}
/*!
* \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);
}
}
diff --git a/lib/pacemaker/pcmk_sched_remote.c b/lib/pacemaker/pcmk_sched_remote.c
index de86bc31f9..c4d20d040b 100644
--- a/lib/pacemaker/pcmk_sched_remote.c
+++ b/lib/pacemaker/pcmk_sched_remote.c
@@ -1,725 +1,725 @@
/*
* 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 <crm/common/xml_internal.h>
#include <glib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
enum remote_connection_state {
remote_state_unknown = 0,
remote_state_alive = 1,
remote_state_resting = 2,
remote_state_failed = 3,
remote_state_stopped = 4
};
static const char *
state2text(enum remote_connection_state state)
{
switch (state) {
case remote_state_unknown:
return "unknown";
case remote_state_alive:
return "alive";
case remote_state_resting:
return "resting";
case remote_state_failed:
return "failed";
case remote_state_stopped:
return "stopped";
}
return "impossible";
}
/* We always use pe_order_preserve with these convenience functions to exempt
* internally generated constraints from the prohibition of user constraints
* involving remote connection resources.
*
* The start ordering additionally uses pe_order_runnable_left so that the
* specified action is not runnable if the start is not runnable.
*/
static inline void
order_start_then_action(pe_resource_t *first_rsc, pe_action_t *then_action,
uint32_t extra)
{
if ((first_rsc != NULL) && (then_action != NULL)) {
pcmk__new_ordering(first_rsc, start_key(first_rsc), NULL,
then_action->rsc, NULL, then_action,
pe_order_preserve|pe_order_runnable_left|extra,
first_rsc->cluster);
}
}
static inline void
order_action_then_stop(pe_action_t *first_action, pe_resource_t *then_rsc,
uint32_t extra)
{
if ((first_action != NULL) && (then_rsc != NULL)) {
pcmk__new_ordering(first_action->rsc, NULL, first_action,
then_rsc, stop_key(then_rsc), NULL,
pe_order_preserve|extra, then_rsc->cluster);
}
}
static enum remote_connection_state
get_remote_node_state(const pe_node_t *node)
{
const pe_resource_t *remote_rsc = NULL;
const pe_node_t *cluster_node = NULL;
CRM_ASSERT(node != NULL);
remote_rsc = node->details->remote_rsc;
CRM_ASSERT(remote_rsc != NULL);
cluster_node = pe__current_node(remote_rsc);
/* If the cluster node the remote connection resource resides on
* is unclean or went offline, we can't process any operations
* on that remote node until after it starts elsewhere.
*/
if ((remote_rsc->next_role == pcmk_role_stopped)
|| (remote_rsc->allocated_to == NULL)) {
// The connection resource is not going to run anywhere
if ((cluster_node != NULL) && cluster_node->details->unclean) {
/* The remote connection is failed because its resource is on a
* failed node and can't be recovered elsewhere, so we must fence.
*/
return remote_state_failed;
}
if (!pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)) {
/* Connection resource is cleanly stopped */
return remote_state_stopped;
}
/* Connection resource is failed */
if ((remote_rsc->next_role == pcmk_role_stopped)
&& remote_rsc->remote_reconnect_ms
&& node->details->remote_was_fenced
&& !pe__shutdown_requested(node)) {
/* We won't know whether the connection is recoverable until the
* reconnect interval expires and we reattempt connection.
*/
return remote_state_unknown;
}
/* The remote connection is in a failed state. If there are any
* resources known to be active on it (stop) or in an unknown state
* (probe), we must assume the worst and fence it.
*/
return remote_state_failed;
} else if (cluster_node == NULL) {
/* Connection is recoverable but not currently running anywhere, so see
* if we can recover it first
*/
return remote_state_unknown;
} else if (cluster_node->details->unclean
|| !(cluster_node->details->online)) {
// Connection is running on a dead node, see if we can recover it first
return remote_state_resting;
} else if (pcmk__list_of_multiple(remote_rsc->running_on)
&& (remote_rsc->partial_migration_source != NULL)
&& (remote_rsc->partial_migration_target != NULL)) {
/* We're in the middle of migrating a connection resource, so wait until
* after the migration completes before performing any actions.
*/
return remote_state_resting;
}
return remote_state_alive;
}
/*!
* \internal
* \brief Order actions on remote node relative to actions for the connection
*
* \param[in,out] action An action scheduled on a Pacemaker Remote node
*/
static void
apply_remote_ordering(pe_action_t *action)
{
pe_resource_t *remote_rsc = NULL;
enum action_tasks task = text2task(action->task);
enum remote_connection_state state = get_remote_node_state(action->node);
uint32_t order_opts = pe_order_none;
if (action->rsc == NULL) {
return;
}
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc != NULL);
crm_trace("Order %s action %s relative to %s%s (state: %s)",
action->task, action->uuid,
pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)? "failed " : "",
remote_rsc->id, state2text(state));
if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* Migration ops map to pcmk_action_unspecified, but we need to apply
* the same ordering as for stop or demote (see get_router_node()).
*/
task = pcmk_action_stop;
}
switch (task) {
case pcmk_action_start:
case pcmk_action_promote:
order_opts = pe_order_none;
if (state == remote_state_failed) {
/* Force recovery, by making this action required */
pe__set_order_flags(order_opts, pe_order_implies_then);
}
/* Ensure connection is up before running this action */
order_start_then_action(remote_rsc, action, order_opts);
break;
case pcmk_action_stop:
if (state == remote_state_alive) {
order_action_then_stop(action, remote_rsc,
pe_order_implies_first);
} else if (state == remote_state_failed) {
/* The resource is active on the node, but since we don't have a
* valid connection, the only way to stop the resource is by
* fencing the node. There is no need to order the stop relative
* to the remote connection, since the stop will become implied
* by the fencing.
*/
pe_fence_node(remote_rsc->cluster, action->node,
"resources are active but "
"connection is unrecoverable",
FALSE);
} else if (remote_rsc->next_role == pcmk_role_stopped) {
/* State must be remote_state_unknown or remote_state_stopped.
* Since the connection is not coming back up in this
* transition, stop this resource first.
*/
order_action_then_stop(action, remote_rsc,
pe_order_implies_first);
} else {
/* The connection is going to be started somewhere else, so
* stop this resource after that completes.
*/
order_start_then_action(remote_rsc, action, pe_order_none);
}
break;
case pcmk_action_demote:
/* Only order this demote relative to the connection start if the
* connection isn't being torn down. Otherwise, the demote would be
* blocked because the connection start would not be allowed.
*/
if ((state == remote_state_resting)
|| (state == remote_state_unknown)) {
order_start_then_action(remote_rsc, action, pe_order_none);
} /* Otherwise we can rely on the stop ordering */
break;
default:
/* Wait for the connection resource to be up */
if (pcmk__action_is_recurring(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
order_start_then_action(remote_rsc, action,
pe_order_implies_then);
} else {
pe_node_t *cluster_node = pe__current_node(remote_rsc);
if ((task == pcmk_action_monitor) && (state == remote_state_failed)) {
/* We would only be here if we do not know the state of the
* resource on the remote node. Since we have no way to find
* out, it is necessary to fence the node.
*/
pe_fence_node(remote_rsc->cluster, action->node,
"resources are in unknown state "
"and connection is unrecoverable", FALSE);
}
if ((cluster_node != NULL) && (state == remote_state_stopped)) {
/* The connection is currently up, but is going down
* permanently. Make sure we check services are actually
* stopped _before_ we let the connection get closed.
*/
order_action_then_stop(action, remote_rsc,
pe_order_runnable_left);
} else {
order_start_then_action(remote_rsc, action, pe_order_none);
}
}
break;
}
}
static void
apply_container_ordering(pe_action_t *action)
{
/* VMs are also classified as containers for these purposes... in
* that they both involve a 'thing' running on a real or remote
* cluster node.
*
* This allows us to be smarter about the type and extent of
* recovery actions required in various scenarios
*/
pe_resource_t *remote_rsc = NULL;
pe_resource_t *container = NULL;
enum action_tasks task = text2task(action->task);
CRM_ASSERT(action->rsc != NULL);
CRM_ASSERT(action->node != NULL);
CRM_ASSERT(pe__is_guest_or_remote_node(action->node));
remote_rsc = action->node->details->remote_rsc;
CRM_ASSERT(remote_rsc != NULL);
container = remote_rsc->container;
CRM_ASSERT(container != NULL);
if (pcmk_is_set(container->flags, pcmk_rsc_failed)) {
pe_fence_node(action->rsc->cluster, action->node, "container failed",
FALSE);
}
crm_trace("Order %s action %s relative to %s%s for %s%s",
action->task, action->uuid,
pcmk_is_set(remote_rsc->flags, pcmk_rsc_failed)? "failed " : "",
remote_rsc->id,
pcmk_is_set(container->flags, pcmk_rsc_failed)? "failed " : "",
container->id);
if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* Migration ops map to pcmk_action_unspecified, but we need to apply
* the same ordering as for stop or demote (see get_router_node()).
*/
task = pcmk_action_stop;
}
switch (task) {
case pcmk_action_start:
case pcmk_action_promote:
// Force resource recovery if the container is recovered
order_start_then_action(container, action, pe_order_implies_then);
// Wait for the connection resource to be up, too
order_start_then_action(remote_rsc, action, pe_order_none);
break;
case pcmk_action_stop:
case pcmk_action_demote:
if (pcmk_is_set(container->flags, pcmk_rsc_failed)) {
/* When the container representing a guest node fails, any stop
* or demote actions for resources running on the guest node
* are implied by the container stopping. This is similar to
* how fencing operations work for cluster nodes and remote
* nodes.
*/
} else {
/* Ensure the operation happens before the connection is brought
* down.
*
* If we really wanted to, we could order these after the
* connection start, IFF the container's current role was
* stopped (otherwise we re-introduce an ordering loop when the
* connection is restarting).
*/
order_action_then_stop(action, remote_rsc, pe_order_none);
}
break;
default:
/* Wait for the connection resource to be up */
if (pcmk__action_is_recurring(action)) {
/* In case we ever get the recovery logic wrong, force
* recurring monitors to be restarted, even if just
* the connection was re-established
*/
if (task != pcmk_action_unspecified) {
order_start_then_action(remote_rsc, action,
pe_order_implies_then);
}
} else {
order_start_then_action(remote_rsc, action, pe_order_none);
}
break;
}
}
/*!
* \internal
* \brief Order all relevant actions relative to remote connection actions
*
* \param[in,out] data_set Cluster working set
*/
void
pcmk__order_remote_connection_actions(pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pcmk_sched_have_remote_nodes)) {
return;
}
crm_trace("Creating remote connection orderings");
for (GList *iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = iter->data;
pe_resource_t *remote = NULL;
// We are only interested in resource actions
if (action->rsc == NULL) {
continue;
}
/* Special case: If we are clearing the failcount of an actual
* remote connection resource, then make sure this happens before
* any start of the resource in this transition.
*/
if (action->rsc->is_remote_node &&
pcmk__str_eq(action->task, PCMK_ACTION_CLEAR_FAILCOUNT,
pcmk__str_none)) {
pcmk__new_ordering(action->rsc, NULL, action, action->rsc,
pcmk__op_key(action->rsc->id, PCMK_ACTION_START,
0),
NULL, pe_order_optional, data_set);
continue;
}
// We are only interested in actions assigned to a node
if (action->node == NULL) {
continue;
}
if (!pe__is_guest_or_remote_node(action->node)) {
continue;
}
/* We are only interested in real actions.
*
* @TODO This is probably wrong; pseudo-actions might be converted to
* real actions and vice versa later in update_actions() at the end of
* pcmk__apply_orderings().
*/
- if (pcmk_is_set(action->flags, pe_action_pseudo)) {
+ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
continue;
}
remote = action->node->details->remote_rsc;
if (remote == NULL) {
// Orphaned
continue;
}
/* Another special case: if a resource is moving to a Pacemaker Remote
* node, order the stop on the original node after any start of the
* remote connection. This ensures that if the connection fails to
* start, we leave the resource running on the original node.
*/
if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_none)) {
for (GList *item = action->rsc->actions; item != NULL;
item = item->next) {
pe_action_t *rsc_action = item->data;
if (!pe__same_node(rsc_action->node, action->node)
&& pcmk__str_eq(rsc_action->task, PCMK_ACTION_STOP,
pcmk__str_none)) {
pcmk__new_ordering(remote, start_key(remote), NULL,
action->rsc, NULL, rsc_action,
pe_order_optional, data_set);
}
}
}
/* The action occurs across a remote connection, so create
* ordering constraints that guarantee the action occurs while the node
* is active (after start, before stop ... things like that).
*
* This is somewhat brittle in that we need to make sure the results of
* this ordering are compatible with the result of get_router_node().
* It would probably be better to add XML_LRM_ATTR_ROUTER_NODE as part
* of this logic rather than create_graph_action().
*/
if (remote->container) {
crm_trace("Container ordering for %s", action->uuid);
apply_container_ordering(action);
} else {
crm_trace("Remote ordering for %s", action->uuid);
apply_remote_ordering(action);
}
}
}
/*!
* \internal
* \brief Check whether a node is a failed remote node
*
* \param[in] node Node to check
*
* \return true if \p node is a failed remote node, false otherwise
*/
bool
pcmk__is_failed_remote_node(const pe_node_t *node)
{
return pe__is_remote_node(node) && (node->details->remote_rsc != NULL)
&& (get_remote_node_state(node) == remote_state_failed);
}
/*!
* \internal
* \brief Check whether a given resource corresponds to a given node as guest
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p node is a guest node and \p rsc is its containing
* resource, otherwise false
*/
bool
pcmk__rsc_corresponds_to_guest(const pe_resource_t *rsc, const pe_node_t *node)
{
return (rsc != NULL) && (rsc->fillers != NULL) && (node != NULL)
&& (node->details->remote_rsc != NULL)
&& (node->details->remote_rsc->container == rsc);
}
/*!
* \internal
* \brief Get proper connection host that a remote action must be routed through
*
* A remote connection resource might be starting, stopping, or migrating in the
* same transition that an action needs to be executed on its Pacemaker Remote
* node. Determine the proper node that the remote action should be routed
* through.
*
* \param[in] action (Potentially remote) action to route
*
* \return Connection host that action should be routed through if remote,
* otherwise NULL
*/
pe_node_t *
pcmk__connection_host_for_action(const pe_action_t *action)
{
pe_node_t *began_on = NULL;
pe_node_t *ended_on = NULL;
bool partial_migration = false;
const char *task = action->task;
if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_none)
|| !pe__is_guest_or_remote_node(action->node)) {
return NULL;
}
CRM_ASSERT(action->node->details->remote_rsc != NULL);
began_on = pe__current_node(action->node->details->remote_rsc);
ended_on = action->node->details->remote_rsc->allocated_to;
if (action->node->details->remote_rsc
&& (action->node->details->remote_rsc->container == NULL)
&& action->node->details->remote_rsc->partial_migration_target) {
partial_migration = true;
}
if (began_on == NULL) {
crm_trace("Routing %s for %s through remote connection's "
"next node %s (starting)%s",
action->task, (action->rsc? action->rsc->id : "no resource"),
(ended_on? ended_on->details->uname : "none"),
partial_migration? " (partial migration)" : "");
return ended_on;
}
if (ended_on == NULL) {
crm_trace("Routing %s for %s through remote connection's "
"current node %s (stopping)%s",
action->task, (action->rsc? action->rsc->id : "no resource"),
(began_on? began_on->details->uname : "none"),
partial_migration? " (partial migration)" : "");
return began_on;
}
if (pe__same_node(began_on, ended_on)) {
crm_trace("Routing %s for %s through remote connection's "
"current node %s (not moving)%s",
action->task, (action->rsc? action->rsc->id : "no resource"),
(began_on? began_on->details->uname : "none"),
partial_migration? " (partial migration)" : "");
return began_on;
}
/* If we get here, the remote connection is moving during this transition.
* This means some actions for resources behind the connection will get
* routed through the cluster node the connection resource is currently on,
* and others are routed through the cluster node the connection will end up
* on.
*/
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
task = g_hash_table_lookup(action->meta, "notify_operation");
}
/*
* Stop, demote, and migration actions must occur before the connection can
* move (these actions are required before the remote resource can stop). In
* this case, we know these actions have to be routed through the initial
* cluster node the connection resource lived on before the move takes
* place.
*
* The exception is a partial migration of a (non-guest) remote connection
* resource; in that case, all actions (even these) will be ordered after
* the connection's pseudo-start on the migration target, so the target is
* the router node.
*/
if (pcmk__strcase_any_of(task, PCMK_ACTION_CANCEL, PCMK_ACTION_STOP,
PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_MIGRATE_TO, NULL)
&& !partial_migration) {
crm_trace("Routing %s for %s through remote connection's "
"current node %s (moving)%s",
action->task, (action->rsc? action->rsc->id : "no resource"),
(began_on? began_on->details->uname : "none"),
partial_migration? " (partial migration)" : "");
return began_on;
}
/* Everything else (start, promote, monitor, probe, refresh,
* clear failcount, delete, ...) must occur after the connection starts on
* the node it is moving to.
*/
crm_trace("Routing %s for %s through remote connection's "
"next node %s (moving)%s",
action->task, (action->rsc? action->rsc->id : "no resource"),
(ended_on? ended_on->details->uname : "none"),
partial_migration? " (partial migration)" : "");
return ended_on;
}
/*!
* \internal
* \brief Replace remote connection's addr="#uname" with actual address
*
* REMOTE_CONTAINER_HACK: If a given resource is a remote connection resource
* with its "addr" parameter set to "#uname", pull the actual value from the
* parameters evaluated without a node (which was put there earlier in
* pcmk__create_graph() when the bundle's expand() method was called).
*
* \param[in,out] rsc Resource to check
* \param[in,out] params Resource parameters evaluated per node
*/
void
pcmk__substitute_remote_addr(pe_resource_t *rsc, GHashTable *params)
{
const char *remote_addr = g_hash_table_lookup(params,
XML_RSC_ATTR_REMOTE_RA_ADDR);
if (pcmk__str_eq(remote_addr, "#uname", pcmk__str_none)) {
GHashTable *base = pe_rsc_params(rsc, NULL, rsc->cluster);
remote_addr = g_hash_table_lookup(base, XML_RSC_ATTR_REMOTE_RA_ADDR);
if (remote_addr != NULL) {
g_hash_table_insert(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR),
strdup(remote_addr));
}
}
}
/*!
* \brief Add special bundle meta-attributes to XML
*
* If a given action will be executed on a guest node (including a bundle),
* add the special bundle meta-attribute "container-attribute-target" and
* environment variable "physical_host" as XML attributes (using meta-attribute
* naming).
*
* \param[in,out] args_xml XML to add attributes to
* \param[in] action Action to check
*/
void
pcmk__add_bundle_meta_to_xml(xmlNode *args_xml, const pe_action_t *action)
{
const pe_node_t *guest = action->node;
const pe_node_t *host = NULL;
enum action_tasks task;
if (!pe__is_guest_node(guest)) {
return;
}
task = text2task(action->task);
if ((task == pcmk_action_notify) || (task == pcmk_action_notified)) {
task = text2task(g_hash_table_lookup(action->meta, "notify_operation"));
}
switch (task) {
case pcmk_action_stop:
case pcmk_action_stopped:
case pcmk_action_demote:
case pcmk_action_demoted:
// "Down" actions take place on guest's current host
host = pe__current_node(guest->details->remote_rsc->container);
break;
case pcmk_action_start:
case pcmk_action_started:
case pcmk_action_monitor:
case pcmk_action_promote:
case pcmk_action_promoted:
// "Up" actions take place on guest's next host
host = guest->details->remote_rsc->container->allocated_to;
break;
default:
break;
}
if (host != NULL) {
hash2metafield((gpointer) XML_RSC_ATTR_TARGET,
(gpointer) g_hash_table_lookup(action->rsc->meta,
XML_RSC_ATTR_TARGET),
(gpointer) args_xml);
hash2metafield((gpointer) PCMK__ENV_PHYSICAL_HOST,
(gpointer) host->details->uname,
(gpointer) args_xml);
}
}
diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c
index e731126de0..6e2618fe7b 100644
--- a/lib/pacemaker/pcmk_scheduler.c
+++ b/lib/pacemaker/pcmk_scheduler.c
@@ -1,816 +1,816 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/scheduler_internal.h>
#include <glib.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
CRM_TRACE_INIT_DATA(pacemaker);
/*!
* \internal
* \brief Do deferred action checks after assignment
*
* When unpacking the resource history, the scheduler checks for resource
* configurations that have changed since an action was run. However, at that
* time, bundles using the REMOTE_CONTAINER_HACK don't have their final
* parameter information, so instead they add a deferred check to a list. This
* function processes one entry in that list.
*
* \param[in,out] rsc Resource that action history is for
* \param[in,out] node Node that action history is for
* \param[in] rsc_op Action history entry
* \param[in] check Type of deferred check to do
*/
static void
check_params(pe_resource_t *rsc, pe_node_t *node, const xmlNode *rsc_op,
enum pcmk__check_parameters check)
{
const char *reason = NULL;
op_digest_cache_t *digest_data = NULL;
switch (check) {
case pcmk__check_active:
if (pcmk__check_action_config(rsc, node, rsc_op)
&& pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL)) {
reason = "action definition changed";
}
break;
case pcmk__check_last_failure:
digest_data = rsc_action_digest_cmp(rsc, rsc_op, node,
rsc->cluster);
switch (digest_data->rc) {
case RSC_DIGEST_UNKNOWN:
crm_trace("Resource %s history entry %s on %s has "
"no digest to compare",
rsc->id, ID(rsc_op), node->details->id);
break;
case RSC_DIGEST_MATCH:
break;
default:
reason = "resource parameters have changed";
break;
}
break;
}
if (reason != NULL) {
pe__clear_failcount(rsc, node, reason, rsc->cluster);
}
}
/*!
* \internal
* \brief Check whether a resource has failcount clearing scheduled on a node
*
* \param[in] node Node to check
* \param[in] rsc Resource to check
*
* \return true if \p rsc has failcount clearing scheduled on \p node,
* otherwise false
*/
static bool
failcount_clear_action_exists(const pe_node_t *node, const pe_resource_t *rsc)
{
GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT,
TRUE);
if (list != NULL) {
g_list_free(list);
return true;
}
return false;
}
/*!
* \internal
* \brief Ban a resource from a node if it reached its failure threshold there
*
* \param[in,out] data Resource to check failure threshold for
* \param[in] user_data Node to check resource on
*/
static void
check_failure_threshold(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
// If this is a collective resource, apply recursively to children instead
if (rsc->children != NULL) {
g_list_foreach(rsc->children, check_failure_threshold, user_data);
return;
}
if (!failcount_clear_action_exists(node, rsc)) {
/* Don't force the resource away from this node due to a failcount
* that's going to be cleared.
*
* @TODO Failcount clearing can be scheduled in
* pcmk__handle_rsc_config_changes() via process_rsc_history(), or in
* schedule_resource_actions() via check_params(). This runs well before
* then, so it cannot detect those, meaning we might check the migration
* threshold when we shouldn't. Worst case, we stop or move the
* resource, then move it back in the next transition.
*/
pe_resource_t *failed = NULL;
if (pcmk__threshold_reached(rsc, node, &failed)) {
resource_location(failed, node, -INFINITY, "__fail_limit__",
rsc->cluster);
}
}
}
/*!
* \internal
* \brief If resource has exclusive discovery, ban node if not allowed
*
* Location constraints have a resource-discovery option that allows users to
* specify where probes are done for the affected resource. If this is set to
* exclusive, probes will only be done on nodes listed in exclusive constraints.
* This function bans the resource from the node if the node is not listed.
*
* \param[in,out] data Resource to check
* \param[in] user_data Node to check resource on
*/
static void
apply_exclusive_discovery(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
const pe_node_t *node = user_data;
if (rsc->exclusive_discover
|| pe__const_top_resource(rsc, false)->exclusive_discover) {
pe_node_t *match = NULL;
// If this is a collective resource, apply recursively to children
g_list_foreach(rsc->children, apply_exclusive_discovery, user_data);
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if ((match != NULL)
&& (match->rsc_discover_mode != pe_discover_exclusive)) {
match->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Apply stickiness to a resource if appropriate
*
* \param[in,out] data Resource to check for stickiness
* \param[in] user_data Ignored
*/
static void
apply_stickiness(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
pe_node_t *node = NULL;
// If this is a collective resource, apply recursively to children instead
if (rsc->children != NULL) {
g_list_foreach(rsc->children, apply_stickiness, NULL);
return;
}
/* A resource is sticky if it is managed, has stickiness configured, and is
* active on a single node.
*/
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)
|| (rsc->stickiness < 1) || !pcmk__list_of_1(rsc->running_on)) {
return;
}
node = rsc->running_on->data;
/* In a symmetric cluster, stickiness can always be used. In an
* asymmetric cluster, we have to check whether the resource is still
* allowed on the node, so we don't keep the resource somewhere it is no
* longer explicitly enabled.
*/
if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_symmetric_cluster)
&& (g_hash_table_lookup(rsc->allowed_nodes,
node->details->id) == NULL)) {
pe_rsc_debug(rsc,
"Ignoring %s stickiness because the cluster is "
"asymmetric and %s is not explicitly allowed",
rsc->id, pe__node_name(node));
return;
}
pe_rsc_debug(rsc, "Resource %s has %d stickiness on %s",
rsc->id, rsc->stickiness, pe__node_name(node));
resource_location(rsc, node, rsc->stickiness, "stickiness", rsc->cluster);
}
/*!
* \internal
* \brief Apply shutdown locks for all resources as appropriate
*
* \param[in,out] data_set Cluster working set
*/
static void
apply_shutdown_locks(pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, pcmk_sched_shutdown_lock)) {
return;
}
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->shutdown_lock(rsc);
}
}
/*!
* \internal
* \brief Calculate the number of available nodes in the cluster
*
* \param[in,out] data_set Cluster working set
*/
static void
count_available_nodes(pe_working_set_t *data_set)
{
if (pcmk_is_set(data_set->flags, pcmk_sched_no_compat)) {
return;
}
// @COMPAT for API backward compatibility only (cluster does not use value)
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
if ((node != NULL) && (node->weight >= 0) && node->details->online
&& (node->details->type != node_ping)) {
data_set->max_valid_nodes++;
}
}
crm_trace("Online node count: %d", data_set->max_valid_nodes);
}
/*
* \internal
* \brief Apply node-specific scheduling criteria
*
* After the CIB has been unpacked, process node-specific scheduling criteria
* including shutdown locks, location constraints, resource stickiness,
* migration thresholds, and exclusive resource discovery.
*/
static void
apply_node_criteria(pe_working_set_t *data_set)
{
crm_trace("Applying node-specific scheduling criteria");
apply_shutdown_locks(data_set);
count_available_nodes(data_set);
pcmk__apply_locations(data_set);
g_list_foreach(data_set->resources, apply_stickiness, NULL);
for (GList *node_iter = data_set->nodes; node_iter != NULL;
node_iter = node_iter->next) {
for (GList *rsc_iter = data_set->resources; rsc_iter != NULL;
rsc_iter = rsc_iter->next) {
check_failure_threshold(rsc_iter->data, node_iter->data);
apply_exclusive_discovery(rsc_iter->data, node_iter->data);
}
}
}
/*!
* \internal
* \brief Assign resources to nodes
*
* \param[in,out] data_set Cluster working set
*/
static void
assign_resources(pe_working_set_t *data_set)
{
GList *iter = NULL;
crm_trace("Assigning resources to nodes");
if (!pcmk__str_eq(data_set->placement_strategy, "default",
pcmk__str_casei)) {
pcmk__sort_resources(data_set);
}
pcmk__show_node_capacities("Original", data_set);
if (pcmk_is_set(data_set->flags, 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 = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (rsc->is_remote_node) {
pe_rsc_trace(rsc, "Assigning remote connection resource '%s'",
rsc->id);
rsc->cmds->assign(rsc, rsc->partial_migration_target, true);
}
}
}
/* now do the rest of the resources */
for (iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
if (!rsc->is_remote_node) {
pe_rsc_trace(rsc, "Assigning %s resource '%s'",
rsc->xml->name, rsc->id);
rsc->cmds->assign(rsc, NULL, true);
}
}
pcmk__show_node_capacities("Remaining", data_set);
}
/*!
* \internal
* \brief Schedule fail count clearing on online nodes if resource is orphaned
*
* \param[in,out] data Resource to check
* \param[in] user_data Ignored
*/
static void
clear_failcounts_if_orphaned(gpointer data, gpointer user_data)
{
pe_resource_t *rsc = data;
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
return;
}
crm_trace("Clear fail counts for orphaned resource %s", rsc->id);
/* There's no need to recurse into rsc->children because those
* should just be unassigned clone instances.
*/
for (GList *iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
pe_action_t *clear_op = NULL;
if (!node->details->online) {
continue;
}
if (pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL) == 0) {
continue;
}
clear_op = pe__clear_failcount(rsc, node, "it is orphaned",
rsc->cluster);
/* We can't use order_action_then_stop() here because its
* pe_order_preserve breaks things
*/
pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc),
NULL, pe_order_optional, rsc->cluster);
}
}
/*!
* \internal
* \brief Schedule any resource actions needed
*
* \param[in,out] data_set Cluster working set
*/
static void
schedule_resource_actions(pe_working_set_t *data_set)
{
// Process deferred action checks
pe__foreach_param_check(data_set, check_params);
pe__free_param_checks(data_set);
if (pcmk_is_set(data_set->flags, pcmk_sched_probe_resources)) {
crm_trace("Scheduling probes");
pcmk__schedule_probes(data_set);
}
if (pcmk_is_set(data_set->flags, pcmk_sched_stop_removed_resources)) {
g_list_foreach(data_set->resources, clear_failcounts_if_orphaned, NULL);
}
crm_trace("Scheduling resource actions");
for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) {
pe_resource_t *rsc = (pe_resource_t *) iter->data;
rsc->cmds->create_actions(rsc);
}
}
/*!
* \internal
* \brief Check whether a resource or any of its descendants are managed
*
* \param[in] rsc Resource to check
*
* \return true if resource or any descendant is managed, otherwise false
*/
static bool
is_managed(const pe_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
return true;
}
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
if (is_managed((pe_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether any resources in the cluster are managed
*
* \param[in] data_set Cluster working set
*
* \return true if any resource is managed, otherwise false
*/
static bool
any_managed_resources(const pe_working_set_t *data_set)
{
for (const GList *iter = data_set->resources;
iter != NULL; iter = iter->next) {
if (is_managed((const pe_resource_t *) iter->data)) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a node requires fencing
*
* \param[in] node Node to check
* \param[in] have_managed Whether any resource in cluster is managed
*
* \return true if \p node should be fenced, otherwise false
*/
static bool
needs_fencing(const pe_node_t *node, bool have_managed)
{
return have_managed && node->details->unclean
&& pe_can_fence(node->details->data_set, node);
}
/*!
* \internal
* \brief Check whether a node requires shutdown
*
* \param[in] node Node to check
*
* \return true if \p node should be shut down, otherwise false
*/
static bool
needs_shutdown(const pe_node_t *node)
{
if (pe__is_guest_or_remote_node(node)) {
/* Do not send shutdown actions for Pacemaker Remote nodes.
* @TODO We might come up with a good use for this in the future.
*/
return false;
}
return node->details->online && node->details->shutdown;
}
/*!
* \internal
* \brief Track and order non-DC fencing
*
* \param[in,out] list List of existing non-DC fencing actions
* \param[in,out] action Fencing action to prepend to \p list
* \param[in] data_set Cluster working set
*
* \return (Possibly new) head of \p list
*/
static GList *
add_nondc_fencing(GList *list, pe_action_t *action,
const pe_working_set_t *data_set)
{
if (!pcmk_is_set(data_set->flags, 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((pe_action_t *) list->data, action, pe_order_optional);
}
return g_list_prepend(list, action);
}
/*!
* \internal
* \brief Schedule a node for fencing
*
* \param[in,out] node Node that requires fencing
*/
static pe_action_t *
schedule_fencing(pe_node_t *node)
{
pe_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean",
FALSE, node->details->data_set);
pe_warn("Scheduling node %s for fencing", pe__node_name(node));
pcmk__order_vs_fence(fencing, node->details->data_set);
return fencing;
}
/*!
* \internal
* \brief Create and order node fencing and shutdown actions
*
* \param[in,out] data_set Cluster working set
*/
static void
schedule_fencing_and_shutdowns(pe_working_set_t *data_set)
{
pe_action_t *dc_down = NULL;
bool integrity_lost = false;
bool have_managed = any_managed_resources(data_set);
GList *fencing_ops = NULL;
GList *shutdown_ops = NULL;
crm_trace("Scheduling fencing and shutdowns as needed");
if (!have_managed) {
crm_notice("No fencing will be done until there are resources "
"to manage");
}
// Check each node for whether it needs fencing or shutdown
for (GList *iter = data_set->nodes; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
pe_action_t *fencing = NULL;
/* Guest nodes are "fenced" by recovering their container resource,
* so handle them separately.
*/
if (pe__is_guest_node(node)) {
if (node->details->remote_requires_reset && have_managed
&& pe_can_fence(data_set, node)) {
pcmk__fence_guest(node);
}
continue;
}
if (needs_fencing(node, have_managed)) {
fencing = schedule_fencing(node);
// Track DC and non-DC fence actions separately
if (node->details->is_dc) {
dc_down = fencing;
} else {
fencing_ops = add_nondc_fencing(fencing_ops, fencing, data_set);
}
} else if (needs_shutdown(node)) {
pe_action_t *down_op = pcmk__new_shutdown_action(node);
// Track DC and non-DC shutdown actions separately
if (node->details->is_dc) {
dc_down = down_op;
} else {
shutdown_ops = g_list_prepend(shutdown_ops, down_op);
}
}
if ((fencing == NULL) && node->details->unclean) {
integrity_lost = true;
pe_warn("Node %s is unclean but cannot be fenced",
pe__node_name(node));
}
}
if (integrity_lost) {
if (!pcmk_is_set(data_set->flags, pcmk_sched_fencing_enabled)) {
pe_warn("Resource functionality and data integrity cannot be "
"guaranteed (configure, enable, and test fencing to "
"correct this)");
} else if (!pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
crm_notice("Unclean nodes will not be fenced until quorum is "
"attained or no-quorum-policy is set to ignore");
}
}
if (dc_down != NULL) {
/* Order any non-DC shutdowns before any DC shutdown, to avoid repeated
* DC elections. However, we don't want to order non-DC shutdowns before
* a DC *fencing*, because even though we don't want a node that's
* shutting down to become DC, the DC fencing could be ordered before a
* clone stop that's also ordered before the shutdowns, thus leading to
* a graph loop.
*/
if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
pcmk__order_after_each(dc_down, shutdown_ops);
}
// Order any non-DC fencing before any DC fencing or shutdown
if (pcmk_is_set(data_set->flags, 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((pe_action_t *) fencing_ops->data, dc_down,
pe_order_optional);
}
}
g_list_free(fencing_ops);
g_list_free(shutdown_ops);
}
static void
log_resource_details(pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
GList *all = NULL;
/* Due to the `crm_mon --node=` feature, out->message() for all the
* resource-related messages expects a list of nodes that we are allowed to
* output information for. Here, we create a wildcard to match all nodes.
*/
all = g_list_prepend(all, (gpointer) "*");
for (GList *item = data_set->resources; item != NULL; item = item->next) {
pe_resource_t *rsc = (pe_resource_t *) item->data;
// Log all resources except inactive orphans
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)
|| (rsc->role != pcmk_role_stopped)) {
out->message(out, crm_map_element_name(rsc->xml), 0, rsc, all, all);
}
}
g_list_free(all);
}
static void
log_all_actions(pe_working_set_t *data_set)
{
/* This only ever outputs to the log, so ignore whatever output object was
* previously set and just log instead.
*/
pcmk__output_t *prev_out = data_set->priv;
pcmk__output_t *out = NULL;
if (pcmk__log_output_new(&out) != pcmk_rc_ok) {
return;
}
pe__register_messages(out);
pcmk__register_lib_messages(out);
pcmk__output_set_log_level(out, LOG_NOTICE);
data_set->priv = out;
out->begin_list(out, NULL, NULL, "Actions");
pcmk__output_actions(data_set);
out->end_list(out);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
data_set->priv = prev_out;
}
/*!
* \internal
* \brief Log all required but unrunnable actions at trace level
*
* \param[in] data_set Cluster working set
*/
static void
log_unrunnable_actions(const pe_working_set_t *data_set)
{
const uint64_t flags = pe_action_optional
|pe_action_runnable
- |pe_action_pseudo;
+ |pcmk_action_pseudo;
crm_trace("Required but unrunnable actions:");
for (const GList *iter = data_set->actions;
iter != NULL; iter = iter->next) {
const pe_action_t *action = (const pe_action_t *) iter->data;
if (!pcmk_any_flags_set(action->flags, flags)) {
pcmk__log_action("\t", action, true);
}
}
}
/*!
* \internal
* \brief Unpack the CIB for scheduling
*
* \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked)
* \param[in] flags Working set flags to set in addition to defaults
* \param[in,out] data_set Cluster working set
*/
static void
unpack_cib(xmlNode *cib, unsigned long long flags, pe_working_set_t *data_set)
{
const char* localhost_save = NULL;
if (pcmk_is_set(data_set->flags, pcmk_sched_have_status)) {
crm_trace("Reusing previously calculated cluster status");
pe__set_working_set_flags(data_set, flags);
return;
}
if (data_set->localhost) {
localhost_save = data_set->localhost;
}
CRM_ASSERT(cib != NULL);
crm_trace("Calculating cluster status");
/* This will zero the entire struct without freeing anything first, so
* callers should never call pcmk__schedule_actions() with a populated data
* set unless pcmk_sched_have_status is set (i.e. cluster_status() was
* previously called, whether directly or via pcmk__schedule_actions()).
*/
set_working_set_defaults(data_set);
if (localhost_save) {
data_set->localhost = localhost_save;
}
pe__set_working_set_flags(data_set, flags);
data_set->input = cib;
cluster_status(data_set); // Sets 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 Working set flags to set in addition to defaults
* \param[in,out] data_set Cluster working set
*/
void
pcmk__schedule_actions(xmlNode *cib, unsigned long long flags,
pe_working_set_t *data_set)
{
unpack_cib(cib, flags, data_set);
pcmk__set_assignment_methods(data_set);
pcmk__apply_node_health(data_set);
pcmk__unpack_constraints(data_set);
if (pcmk_is_set(data_set->flags, pcmk_sched_validate_only)) {
return;
}
if (!pcmk_is_set(data_set->flags, pcmk_sched_location_only)
&& pcmk__is_daemon) {
log_resource_details(data_set);
}
apply_node_criteria(data_set);
if (pcmk_is_set(data_set->flags, pcmk_sched_location_only)) {
return;
}
pcmk__create_internal_constraints(data_set);
pcmk__handle_rsc_config_changes(data_set);
assign_resources(data_set);
schedule_resource_actions(data_set);
/* Remote ordering constraints need to happen prior to calculating fencing
* because it is one more place we can mark nodes as needing fencing.
*/
pcmk__order_remote_connection_actions(data_set);
schedule_fencing_and_shutdowns(data_set);
pcmk__apply_orderings(data_set);
log_all_actions(data_set);
pcmk__create_graph(data_set);
if (get_crm_log_level() == LOG_TRACE) {
log_unrunnable_actions(data_set);
}
}
diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c
index 60e07993cd..f1c794800d 100644
--- a/lib/pacemaker/pcmk_simulate.c
+++ b/lib/pacemaker/pcmk_simulate.c
@@ -1,1005 +1,1005 @@
/*
* Copyright 2021-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/cib/internal.h>
#include <crm/common/output.h>
#include <crm/common/results.h>
#include <crm/pengine/pe_types.h>
#include <pacemaker-internal.h>
#include <pacemaker.h>
#include <stdint.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include "libpacemaker_private.h"
static pcmk__output_t *out = NULL;
static cib_t *fake_cib = NULL;
static GList *fake_resource_list = NULL;
static const GList *fake_op_fail_list = NULL;
static void set_effective_date(pe_working_set_t *data_set, bool print_original,
const char *use_date);
/*!
* \internal
* \brief Create an action name for use in a dot graph
*
* \param[in] action Action to create name for
* \param[in] verbose If true, add action ID to name
*
* \return Newly allocated string with action name
* \note It is the caller's responsibility to free the result.
*/
static char *
create_action_name(const pe_action_t *action, bool verbose)
{
char *action_name = NULL;
const char *prefix = "";
const char *action_host = NULL;
const char *clone_name = NULL;
const char *task = action->task;
if (action->node != NULL) {
action_host = action->node->details->uname;
- } else if (!pcmk_is_set(action->flags, pe_action_pseudo)) {
+ } else if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) {
action_host = "<none>";
}
if (pcmk__str_eq(action->task, PCMK_ACTION_CANCEL, pcmk__str_none)) {
prefix = "Cancel ";
task = action->cancel_task;
}
if (action->rsc != NULL) {
clone_name = action->rsc->clone_name;
}
if (clone_name != NULL) {
char *key = NULL;
guint interval_ms = 0;
if (pcmk__guint_from_hash(action->meta,
XML_LRM_ATTR_INTERVAL_MS, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY,
PCMK_ACTION_NOTIFIED, NULL)) {
const char *n_type = g_hash_table_lookup(action->meta,
"notify_key_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_key_operation");
CRM_ASSERT(n_type != NULL);
CRM_ASSERT(n_task != NULL);
key = pcmk__notify_key(clone_name, n_type, n_task);
} else {
key = pcmk__op_key(clone_name, task, interval_ms);
}
if (action_host != NULL) {
action_name = crm_strdup_printf("%s%s %s",
prefix, key, action_host);
} else {
action_name = crm_strdup_printf("%s%s", prefix, key);
}
free(key);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta, "stonith_action");
action_name = crm_strdup_printf("%s%s '%s' %s",
prefix, action->task, op, action_host);
} else if (action->rsc && action_host) {
action_name = crm_strdup_printf("%s%s %s",
prefix, action->uuid, action_host);
} else if (action_host) {
action_name = crm_strdup_printf("%s%s %s",
prefix, action->task, action_host);
} else {
action_name = crm_strdup_printf("%s", action->uuid);
}
if (verbose) {
char *with_id = crm_strdup_printf("%s (%d)", action_name, action->id);
free(action_name);
action_name = with_id;
}
return action_name;
}
/*!
* \internal
* \brief Display the status of a cluster
*
* \param[in,out] data_set Cluster working set
* \param[in] show_opts How to modify display (as pcmk_show_opt_e flags)
* \param[in] section_opts Sections to display (as pcmk_section_e flags)
* \param[in] title What to use as list title
* \param[in] print_spacer Whether to display a spacer first
*/
static void
print_cluster_status(pe_working_set_t *data_set, uint32_t show_opts,
uint32_t section_opts, const char *title,
bool print_spacer)
{
pcmk__output_t *out = data_set->priv;
GList *all = NULL;
crm_exit_t stonith_rc = 0;
enum pcmk_pacemakerd_state state = pcmk_pacemakerd_state_invalid;
section_opts |= pcmk_section_nodes | pcmk_section_resources;
show_opts |= pcmk_show_inactive_rscs | pcmk_show_failed_detail;
all = g_list_prepend(all, (gpointer) "*");
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
out->begin_list(out, NULL, NULL, "%s", title);
out->message(out, "cluster-status",
data_set, state, stonith_rc, NULL,
false, section_opts, show_opts, NULL, all, all);
out->end_list(out);
g_list_free(all);
}
/*!
* \internal
* \brief Display a summary of all actions scheduled in a transition
*
* \param[in,out] data_set Cluster working set (fully scheduled)
* \param[in] print_spacer Whether to display a spacer first
*/
static void
print_transition_summary(pe_working_set_t *data_set, bool print_spacer)
{
pcmk__output_t *out = data_set->priv;
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
out->begin_list(out, NULL, NULL, "Transition Summary");
pcmk__output_actions(data_set);
out->end_list(out);
}
/*!
* \internal
* \brief Reset a cluster working set's input, output, date, and flags
*
* \param[in,out] data_set Cluster working set
* \param[in] input What to set as cluster input
* \param[in] out What to set as cluster output object
* \param[in] use_date What to set as cluster's current timestamp
* \param[in] flags Group of enum pcmk_scheduler_flags to set
*/
static void
reset(pe_working_set_t *data_set, xmlNodePtr input, pcmk__output_t *out,
const char *use_date, unsigned int flags)
{
data_set->input = input;
data_set->priv = out;
set_effective_date(data_set, true, use_date);
if (pcmk_is_set(flags, pcmk_sim_sanitized)) {
pe__set_working_set_flags(data_set, pcmk_sched_sanitized);
}
if (pcmk_is_set(flags, pcmk_sim_show_scores)) {
pe__set_working_set_flags(data_set, pcmk_sched_output_scores);
}
if (pcmk_is_set(flags, pcmk_sim_show_utilization)) {
pe__set_working_set_flags(data_set, pcmk_sched_show_utilization);
}
}
/*!
* \brief Write out a file in dot(1) format describing the actions that will
* be taken by the scheduler in response to an input CIB file.
*
* \param[in,out] data_set Working set for the cluster
* \param[in] dot_file The filename to write
* \param[in] all_actions Write all actions, even those that are optional
* or are on unmanaged resources
* \param[in] verbose Add extra information, such as action IDs, to the
* output
*
* \return Standard Pacemaker return code
*/
static int
write_sim_dotfile(pe_working_set_t *data_set, const char *dot_file,
bool all_actions, bool verbose)
{
GList *iter = NULL;
FILE *dot_strm = fopen(dot_file, "w");
if (dot_strm == NULL) {
return errno;
}
fprintf(dot_strm, " digraph \"g\" {\n");
for (iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = (pe_action_t *) iter->data;
const char *style = "dashed";
const char *font = "black";
const char *color = "black";
char *action_name = create_action_name(action, verbose);
- if (pcmk_is_set(action->flags, pe_action_pseudo)) {
+ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
font = "orange";
}
if (pcmk_is_set(action->flags, pe_action_dumped)) {
style = "bold";
color = "green";
} else if ((action->rsc != NULL)
&& !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)) {
color = "red";
font = "purple";
if (!all_actions) {
goto do_not_write;
}
} else if (pcmk_is_set(action->flags, pe_action_optional)) {
color = "blue";
if (!all_actions) {
goto do_not_write;
}
} else {
color = "red";
CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pe_action_runnable));
}
pe__set_action_flags(action, pe_action_dumped);
fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n",
action_name, style, color, font);
do_not_write:
free(action_name);
}
for (iter = data_set->actions; iter != NULL; iter = iter->next) {
pe_action_t *action = (pe_action_t *) iter->data;
for (GList *before_iter = action->actions_before;
before_iter != NULL; before_iter = before_iter->next) {
pe_action_wrapper_t *before = before_iter->data;
char *before_name = NULL;
char *after_name = NULL;
const char *style = "dashed";
bool optional = true;
if (before->state == pe_link_dumped) {
optional = false;
style = "bold";
} else if (before->type == pe_order_none) {
continue;
} else if (pcmk_is_set(before->action->flags, pe_action_dumped)
&& pcmk_is_set(action->flags, pe_action_dumped)
&& before->type != pe_order_load) {
optional = false;
}
if (all_actions || !optional) {
before_name = create_action_name(before->action, verbose);
after_name = create_action_name(action, verbose);
fprintf(dot_strm, "\"%s\" -> \"%s\" [ style = %s]\n",
before_name, after_name, style);
free(before_name);
free(after_name);
}
}
}
fprintf(dot_strm, "}\n");
fflush(dot_strm);
fclose(dot_strm);
return pcmk_rc_ok;
}
/*!
* \brief Profile the configuration updates and scheduler actions in a single
* CIB file, printing the profiling timings.
*
* \note \p data_set->priv must have been set to a valid \p pcmk__output_t
* object before this function is called.
*
* \param[in] xml_file The CIB file to profile
* \param[in] repeat Number of times to run
* \param[in,out] data_set Working set for the cluster
* \param[in] use_date The date to set the cluster's time to (may be NULL)
*/
static void
profile_file(const char *xml_file, long long repeat, pe_working_set_t *data_set,
const char *use_date)
{
pcmk__output_t *out = data_set->priv;
xmlNode *cib_object = NULL;
clock_t start = 0;
clock_t end;
unsigned long long data_set_flags = pcmk_sched_no_compat;
CRM_ASSERT(out != NULL);
cib_object = filename2xml(xml_file);
start = clock();
if (pcmk_find_cib_element(cib_object, XML_CIB_TAG_STATUS) == NULL) {
create_xml_node(cib_object, XML_CIB_TAG_STATUS);
}
if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) {
free_xml(cib_object);
return;
}
if (validate_xml(cib_object, NULL, FALSE) != TRUE) {
free_xml(cib_object);
return;
}
if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) {
data_set_flags |= pcmk_sched_output_scores;
}
if (pcmk_is_set(data_set->flags, pcmk_sched_show_utilization)) {
data_set_flags |= pcmk_sched_show_utilization;
}
for (int i = 0; i < repeat; ++i) {
xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object);
data_set->input = input;
set_effective_date(data_set, false, use_date);
pcmk__schedule_actions(input, data_set_flags, data_set);
pe_reset_working_set(data_set);
}
end = clock();
out->message(out, "profile", xml_file, start, end);
}
void
pcmk__profile_dir(const char *dir, long long repeat, pe_working_set_t *data_set,
const char *use_date)
{
pcmk__output_t *out = data_set->priv;
struct dirent **namelist;
int file_num = scandir(dir, &namelist, 0, alphasort);
CRM_ASSERT(out != NULL);
if (file_num > 0) {
struct stat prop;
char buffer[FILENAME_MAX];
out->begin_list(out, NULL, NULL, "Timings");
while (file_num--) {
if ('.' == namelist[file_num]->d_name[0]) {
free(namelist[file_num]);
continue;
} else if (!pcmk__ends_with_ext(namelist[file_num]->d_name,
".xml")) {
free(namelist[file_num]);
continue;
}
snprintf(buffer, sizeof(buffer), "%s/%s",
dir, namelist[file_num]->d_name);
if (stat(buffer, &prop) == 0 && S_ISREG(prop.st_mode)) {
profile_file(buffer, repeat, data_set, use_date);
}
free(namelist[file_num]);
}
free(namelist);
out->end_list(out);
}
}
/*!
* \brief Set the date of the cluster, either to the value given by
* \p use_date, or to the "execution-date" value in the CIB.
*
* \note \p data_set->priv must have been set to a valid \p pcmk__output_t
* object before this function is called.
*
* \param[in,out] data_set Working set for the cluster
* \param[in] print_original If \p true, the "execution-date" should
* also be printed
* \param[in] use_date The date to set the cluster's time to
* (may be NULL)
*/
static void
set_effective_date(pe_working_set_t *data_set, bool print_original,
const char *use_date)
{
pcmk__output_t *out = data_set->priv;
time_t original_date = 0;
CRM_ASSERT(out != NULL);
crm_element_value_epoch(data_set->input, "execution-date", &original_date);
if (use_date) {
data_set->now = crm_time_new(use_date);
out->info(out, "Setting effective cluster time: %s", use_date);
crm_time_log(LOG_NOTICE, "Pretending 'now' is", data_set->now,
crm_time_log_date | crm_time_log_timeofday);
} else if (original_date != 0) {
data_set->now = pcmk__copy_timet(original_date);
if (print_original) {
char *when = crm_time_as_string(data_set->now,
crm_time_log_date|crm_time_log_timeofday);
out->info(out, "Using the original execution date of: %s", when);
free(when);
}
}
}
/*!
* \internal
* \brief Simulate successfully executing a pseudo-action in a graph
*
* \param[in,out] graph Graph to update with pseudo-action result
* \param[in,out] action Pseudo-action to simulate executing
*
* \return Standard Pacemaker return code
*/
static int
simulate_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
out->message(out, "inject-pseudo-action", node, task);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate executing a resource action in a graph
*
* \param[in,out] graph Graph to update with resource action result
* \param[in,out] action Resource action to simulate executing
*
* \return Standard Pacemaker return code
*/
static int
simulate_resource_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
int rc;
lrmd_event_data_t *op = NULL;
int target_outcome = PCMK_OCF_OK;
const char *rtype = NULL;
const char *rclass = NULL;
const char *resource = NULL;
const char *rprovider = NULL;
const char *resource_config_name = NULL;
const char *operation = crm_element_value(action->xml, "operation");
const char *target_rc_s = crm_meta_value(action->params,
XML_ATTR_TE_TARGET_RC);
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET);
char *uuid = NULL;
const char *router_node = crm_element_value(action->xml,
XML_LRM_ATTR_ROUTER_NODE);
// Certain actions don't need to be displayed or history entries
if (pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) {
crm_debug("No history injection for %s op on %s", operation, node);
goto done; // Confirm action and update graph
}
if (action_rsc == NULL) { // Shouldn't be possible
crm_log_xml_err(action->xml, "Bad");
free(node);
return EPROTO;
}
/* A resource might be known by different names in the configuration and in
* the action (for example, a clone instance). Grab the configuration name
* (which is preferred when writing history), and if necessary, the instance
* name.
*/
resource_config_name = crm_element_value(action_rsc, XML_ATTR_ID);
if (resource_config_name == NULL) { // Shouldn't be possible
crm_log_xml_err(action->xml, "No ID");
free(node);
return EPROTO;
}
resource = resource_config_name;
if (pe_find_resource(fake_resource_list, resource) == NULL) {
const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG);
if ((longname != NULL)
&& (pe_find_resource(fake_resource_list, longname) != NULL)) {
resource = longname;
}
}
// Certain actions need to be displayed but don't need history entries
if (pcmk__strcase_any_of(operation, PCMK_ACTION_DELETE,
PCMK_ACTION_META_DATA, NULL)) {
out->message(out, "inject-rsc-action", resource, operation, node,
(guint) 0);
goto done; // Confirm action and update graph
}
rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS);
rtype = crm_element_value(action_rsc, XML_ATTR_TYPE);
rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER);
pcmk__scan_min_int(target_rc_s, &target_outcome, 0);
CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL,
cib_sync_call|cib_scope_local) == pcmk_ok);
// Ensure the action node is in the CIB
uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID);
cib_node = pcmk__inject_node(fake_cib, node,
((router_node == NULL)? uuid: node));
free(uuid);
CRM_ASSERT(cib_node != NULL);
// Add a history entry for the action
cib_resource = pcmk__inject_resource_history(out, cib_node, resource,
resource_config_name,
rclass, rtype, rprovider);
if (cib_resource == NULL) {
crm_err("Could not simulate action %d history for resource %s",
action->id, resource);
free(node);
free_xml(cib_node);
return EINVAL;
}
// Simulate and display an executor event for the action result
op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE,
target_outcome, "User-injected result");
out->message(out, "inject-rsc-action", resource, op->op_type, node,
op->interval_ms);
// Check whether action is in a list of desired simulated failures
for (const GList *iter = fake_op_fail_list;
iter != NULL; iter = iter->next) {
const char *spec = (const char *) iter->data;
char *key = NULL;
const char *match_name = NULL;
// Allow user to specify anonymous clone with or without instance number
key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type,
op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = resource;
}
free(key);
// If not found, try the resource's name in the configuration
if ((match_name == NULL)
&& (strcmp(resource, resource_config_name) != 0)) {
key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource_config_name,
op->op_type, op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = resource_config_name;
}
free(key);
}
if (match_name == NULL) {
continue; // This failed action entry doesn't match
}
// ${match_name}_${task}_${interval_in_ms}@${node}=${rc}
rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc);
if (rc != 1) {
out->err(out, "Invalid failed operation '%s' "
"(result code must be integer)", spec);
continue; // Keep checking other list entries
}
out->info(out, "Pretending action %d failed with rc=%d",
action->id, op->rc);
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
graph->abort_priority = INFINITY;
pcmk__inject_failcount(out, cib_node, match_name, op->op_type,
op->interval_ms, op->rc);
break;
}
pcmk__inject_action_result(cib_resource, op, target_outcome);
lrmd_free_event(op);
rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
done:
free(node);
free_xml(cib_node);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate successfully executing a cluster action
*
* \param[in,out] graph Graph to update with action result
* \param[in,out] action Cluster action to simulate
*
* \return Standard Pacemaker return code
*/
static int
simulate_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
out->message(out, "inject-cluster-action", node, task, rsc);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate successfully executing a fencing action
*
* \param[in,out] graph Graph to update with action result
* \param[in,out] action Fencing action to simulate
*
* \return Standard Pacemaker return code
*/
static int
simulate_fencing_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *op = crm_meta_value(action->params, "stonith_action");
char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET);
out->message(out, "inject-fencing-action", target, op);
if (!pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) {
int rc = pcmk_ok;
GString *xpath = g_string_sized_new(512);
// Set node state to offline
xmlNode *cib_node = pcmk__inject_node_state_change(fake_cib, target,
false);
CRM_ASSERT(cib_node != NULL);
crm_xml_add(cib_node, XML_ATTR_ORIGIN, __func__);
rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
// Simulate controller clearing node's resource history and attributes
pcmk__g_strcat(xpath,
"//" XML_CIB_TAG_STATE
"[@" XML_ATTR_UNAME "='", target, "']/" XML_CIB_TAG_LRM,
NULL);
fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
g_string_truncate(xpath, 0);
pcmk__g_strcat(xpath,
"//" XML_CIB_TAG_STATE
"[@" XML_ATTR_UNAME "='", target, "']"
"/" XML_TAG_TRANSIENT_NODEATTRS, NULL);
fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
free_xml(cib_node);
g_string_free(xpath, TRUE);
}
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
pcmk__update_graph(graph, action);
free(target);
return pcmk_rc_ok;
}
enum pcmk__graph_status
pcmk__simulate_transition(pe_working_set_t *data_set, cib_t *cib,
const GList *op_fail_list)
{
pcmk__graph_t *transition = NULL;
enum pcmk__graph_status graph_rc;
pcmk__graph_functions_t simulation_fns = {
simulate_pseudo_action,
simulate_resource_action,
simulate_cluster_action,
simulate_fencing_action,
};
out = data_set->priv;
fake_cib = cib;
fake_op_fail_list = op_fail_list;
if (!out->is_quiet(out)) {
out->begin_list(out, NULL, NULL, "Executing Cluster Transition");
}
pcmk__set_graph_functions(&simulation_fns);
transition = pcmk__unpack_graph(data_set->graph, crm_system_name);
pcmk__log_graph(LOG_DEBUG, transition);
fake_resource_list = data_set->resources;
do {
graph_rc = pcmk__execute_graph(transition);
} while (graph_rc == pcmk__graph_active);
fake_resource_list = NULL;
if (graph_rc != pcmk__graph_complete) {
out->err(out, "Transition failed: %s",
pcmk__graph_status2text(graph_rc));
pcmk__log_graph(LOG_ERR, transition);
out->err(out, "An invalid transition was produced");
}
pcmk__free_graph(transition);
if (!out->is_quiet(out)) {
// If not quiet, we'll need the resulting CIB for later display
xmlNode *cib_object = NULL;
int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
pe_reset_working_set(data_set);
data_set->input = cib_object;
out->end_list(out);
}
return graph_rc;
}
int
pcmk__simulate(pe_working_set_t *data_set, pcmk__output_t *out,
const pcmk_injections_t *injections, unsigned int flags,
uint32_t section_opts, const char *use_date,
const char *input_file, const char *graph_file,
const char *dot_file)
{
int printed = pcmk_rc_no_output;
int rc = pcmk_rc_ok;
xmlNodePtr input = NULL;
cib_t *cib = NULL;
rc = cib__signon_query(out, &cib, &input);
if (rc != pcmk_rc_ok) {
goto simulate_done;
}
reset(data_set, input, out, use_date, flags);
cluster_status(data_set);
if ((cib->variant == cib_native)
&& pcmk_is_set(section_opts, pcmk_section_times)) {
if (pcmk__our_nodename == NULL) {
// Currently used only in the times section
pcmk__query_node_name(out, 0, &pcmk__our_nodename, 0);
}
data_set->localhost = pcmk__our_nodename;
}
if (!out->is_quiet(out)) {
const bool show_pending = pcmk_is_set(flags, pcmk_sim_show_pending);
if (pcmk_is_set(data_set->flags, pcmk_sched_in_maintenance)) {
printed = out->message(out, "maint-mode", data_set->flags);
}
if (data_set->disabled_resources || data_set->blocked_resources) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
printed = out->info(out,
"%d of %d resource instances DISABLED and "
"%d BLOCKED from further action due to failure",
data_set->disabled_resources,
data_set->ninstances,
data_set->blocked_resources);
}
/* Most formatted output headers use caps for each word, but this one
* only has the first word capitalized for compatibility with pcs.
*/
print_cluster_status(data_set, (show_pending? pcmk_show_pending : 0),
section_opts, "Current cluster status",
(printed == pcmk_rc_ok));
printed = pcmk_rc_ok;
}
// If the user requested any injections, handle them
if ((injections->node_down != NULL)
|| (injections->node_fail != NULL)
|| (injections->node_up != NULL)
|| (injections->op_inject != NULL)
|| (injections->ticket_activate != NULL)
|| (injections->ticket_grant != NULL)
|| (injections->ticket_revoke != NULL)
|| (injections->ticket_standby != NULL)
|| (injections->watchdog != NULL)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
pcmk__inject_scheduler_input(data_set, cib, injections);
printed = pcmk_rc_ok;
rc = cib->cmds->query(cib, NULL, &input, cib_sync_call);
if (rc != pcmk_rc_ok) {
rc = pcmk_legacy2rc(rc);
goto simulate_done;
}
cleanup_calculations(data_set);
reset(data_set, input, out, use_date, flags);
cluster_status(data_set);
}
if (input_file != NULL) {
rc = write_xml_file(input, input_file, FALSE);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
goto simulate_done;
}
}
if (pcmk_any_flags_set(flags, pcmk_sim_process | pcmk_sim_simulate)) {
pcmk__output_t *logger_out = NULL;
unsigned long long data_set_flags = pcmk_sched_no_compat;
if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) {
data_set_flags |= pcmk_sched_output_scores;
}
if (pcmk_is_set(data_set->flags, pcmk_sched_show_utilization)) {
data_set_flags |= pcmk_sched_show_utilization;
}
if (pcmk_all_flags_set(data_set->flags,
pcmk_sched_output_scores
|pcmk_sched_show_utilization)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL,
"Assignment Scores and Utilization Information");
printed = pcmk_rc_ok;
} else if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL, "Assignment Scores");
printed = pcmk_rc_ok;
} else if (pcmk_is_set(data_set->flags, pcmk_sched_show_utilization)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL, "Utilization Information");
printed = pcmk_rc_ok;
} else {
rc = pcmk__log_output_new(&logger_out);
if (rc != pcmk_rc_ok) {
goto simulate_done;
}
pe__register_messages(logger_out);
pcmk__register_lib_messages(logger_out);
data_set->priv = logger_out;
}
pcmk__schedule_actions(input, data_set_flags, data_set);
if (logger_out == NULL) {
out->end_list(out);
} else {
logger_out->finish(logger_out, CRM_EX_OK, true, NULL);
pcmk__output_free(logger_out);
data_set->priv = out;
}
input = NULL; /* Don't try and free it twice */
if (graph_file != NULL) {
rc = write_xml_file(data_set->graph, graph_file, FALSE);
if (rc < 0) {
rc = pcmk_rc_graph_error;
goto simulate_done;
}
}
if (dot_file != NULL) {
rc = write_sim_dotfile(data_set, dot_file,
pcmk_is_set(flags, pcmk_sim_all_actions),
pcmk_is_set(flags, pcmk_sim_verbose));
if (rc != pcmk_rc_ok) {
rc = pcmk_rc_dot_error;
goto simulate_done;
}
}
if (!out->is_quiet(out)) {
print_transition_summary(data_set, printed == pcmk_rc_ok);
}
}
rc = pcmk_rc_ok;
if (!pcmk_is_set(flags, pcmk_sim_simulate)) {
goto simulate_done;
}
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
if (pcmk__simulate_transition(data_set, cib, injections->op_fail)
!= pcmk__graph_complete) {
rc = pcmk_rc_invalid_transition;
}
if (out->is_quiet(out)) {
goto simulate_done;
}
set_effective_date(data_set, true, use_date);
if (pcmk_is_set(flags, pcmk_sim_show_scores)) {
pe__set_working_set_flags(data_set, pcmk_sched_output_scores);
}
if (pcmk_is_set(flags, pcmk_sim_show_utilization)) {
pe__set_working_set_flags(data_set, pcmk_sched_show_utilization);
}
cluster_status(data_set);
print_cluster_status(data_set, 0, section_opts, "Revised Cluster Status",
true);
simulate_done:
cib__clean_up_connection(&cib);
return rc;
}
int
pcmk_simulate(xmlNodePtr *xml, pe_working_set_t *data_set,
const pcmk_injections_t *injections, unsigned int flags,
unsigned int section_opts, const char *use_date,
const char *input_file, const char *graph_file,
const char *dot_file)
{
pcmk__output_t *out = NULL;
int rc = pcmk_rc_ok;
rc = pcmk__xml_output_new(&out, xml);
if (rc != pcmk_rc_ok) {
return rc;
}
pe__register_messages(out);
pcmk__register_lib_messages(out);
rc = pcmk__simulate(data_set, out, injections, flags, section_opts,
use_date, input_file, graph_file, dot_file);
pcmk__xml_output_finish(out, xml);
return rc;
}
diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c
index 4dfb46a4b9..8ea5884520 100644
--- a/lib/pengine/pe_actions.c
+++ b/lib/pengine/pe_actions.c
@@ -1,1745 +1,1745 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/scheduler_internal.h>
#include <crm/pengine/internal.h>
#include <crm/common/xml_internal.h>
#include "pe_status_private.h"
static void unpack_operation(pe_action_t *action, const xmlNode *xml_obj,
const pe_resource_t *container, guint interval_ms);
static void
add_singleton(pe_working_set_t *data_set, pe_action_t *action)
{
if (data_set->singletons == NULL) {
data_set->singletons = pcmk__strkey_table(NULL, NULL);
}
g_hash_table_insert(data_set->singletons, action->uuid, action);
}
static pe_action_t *
lookup_singleton(pe_working_set_t *data_set, const char *action_uuid)
{
if (data_set->singletons == NULL) {
return NULL;
}
return g_hash_table_lookup(data_set->singletons, action_uuid);
}
/*!
* \internal
* \brief Find an existing action that matches arguments
*
* \param[in] key Action key to match
* \param[in] rsc Resource to match (if any)
* \param[in] node Node to match (if any)
* \param[in] data_set Cluster working set
*
* \return Existing action that matches arguments (or NULL if none)
*/
static pe_action_t *
find_existing_action(const char *key, const pe_resource_t *rsc,
const pe_node_t *node, const pe_working_set_t *data_set)
{
GList *matches = NULL;
pe_action_t *action = NULL;
/* When rsc is NULL, it would be quicker to check data_set->singletons,
* but checking all data_set->actions takes the node into account.
*/
matches = find_actions(((rsc == NULL)? data_set->actions : rsc->actions),
key, node);
if (matches == NULL) {
return NULL;
}
CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches));
action = matches->data;
g_list_free(matches);
return action;
}
static xmlNode *
find_rsc_op_entry_helper(const pe_resource_t *rsc, const char *key,
gboolean include_disabled)
{
guint interval_ms = 0;
gboolean do_retry = TRUE;
char *local_key = NULL;
const char *name = NULL;
const char *interval_spec = NULL;
char *match_key = NULL;
xmlNode *op = NULL;
xmlNode *operation = NULL;
retry:
for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL;
operation = pcmk__xe_next(operation)) {
if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) {
bool enabled = false;
name = crm_element_value(operation, "name");
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (!include_disabled && pcmk__xe_get_bool_attr(operation, "enabled", &enabled) == pcmk_rc_ok &&
!enabled) {
continue;
}
interval_ms = crm_parse_interval_spec(interval_spec);
match_key = pcmk__op_key(rsc->id, name, interval_ms);
if (pcmk__str_eq(key, match_key, pcmk__str_casei)) {
op = operation;
}
free(match_key);
if (rsc->clone_name) {
match_key = pcmk__op_key(rsc->clone_name, name, interval_ms);
if (pcmk__str_eq(key, match_key, pcmk__str_casei)) {
op = operation;
}
free(match_key);
}
if (op != NULL) {
free(local_key);
return op;
}
}
}
free(local_key);
if (do_retry == FALSE) {
return NULL;
}
do_retry = FALSE;
if ((strstr(key, PCMK_ACTION_MIGRATE_TO) != NULL)
|| (strstr(key, PCMK_ACTION_MIGRATE_FROM) != NULL)) {
local_key = pcmk__op_key(rsc->id, "migrate", 0);
key = local_key;
goto retry;
} else if (strstr(key, "_notify_")) {
local_key = pcmk__op_key(rsc->id, PCMK_ACTION_NOTIFY, 0);
key = local_key;
goto retry;
}
return NULL;
}
xmlNode *
find_rsc_op_entry(const pe_resource_t *rsc, const char *key)
{
return find_rsc_op_entry_helper(rsc, key, FALSE);
}
/*!
* \internal
* \brief Create a new action object
*
* \param[in] key Action key
* \param[in] task Action name
* \param[in,out] rsc Resource that action is for (if any)
* \param[in] node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in] for_graph Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \return Newly allocated action
* \note This function takes ownership of \p key. It is the caller's
* responsibility to free the return value with pe_free_action().
*/
static pe_action_t *
new_action(char *key, const char *task, pe_resource_t *rsc,
const pe_node_t *node, bool optional, bool for_graph,
pe_working_set_t *data_set)
{
pe_action_t *action = calloc(1, sizeof(pe_action_t));
CRM_ASSERT(action != NULL);
action->rsc = rsc;
action->task = strdup(task); CRM_ASSERT(action->task != NULL);
action->uuid = key;
action->extra = pcmk__strkey_table(free, free);
action->meta = pcmk__strkey_table(free, free);
if (node) {
action->node = pe__copy_node(node);
}
if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_casei)) {
// Resource history deletion for a node can be done on the DC
pe__set_action_flags(action, pe_action_dc);
}
pe__set_action_flags(action, pe_action_runnable);
if (optional) {
pe__set_action_flags(action, pe_action_optional);
} else {
pe__clear_action_flags(action, pe_action_optional);
}
if (rsc != NULL) {
guint interval_ms = 0;
action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE);
parse_op_key(key, NULL, NULL, &interval_ms);
unpack_operation(action, action->op_entry, rsc->container, interval_ms);
}
if (for_graph) {
pe_rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s",
(optional? "optional" : "required"),
data_set->action_id, key, task,
((rsc == NULL)? "no resource" : rsc->id),
pe__node_name(node));
action->id = data_set->action_id++;
data_set->actions = g_list_prepend(data_set->actions, action);
if (rsc == NULL) {
add_singleton(data_set, action);
} else {
rsc->actions = g_list_prepend(rsc->actions, action);
}
}
return action;
}
/*!
* \internal
* \brief Evaluate node attribute values for an action
*
* \param[in,out] action Action to unpack attributes for
* \param[in,out] data_set Cluster working set
*/
static void
unpack_action_node_attributes(pe_action_t *action, pe_working_set_t *data_set)
{
if (!pcmk_is_set(action->flags, pe_action_have_node_attrs)
&& (action->op_entry != NULL)) {
pe_rule_eval_data_t rule_data = {
.node_hash = action->node->details->attrs,
.role = pcmk_role_unknown,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
pe__set_action_flags(action, pe_action_have_node_attrs);
pe__unpack_dataset_nvpairs(action->op_entry, XML_TAG_ATTR_SETS,
&rule_data, action->extra, NULL,
FALSE, data_set);
}
}
/*!
* \internal
* \brief Update an action's optional flag
*
* \param[in,out] action Action to update
* \param[in] optional Requested optional status
*/
static void
update_action_optional(pe_action_t *action, gboolean optional)
{
// Force a non-recurring action to be optional if its resource is unmanaged
if ((action->rsc != NULL) && (action->node != NULL)
- && !pcmk_is_set(action->flags, pe_action_pseudo)
+ && !pcmk_is_set(action->flags, pcmk_action_pseudo)
&& !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)
&& (g_hash_table_lookup(action->meta,
XML_LRM_ATTR_INTERVAL_MS) == NULL)) {
pe_rsc_debug(action->rsc, "%s on %s is optional (%s is unmanaged)",
action->uuid, pe__node_name(action->node),
action->rsc->id);
pe__set_action_flags(action, pe_action_optional);
// We shouldn't clear runnable here because ... something
// Otherwise require the action if requested
} else if (!optional) {
pe__clear_action_flags(action, pe_action_optional);
}
}
static enum pe_quorum_policy
effective_quorum_policy(pe_resource_t *rsc, pe_working_set_t *data_set)
{
enum pe_quorum_policy policy = data_set->no_quorum_policy;
if (pcmk_is_set(data_set->flags, pcmk_sched_quorate)) {
policy = pcmk_no_quorum_ignore;
} else if (data_set->no_quorum_policy == pcmk_no_quorum_demote) {
switch (rsc->role) {
case pcmk_role_promoted:
case pcmk_role_unpromoted:
if (rsc->next_role > pcmk_role_unpromoted) {
pe__set_next_role(rsc, pcmk_role_unpromoted,
"no-quorum-policy=demote");
}
policy = pcmk_no_quorum_ignore;
break;
default:
policy = pcmk_no_quorum_stop;
break;
}
}
return policy;
}
/*!
* \internal
* \brief Update a resource action's runnable flag
*
* \param[in,out] action Action to update
* \param[in] for_graph Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \note This may also schedule fencing if a stop is unrunnable.
*/
static void
update_resource_action_runnable(pe_action_t *action, bool for_graph,
pe_working_set_t *data_set)
{
- if (pcmk_is_set(action->flags, pe_action_pseudo)) {
+ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
return;
}
if (action->node == NULL) {
pe_rsc_trace(action->rsc, "%s is unrunnable (unallocated)",
action->uuid);
pe__clear_action_flags(action, pe_action_runnable);
} else if (!pcmk_is_set(action->flags, pe_action_dc)
&& !(action->node->details->online)
&& (!pe__is_guest_node(action->node)
|| action->node->details->remote_requires_reset)) {
pe__clear_action_flags(action, pe_action_runnable);
do_crm_log((for_graph? LOG_WARNING: LOG_TRACE),
"%s on %s is unrunnable (node is offline)",
action->uuid, pe__node_name(action->node));
if (pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)
&& for_graph
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)
&& !(action->node->details->unclean)) {
pe_fence_node(data_set, action->node, "stop is unrunnable", false);
}
} else if (!pcmk_is_set(action->flags, pe_action_dc)
&& action->node->details->pending) {
pe__clear_action_flags(action, pe_action_runnable);
do_crm_log((for_graph? LOG_WARNING: LOG_TRACE),
"Action %s on %s is unrunnable (node is pending)",
action->uuid, pe__node_name(action->node));
} else if (action->needs == pcmk_requires_nothing) {
pe_action_set_reason(action, NULL, TRUE);
if (pe__is_guest_node(action->node)
&& !pe_can_fence(data_set, action->node)) {
/* An action that requires nothing usually does not require any
* fencing in order to be runnable. However, there is an exception:
* such an action cannot be completed if it is on a guest node whose
* host is unclean and cannot be fenced.
*/
pe_rsc_debug(action->rsc, "%s on %s is unrunnable "
"(node's host cannot be fenced)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
} else {
pe_rsc_trace(action->rsc,
"%s on %s does not require fencing or quorum",
action->uuid, pe__node_name(action->node));
pe__set_action_flags(action, pe_action_runnable);
}
} else {
switch (effective_quorum_policy(action->rsc, data_set)) {
case pcmk_no_quorum_stop:
pe_rsc_debug(action->rsc, "%s on %s is unrunnable (no quorum)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
pe_action_set_reason(action, "no quorum", true);
break;
case pcmk_no_quorum_freeze:
if (!action->rsc->fns->active(action->rsc, TRUE)
|| (action->rsc->next_role > action->rsc->role)) {
pe_rsc_debug(action->rsc,
"%s on %s is unrunnable (no quorum)",
action->uuid, pe__node_name(action->node));
pe__clear_action_flags(action, pe_action_runnable);
pe_action_set_reason(action, "quorum freeze", true);
}
break;
default:
//pe_action_set_reason(action, NULL, TRUE);
pe__set_action_flags(action, pe_action_runnable);
break;
}
}
}
/*!
* \internal
* \brief Update a resource object's flags for a new action on it
*
* \param[in,out] rsc Resource that action is for (if any)
* \param[in] action New action
*/
static void
update_resource_flags_for_action(pe_resource_t *rsc, const pe_action_t *action)
{
/* @COMPAT pcmk_rsc_starting and pcmk_rsc_stopping are deprecated and unused
* within Pacemaker, and will eventually be removed
*/
if (pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)) {
pe__set_resource_flags(rsc, pcmk_rsc_stopping);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)) {
if (pcmk_is_set(action->flags, pe_action_runnable)) {
pe__set_resource_flags(rsc, pcmk_rsc_starting);
} else {
pe__clear_resource_flags(rsc, pcmk_rsc_starting);
}
}
}
static bool
valid_stop_on_fail(const char *value)
{
return !pcmk__strcase_any_of(value, "standby", "demote", "stop", NULL);
}
/*!
* \internal
* \brief Validate (and possibly reset) resource action's on_fail meta-attribute
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Action name
* \param[in] action_config Action configuration XML from CIB (if any)
* \param[in,out] meta Table of action meta-attributes
*
* \return (Possibly new) value of on-fail meta-attribute
*/
static const char *
validate_on_fail(const pe_resource_t *rsc, const char *action_name,
const xmlNode *action_config, GHashTable *meta)
{
const char *name = NULL;
const char *role = NULL;
const char *interval_spec = NULL;
const char *value = g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
char *key = NULL;
char *new_value = NULL;
// Stop actions can only use certain on-fail values
if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)
&& !valid_stop_on_fail(value)) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s stop "
"action to default value because '%s' is not "
"allowed for stop", rsc->id, value);
g_hash_table_remove(meta, XML_OP_ATTR_ON_FAIL);
return NULL;
}
/* Demote actions default on-fail to the on-fail value for the first
* recurring monitor for the promoted role (if any).
*/
if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none)
&& (value == NULL)) {
/* @TODO This does not consider promote options set in a meta-attribute
* block (which may have rules that need to be evaluated) rather than
* XML properties.
*/
for (xmlNode *operation = first_named_child(rsc->ops_xml, XML_ATTR_OP);
operation != NULL; operation = crm_next_same_xml(operation)) {
bool enabled = false;
const char *promote_on_fail = NULL;
/* We only care about explicit on-fail (if promote uses default, so
* can demote)
*/
promote_on_fail = crm_element_value(operation, XML_OP_ATTR_ON_FAIL);
if (promote_on_fail == NULL) {
continue;
}
// We only care about recurring monitors for the promoted role
name = crm_element_value(operation, "name");
role = crm_element_value(operation, "role");
if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)) {
continue;
}
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (crm_parse_interval_spec(interval_spec) == 0) {
continue;
}
// We only care about enabled monitors
if ((pcmk__xe_get_bool_attr(operation, "enabled",
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
// Demote actions can't default to on-fail="demote"
if (pcmk__str_eq(promote_on_fail, "demote", pcmk__str_casei)) {
continue;
}
// Use value from first applicable promote action found
key = strdup(XML_OP_ATTR_ON_FAIL);
new_value = strdup(promote_on_fail);
CRM_ASSERT((key != NULL) && (new_value != NULL));
g_hash_table_insert(meta, key, new_value);
return g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
}
return NULL;
}
if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none)
&& !pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
key = strdup(XML_OP_ATTR_ON_FAIL);
new_value = strdup("ignore");
CRM_ASSERT((key != NULL) && (new_value != NULL));
g_hash_table_insert(meta, key, new_value);
return g_hash_table_lookup(meta, XML_OP_ATTR_ON_FAIL);
}
// on-fail="demote" is allowed only for certain actions
if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
name = crm_element_value(action_config, "name");
role = crm_element_value(action_config, "role");
interval_spec = crm_element_value(action_config,
XML_LRM_ATTR_INTERVAL);
if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none)
&& (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK__ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL)
|| (crm_parse_interval_spec(interval_spec) == 0))) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s %s "
"action to default value because 'demote' is not "
"allowed for it", rsc->id, name);
g_hash_table_remove(meta, XML_OP_ATTR_ON_FAIL);
return NULL;
}
}
return value;
}
static int
unpack_timeout(const char *value)
{
int timeout_ms = crm_get_msec(value);
if (timeout_ms < 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
return timeout_ms;
}
// true if value contains valid, non-NULL interval origin for recurring op
static bool
unpack_interval_origin(const char *value, const xmlNode *xml_obj,
guint interval_ms, const crm_time_t *now,
long long *start_delay)
{
long long result = 0;
guint interval_sec = interval_ms / 1000;
crm_time_t *origin = NULL;
// Ignore unspecified values and non-recurring operations
if ((value == NULL) || (interval_ms == 0) || (now == NULL)) {
return false;
}
// Parse interval origin from text
origin = crm_time_new(value);
if (origin == NULL) {
pcmk__config_err("Ignoring '" XML_OP_ATTR_ORIGIN "' for operation "
"'%s' because '%s' is not valid",
(ID(xml_obj)? ID(xml_obj) : "(missing ID)"), value);
return false;
}
// Get seconds since origin (negative if origin is in the future)
result = crm_time_get_seconds(now) - crm_time_get_seconds(origin);
crm_time_free(origin);
// Calculate seconds from closest interval to now
result = result % interval_sec;
// Calculate seconds remaining until next interval
result = ((result <= 0)? 0 : interval_sec) - result;
crm_info("Calculated a start delay of %llds for operation '%s'",
result,
(ID(xml_obj)? ID(xml_obj) : "(unspecified)"));
if (start_delay != NULL) {
*start_delay = result * 1000; // milliseconds
}
return true;
}
static int
unpack_start_delay(const char *value, GHashTable *meta)
{
int start_delay = 0;
if (value != NULL) {
start_delay = crm_get_msec(value);
if (start_delay < 0) {
start_delay = 0;
}
if (meta) {
g_hash_table_replace(meta, strdup(XML_OP_ATTR_START_DELAY),
pcmk__itoa(start_delay));
}
}
return start_delay;
}
static xmlNode *
find_min_interval_mon(pe_resource_t * rsc, gboolean include_disabled)
{
guint interval_ms = 0;
guint min_interval_ms = G_MAXUINT;
const char *name = NULL;
const char *interval_spec = NULL;
xmlNode *op = NULL;
xmlNode *operation = NULL;
for (operation = pcmk__xe_first_child(rsc->ops_xml);
operation != NULL;
operation = pcmk__xe_next(operation)) {
if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) {
bool enabled = false;
name = crm_element_value(operation, "name");
interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL);
if (!include_disabled && pcmk__xe_get_bool_attr(operation, "enabled", &enabled) == pcmk_rc_ok &&
!enabled) {
continue;
}
if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
continue;
}
interval_ms = crm_parse_interval_spec(interval_spec);
if (interval_ms && (interval_ms < min_interval_ms)) {
min_interval_ms = interval_ms;
op = operation;
}
}
}
return op;
}
/*!
* \internal
* \brief Unpack action configuration
*
* Unpack a resource action's meta-attributes (normalizing the interval,
* timeout, and start delay values as integer milliseconds), requirements, and
* failure policy from its CIB XML configuration (including defaults).
*
* \param[in,out] action Resource action to unpack into
* \param[in] xml_obj Action configuration XML (NULL for defaults only)
* \param[in] container Resource that contains affected resource, if any
* \param[in] interval_ms How frequently to perform the operation
*/
static void
unpack_operation(pe_action_t *action, const xmlNode *xml_obj,
const pe_resource_t *container, guint interval_ms)
{
int timeout_ms = 0;
const char *value = NULL;
bool is_probe = false;
pe_rsc_eval_data_t rsc_rule_data = {
.standard = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_CLASS),
.provider = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_PROVIDER),
.agent = crm_element_value(action->rsc->xml, XML_EXPR_ATTR_TYPE)
};
pe_op_eval_data_t op_rule_data = {
.op_name = action->task,
.interval = interval_ms
};
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = action->rsc->cluster->now,
.match_data = NULL,
.rsc_data = &rsc_rule_data,
.op_data = &op_rule_data
};
CRM_CHECK(action && action->rsc, return);
is_probe = pcmk_is_probe(action->task, interval_ms);
// Cluster-wide <op_defaults> <meta_attributes>
pe__unpack_dataset_nvpairs(action->rsc->cluster->op_defaults,
XML_TAG_META_SETS, &rule_data, action->meta,
NULL, FALSE, action->rsc->cluster);
// Determine probe default timeout differently
if (is_probe) {
xmlNode *min_interval_mon = find_min_interval_mon(action->rsc, FALSE);
if (min_interval_mon) {
value = crm_element_value(min_interval_mon, XML_ATTR_TIMEOUT);
if (value) {
crm_trace("\t%s: Setting default timeout to minimum-interval "
"monitor's timeout '%s'", action->uuid, value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
strdup(value));
}
}
}
if (xml_obj) {
xmlAttrPtr xIter = NULL;
// <op> <meta_attributes> take precedence over defaults
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_META_SETS, &rule_data,
action->meta, NULL, TRUE,
action->rsc->cluster);
/* Anything set as an <op> XML property has highest precedence.
* This ensures we use the name and interval from the <op> tag.
*/
for (xIter = xml_obj->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
g_hash_table_replace(action->meta, strdup(prop_name), strdup(prop_value));
}
}
g_hash_table_remove(action->meta, "id");
// Normalize interval to milliseconds
if (interval_ms > 0) {
g_hash_table_replace(action->meta, strdup(XML_LRM_ATTR_INTERVAL),
crm_strdup_printf("%u", interval_ms));
} else {
g_hash_table_remove(action->meta, XML_LRM_ATTR_INTERVAL);
}
/*
* Timeout order of precedence:
* 1. pcmk_monitor_timeout (if rsc has pcmk_ra_cap_fence_params
* and task is start or a probe; pcmk_monitor_timeout works
* by default for a recurring monitor)
* 2. explicit op timeout on the primitive
* 3. default op timeout
* a. if probe, then min-interval monitor's timeout
* b. else, in XML_CIB_TAG_OPCONFIG
* 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS
*
* #1 overrides general rule of <op> XML property having highest
* precedence.
*/
if (pcmk_is_set(pcmk_get_ra_caps(rsc_rule_data.standard),
pcmk_ra_cap_fence_params)
&& (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)
|| is_probe)) {
GHashTable *params = pe_rsc_params(action->rsc, action->node,
action->rsc->cluster);
value = g_hash_table_lookup(params, "pcmk_monitor_timeout");
if (value) {
crm_trace("\t%s: Setting timeout to pcmk_monitor_timeout '%s', "
"overriding default", action->uuid, value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
strdup(value));
}
}
// Normalize timeout to positive milliseconds
value = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT);
timeout_ms = unpack_timeout(value);
g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT),
pcmk__itoa(timeout_ms));
if (!pcmk__strcase_any_of(action->task, PCMK_ACTION_START,
PCMK_ACTION_PROMOTE, NULL)) {
action->needs = pcmk_requires_nothing;
value = "nothing (not start or promote)";
} else if (pcmk_is_set(action->rsc->flags, pcmk_rsc_needs_fencing)) {
action->needs = pcmk_requires_fencing;
value = "fencing";
} else if (pcmk_is_set(action->rsc->flags, pcmk_rsc_needs_quorum)) {
action->needs = pcmk_requires_quorum;
value = "quorum";
} else {
action->needs = pcmk_requires_nothing;
value = "nothing";
}
pe_rsc_trace(action->rsc, "%s requires %s", action->uuid, value);
value = validate_on_fail(action->rsc, action->task, xml_obj, action->meta);
if (value == NULL) {
} else if (pcmk__str_eq(value, "block", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_block;
} else if (pcmk__str_eq(value, "fence", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_fence_node;
value = "node fencing";
if (!pcmk_is_set(action->rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for "
"operation '%s' to 'stop' because 'fence' is not "
"valid when fencing is disabled", action->uuid);
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop resource";
}
} else if (pcmk__str_eq(value, "standby", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_standby_node;
value = "node standby";
} else if (pcmk__strcase_any_of(value, "ignore", PCMK__VALUE_NOTHING,
NULL)) {
action->on_fail = pcmk_on_fail_ignore;
value = "ignore";
} else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_ban;
value = "force migration";
} else if (pcmk__str_eq(value, "stop", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop resource";
} else if (pcmk__str_eq(value, "restart", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_restart;
value = "restart (and possibly migrate)";
} else if (pcmk__str_eq(value, "restart-container", pcmk__str_casei)) {
if (container) {
action->on_fail = pcmk_on_fail_restart_container;
value = "restart container (and possibly migrate)";
} else {
value = NULL;
}
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
action->on_fail = pcmk_on_fail_demote;
value = "demote instance";
} else {
pe_err("Resource %s: Unknown failure type (%s)", action->rsc->id, value);
value = NULL;
}
/* defaults */
if (value == NULL && container) {
action->on_fail = pcmk_on_fail_restart_container;
value = "restart container (and possibly migrate) (default)";
/* For remote nodes, ensure that any failure that results in dropping an
* active connection to the node results in fencing of the node.
*
* There are only two action failures that don't result in fencing.
* 1. probes - probe failures are expected.
* 2. start - a start failure indicates that an active connection does not already
* exist. The user can set op on-fail=fence if they really want to fence start
* failures. */
} else if (((value == NULL)
|| !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed))
&& pe__resource_is_remote_conn(action->rsc)
&& !(pcmk__str_eq(action->task, PCMK_ACTION_MONITOR,
pcmk__str_casei)
&& (interval_ms == 0))
&& !pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_casei)) {
if (!pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)) {
action->on_fail = pcmk_on_fail_stop;
action->fail_role = pcmk_role_stopped;
value = "stop unmanaged remote node (enforcing default)";
} else {
if (pcmk_is_set(action->rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
value = "fence remote node (default)";
} else {
value = "recover remote node connection (default)";
}
if (action->rsc->remote_reconnect_ms) {
action->fail_role = pcmk_role_stopped;
}
action->on_fail = pcmk_on_fail_reset_remote;
}
} else if ((value == NULL)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP,
pcmk__str_casei)) {
if (pcmk_is_set(action->rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
action->on_fail = pcmk_on_fail_fence_node;
value = "resource fence (default)";
} else {
action->on_fail = pcmk_on_fail_block;
value = "resource block (default)";
}
} else if (value == NULL) {
action->on_fail = pcmk_on_fail_restart;
value = "restart (and possibly migrate) (default)";
}
pe_rsc_trace(action->rsc, "%s failure handling: %s",
action->uuid, value);
value = NULL;
if (xml_obj != NULL) {
value = g_hash_table_lookup(action->meta, "role_after_failure");
if (value) {
pe_warn_once(pe_wo_role_after,
"Support for role_after_failure is deprecated and will be removed in a future release");
}
}
if (value != NULL && action->fail_role == pcmk_role_unknown) {
action->fail_role = text2role(value);
}
/* defaults */
if (action->fail_role == pcmk_role_unknown) {
if (pcmk__str_eq(action->task, PCMK_ACTION_PROMOTE, pcmk__str_casei)) {
action->fail_role = pcmk_role_unpromoted;
} else {
action->fail_role = pcmk_role_started;
}
}
pe_rsc_trace(action->rsc, "%s failure results in: %s",
action->uuid, role2text(action->fail_role));
value = g_hash_table_lookup(action->meta, XML_OP_ATTR_START_DELAY);
if (value) {
unpack_start_delay(value, action->meta);
} else {
long long start_delay = 0;
value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ORIGIN);
if (unpack_interval_origin(value, xml_obj, interval_ms,
action->rsc->cluster->now, &start_delay)) {
g_hash_table_replace(action->meta, strdup(XML_OP_ATTR_START_DELAY),
crm_strdup_printf("%lld", start_delay));
}
}
}
/*!
* \brief Create or update an action object
*
* \param[in,out] rsc Resource that action is for (if any)
* \param[in,out] key Action key (must be non-NULL)
* \param[in] task Action name (must be non-NULL)
* \param[in] on_node Node that action is on (if any)
* \param[in] optional Whether action should be considered optional
* \param[in] save_action Whether action should be recorded in transition graph
* \param[in,out] data_set Cluster working set
*
* \return Action object corresponding to arguments
* \note This function takes ownership of (and might free) \p key. If
* \p save_action is true, \p data_set will own the returned action,
* otherwise it is the caller's responsibility to free the return value
* with pe_free_action().
*/
pe_action_t *
custom_action(pe_resource_t *rsc, char *key, const char *task,
const pe_node_t *on_node, gboolean optional, gboolean save_action,
pe_working_set_t *data_set)
{
pe_action_t *action = NULL;
CRM_ASSERT((key != NULL) && (task != NULL) && (data_set != NULL));
if (save_action) {
action = find_existing_action(key, rsc, on_node, data_set);
}
if (action == NULL) {
action = new_action(key, task, rsc, on_node, optional, save_action,
data_set);
} else {
free(key);
}
update_action_optional(action, optional);
if (rsc != NULL) {
if (action->node != NULL) {
unpack_action_node_attributes(action, data_set);
}
update_resource_action_runnable(action, save_action, data_set);
if (save_action) {
update_resource_flags_for_action(rsc, action);
}
}
return action;
}
pe_action_t *
get_pseudo_op(const char *name, pe_working_set_t * data_set)
{
pe_action_t *op = lookup_singleton(data_set, name);
if (op == NULL) {
op = custom_action(NULL, strdup(name), name, NULL, TRUE, TRUE, data_set);
- pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable);
+ pe__set_action_flags(op, pcmk_action_pseudo|pe_action_runnable);
}
return op;
}
static GList *
find_unfencing_devices(GList *candidates, GList *matches)
{
for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) {
pe_resource_t *candidate = gIter->data;
if (candidate->children != NULL) {
matches = find_unfencing_devices(candidate->children, matches);
} else if (!pcmk_is_set(candidate->flags, pcmk_rsc_fence_device)) {
continue;
} else if (pcmk_is_set(candidate->flags, pcmk_rsc_needs_unfencing)) {
matches = g_list_prepend(matches, candidate);
} else if (pcmk__str_eq(g_hash_table_lookup(candidate->meta,
PCMK_STONITH_PROVIDES),
PCMK__VALUE_UNFENCING,
pcmk__str_casei)) {
matches = g_list_prepend(matches, candidate);
}
}
return matches;
}
static int
node_priority_fencing_delay(const pe_node_t *node,
const pe_working_set_t *data_set)
{
int member_count = 0;
int online_count = 0;
int top_priority = 0;
int lowest_priority = 0;
GList *gIter = NULL;
// `priority-fencing-delay` is disabled
if (data_set->priority_fencing_delay <= 0) {
return 0;
}
/* No need to request a delay if the fencing target is not a normal cluster
* member, for example if it's a remote node or a guest node. */
if (node->details->type != pcmk_node_variant_cluster) {
return 0;
}
// No need to request a delay if the fencing target is in our partition
if (node->details->online) {
return 0;
}
for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) {
pe_node_t *n = gIter->data;
if (n->details->type != pcmk_node_variant_cluster) {
continue;
}
member_count ++;
if (n->details->online) {
online_count++;
}
if (member_count == 1
|| n->details->priority > top_priority) {
top_priority = n->details->priority;
}
if (member_count == 1
|| n->details->priority < lowest_priority) {
lowest_priority = n->details->priority;
}
}
// No need to delay if we have more than half of the cluster members
if (online_count > member_count / 2) {
return 0;
}
/* All the nodes have equal priority.
* Any configured corresponding `pcmk_delay_base/max` will be applied. */
if (lowest_priority == top_priority) {
return 0;
}
if (node->details->priority < top_priority) {
return 0;
}
return data_set->priority_fencing_delay;
}
pe_action_t *
pe_fence_op(pe_node_t *node, const char *op, bool optional,
const char *reason, bool priority_delay, pe_working_set_t *data_set)
{
char *op_key = NULL;
pe_action_t *stonith_op = NULL;
if(op == NULL) {
op = data_set->stonith_action;
}
op_key = crm_strdup_printf("%s-%s-%s",
PCMK_ACTION_STONITH, node->details->uname, op);
stonith_op = lookup_singleton(data_set, op_key);
if(stonith_op == NULL) {
stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node,
TRUE, TRUE, data_set);
add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET, node->details->uname);
add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET_UUID, node->details->id);
add_hash_param(stonith_op->meta, "stonith_action", op);
if (pcmk_is_set(data_set->flags, pcmk_sched_enable_unfencing)) {
/* Extra work to detect device changes
*/
GString *digests_all = g_string_sized_new(1024);
GString *digests_secure = g_string_sized_new(1024);
GList *matches = find_unfencing_devices(data_set->resources, NULL);
char *key = NULL;
char *value = NULL;
for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) {
pe_resource_t *match = gIter->data;
const char *agent = g_hash_table_lookup(match->meta,
XML_ATTR_TYPE);
op_digest_cache_t *data = NULL;
data = pe__compare_fencing_digest(match, agent, node, data_set);
if(data->rc == RSC_DIGEST_ALL) {
optional = FALSE;
crm_notice("Unfencing node %s because the definition of "
"%s changed", pe__node_name(node), match->id);
if (!pcmk__is_daemon && data_set->priv != NULL) {
pcmk__output_t *out = data_set->priv;
out->info(out,
"notice: Unfencing node %s because the "
"definition of %s changed",
pe__node_name(node), match->id);
}
}
pcmk__g_strcat(digests_all,
match->id, ":", agent, ":",
data->digest_all_calc, ",", NULL);
pcmk__g_strcat(digests_secure,
match->id, ":", agent, ":",
data->digest_secure_calc, ",", NULL);
}
key = strdup(XML_OP_ATTR_DIGESTS_ALL);
value = strdup((const char *) digests_all->str);
CRM_ASSERT((key != NULL) && (value != NULL));
g_hash_table_insert(stonith_op->meta, key, value);
g_string_free(digests_all, TRUE);
key = strdup(XML_OP_ATTR_DIGESTS_SECURE);
value = strdup((const char *) digests_secure->str);
CRM_ASSERT((key != NULL) && (value != NULL));
g_hash_table_insert(stonith_op->meta, key, value);
g_string_free(digests_secure, TRUE);
}
} else {
free(op_key);
}
if (data_set->priority_fencing_delay > 0
/* It's a suitable case where `priority-fencing-delay` applies.
* At least add `priority-fencing-delay` field as an indicator. */
&& (priority_delay
/* The priority delay needs to be recalculated if this function has
* been called by schedule_fencing_and_shutdowns() after node
* priority has already been calculated by native_add_running().
*/
|| g_hash_table_lookup(stonith_op->meta,
XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY) != NULL)) {
/* Add `priority-fencing-delay` to the fencing op even if it's 0 for
* the targeting node. So that it takes precedence over any possible
* `pcmk_delay_base/max`.
*/
char *delay_s = pcmk__itoa(node_priority_fencing_delay(node, data_set));
g_hash_table_insert(stonith_op->meta,
strdup(XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY),
delay_s);
}
if(optional == FALSE && pe_can_fence(data_set, node)) {
pe__clear_action_flags(stonith_op, pe_action_optional);
pe_action_set_reason(stonith_op, reason, false);
} else if(reason && stonith_op->reason == NULL) {
stonith_op->reason = strdup(reason);
}
return stonith_op;
}
void
pe_free_action(pe_action_t * action)
{
if (action == NULL) {
return;
}
g_list_free_full(action->actions_before, free); /* pe_action_wrapper_t* */
g_list_free_full(action->actions_after, free); /* pe_action_wrapper_t* */
if (action->extra) {
g_hash_table_destroy(action->extra);
}
if (action->meta) {
g_hash_table_destroy(action->meta);
}
free(action->cancel_task);
free(action->reason);
free(action->task);
free(action->uuid);
free(action->node);
free(action);
}
int
pe_get_configured_timeout(pe_resource_t *rsc, const char *action, pe_working_set_t *data_set)
{
xmlNode *child = NULL;
GHashTable *action_meta = NULL;
const char *timeout_spec = NULL;
int timeout_ms = 0;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = data_set->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
for (child = first_named_child(rsc->ops_xml, XML_ATTR_OP);
child != NULL; child = crm_next_same_xml(child)) {
if (pcmk__str_eq(action, crm_element_value(child, XML_NVPAIR_ATTR_NAME),
pcmk__str_casei)) {
timeout_spec = crm_element_value(child, XML_ATTR_TIMEOUT);
break;
}
}
if (timeout_spec == NULL && data_set->op_defaults) {
action_meta = pcmk__strkey_table(free, free);
pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS,
&rule_data, action_meta, NULL, FALSE, data_set);
timeout_spec = g_hash_table_lookup(action_meta, XML_ATTR_TIMEOUT);
}
// @TODO check meta-attributes
// @TODO maybe use min-interval monitor timeout as default for monitors
timeout_ms = crm_get_msec(timeout_spec);
if (timeout_ms < 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
if (action_meta != NULL) {
g_hash_table_destroy(action_meta);
}
return timeout_ms;
}
enum action_tasks
get_complex_task(const pe_resource_t *rsc, const char *name)
{
enum action_tasks task = text2task(name);
if ((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)) {
switch (task) {
case pcmk_action_stopped:
case pcmk_action_started:
case pcmk_action_demoted:
case pcmk_action_promoted:
crm_trace("Folding %s back into its atomic counterpart for %s",
name, rsc->id);
--task;
break;
default:
break;
}
}
return task;
}
/*!
* \internal
* \brief Find first matching action in a list
*
* \param[in] input List of actions to search
* \param[in] uuid If not NULL, action must have this UUID
* \param[in] task If not NULL, action must have this action name
* \param[in] on_node If not NULL, action must be on this node
*
* \return First action in list that matches criteria, or NULL if none
*/
pe_action_t *
find_first_action(const GList *input, const char *uuid, const char *task,
const pe_node_t *on_node)
{
CRM_CHECK(uuid || task, return NULL);
for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) {
continue;
} else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
return action;
} else if (action->node == NULL) {
continue;
} else if (on_node->details == action->node->details) {
return action;
}
}
return NULL;
}
GList *
find_actions(GList *input, const char *key, const pe_node_t *on_node)
{
GList *gIter = input;
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
for (; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) {
continue;
} else if (on_node == NULL) {
crm_trace("Action %s matches (ignoring node)", key);
result = g_list_prepend(result, action);
} else if (action->node == NULL) {
crm_trace("Action %s matches (unallocated, assigning to %s)",
key, pe__node_name(on_node));
action->node = pe__copy_node(on_node);
result = g_list_prepend(result, action);
} else if (on_node->details == action->node->details) {
crm_trace("Action %s on %s matches", key, pe__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
GList *
find_actions_exact(GList *input, const char *key, const pe_node_t *on_node)
{
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
if (on_node == NULL) {
return NULL;
}
for (GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pe_action_t *action = (pe_action_t *) gIter->data;
if ((action->node != NULL)
&& pcmk__str_eq(key, action->uuid, pcmk__str_casei)
&& pcmk__str_eq(on_node->details->id, action->node->details->id,
pcmk__str_casei)) {
crm_trace("Action %s on %s matches", key, pe__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
/*!
* \brief Find all actions of given type for a resource
*
* \param[in] rsc Resource to search
* \param[in] node Find only actions scheduled on this node
* \param[in] task Action name to search for
* \param[in] require_node If TRUE, NULL node or action node will not match
*
* \return List of actions found (or NULL if none)
* \note If node is not NULL and require_node is FALSE, matching actions
* without a node will be assigned to node.
*/
GList *
pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node,
const char *task, bool require_node)
{
GList *result = NULL;
char *key = pcmk__op_key(rsc->id, task, 0);
if (require_node) {
result = find_actions_exact(rsc->actions, key, node);
} else {
result = find_actions(rsc->actions, key, node);
}
free(key);
return result;
}
/*!
* \internal
* \brief Create an action reason string based on the action itself
*
* \param[in] action Action to create reason string for
* \param[in] flag Action flag that was cleared
*
* \return Newly allocated string suitable for use as action reason
* \note It is the caller's responsibility to free() the result.
*/
char *
pe__action2reason(const pe_action_t *action, enum pe_action_flags flag)
{
const char *change = NULL;
switch (flag) {
case pe_action_runnable:
change = "unrunnable";
break;
case pe_action_migrate_runnable:
change = "unmigrateable";
break;
case pe_action_optional:
change = "required";
break;
default:
// Bug: caller passed unsupported flag
CRM_CHECK(change != NULL, change = "");
break;
}
return crm_strdup_printf("%s%s%s %s", change,
(action->rsc == NULL)? "" : " ",
(action->rsc == NULL)? "" : action->rsc->id,
action->task);
}
void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite)
{
if (action->reason != NULL && overwrite) {
pe_rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'",
action->uuid, action->reason, pcmk__s(reason, "(none)"));
} else if (action->reason == NULL) {
pe_rsc_trace(action->rsc, "Set %s reason to '%s'",
action->uuid, pcmk__s(reason, "(none)"));
} else {
// crm_assert(action->reason != NULL && !overwrite);
return;
}
pcmk__str_update(&action->reason, reason);
}
/*!
* \internal
* \brief Create an action to clear a resource's history from CIB
*
* \param[in,out] rsc Resource to clear
* \param[in] node Node to clear history on
* \param[in,out] data_set Cluster working set
*
* \return New action to clear resource history
*/
pe_action_t *
pe__clear_resource_history(pe_resource_t *rsc, const pe_node_t *node,
pe_working_set_t *data_set)
{
char *key = NULL;
CRM_ASSERT(rsc && node);
key = pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0);
return custom_action(rsc, key, PCMK_ACTION_LRM_DELETE, node, FALSE, TRUE,
data_set);
}
#define sort_return(an_int, why) do { \
free(a_uuid); \
free(b_uuid); \
crm_trace("%s (%d) %c %s (%d) : %s", \
a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \
b_xml_id, b_call_id, why); \
return an_int; \
} while(0)
int
pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b,
bool same_node_default)
{
int a_call_id = -1;
int b_call_id = -1;
char *a_uuid = NULL;
char *b_uuid = NULL;
const char *a_xml_id = crm_element_value(xml_a, XML_ATTR_ID);
const char *b_xml_id = crm_element_value(xml_b, XML_ATTR_ID);
const char *a_node = crm_element_value(xml_a, XML_LRM_ATTR_TARGET);
const char *b_node = crm_element_value(xml_b, XML_LRM_ATTR_TARGET);
bool same_node = true;
/* @COMPAT The on_node attribute was added to last_failure as of 1.1.13 (via
* 8b3ca1c) and the other entries as of 1.1.12 (via 0b07b5c).
*
* In case that any of the lrm_rsc_op entries doesn't have on_node
* attribute, we need to explicitly tell whether the two operations are on
* the same node.
*/
if (a_node == NULL || b_node == NULL) {
same_node = same_node_default;
} else {
same_node = pcmk__str_eq(a_node, b_node, pcmk__str_casei);
}
if (same_node && pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_none)) {
/* We have duplicate lrm_rsc_op entries in the status
* section which is unlikely to be a good thing
* - we can handle it easily enough, but we need to get
* to the bottom of why it's happening.
*/
pe_err("Duplicate lrm_rsc_op entries named %s", a_xml_id);
sort_return(0, "duplicate");
}
crm_element_value_int(xml_a, XML_LRM_ATTR_CALLID, &a_call_id);
crm_element_value_int(xml_b, XML_LRM_ATTR_CALLID, &b_call_id);
if (a_call_id == -1 && b_call_id == -1) {
/* both are pending ops so it doesn't matter since
* stops are never pending
*/
sort_return(0, "pending");
} else if (same_node && a_call_id >= 0 && a_call_id < b_call_id) {
sort_return(-1, "call id");
} else if (same_node && b_call_id >= 0 && a_call_id > b_call_id) {
sort_return(1, "call id");
} else if (a_call_id >= 0 && b_call_id >= 0
&& (!same_node || a_call_id == b_call_id)) {
/*
* The op and last_failed_op are the same
* Order on last-rc-change
*/
time_t last_a = -1;
time_t last_b = -1;
crm_element_value_epoch(xml_a, XML_RSC_OP_LAST_CHANGE, &last_a);
crm_element_value_epoch(xml_b, XML_RSC_OP_LAST_CHANGE, &last_b);
crm_trace("rc-change: %lld vs %lld",
(long long) last_a, (long long) last_b);
if (last_a >= 0 && last_a < last_b) {
sort_return(-1, "rc-change");
} else if (last_b >= 0 && last_a > last_b) {
sort_return(1, "rc-change");
}
sort_return(0, "rc-change");
} else {
/* One of the inputs is a pending operation
* Attempt to use XML_ATTR_TRANSITION_MAGIC to determine its age relative to the other
*/
int a_id = -1;
int b_id = -1;
const char *a_magic = crm_element_value(xml_a, XML_ATTR_TRANSITION_MAGIC);
const char *b_magic = crm_element_value(xml_b, XML_ATTR_TRANSITION_MAGIC);
CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic"));
if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic a");
}
if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL,
NULL)) {
sort_return(0, "bad magic b");
}
/* try to determine the relative age of the operation...
* some pending operations (e.g. a start) may have been superseded
* by a subsequent stop
*
* [a|b]_id == -1 means it's a shutdown operation and _always_ comes last
*/
if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) {
/*
* some of the logic in here may be redundant...
*
* if the UUID from the TE doesn't match then one better
* be a pending operation.
* pending operations don't survive between elections and joins
* because we query the LRM directly
*/
if (b_call_id == -1) {
sort_return(-1, "transition + call");
} else if (a_call_id == -1) {
sort_return(1, "transition + call");
}
} else if ((a_id >= 0 && a_id < b_id) || b_id == -1) {
sort_return(-1, "transition");
} else if ((b_id >= 0 && a_id > b_id) || a_id == -1) {
sort_return(1, "transition");
}
}
/* we should never end up here */
CRM_CHECK(FALSE, sort_return(0, "default"));
}
gint
sort_op_by_callid(gconstpointer a, gconstpointer b)
{
const xmlNode *xml_a = a;
const xmlNode *xml_b = b;
return pe__is_newer_op(xml_a, xml_b, true);
}
/*!
* \internal
* \brief Create a new pseudo-action for a resource
*
* \param[in,out] rsc Resource to create action for
* \param[in] task Action name
* \param[in] optional Whether action should be considered optional
* \param[in] runnable Whethe action should be considered runnable
*
* \return New action object corresponding to arguments
*/
pe_action_t *
pe__new_rsc_pseudo_action(pe_resource_t *rsc, const char *task, bool optional,
bool runnable)
{
pe_action_t *action = NULL;
CRM_ASSERT((rsc != NULL) && (task != NULL));
action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL,
optional, TRUE, rsc->cluster);
- pe__set_action_flags(action, pe_action_pseudo);
+ pe__set_action_flags(action, pcmk_action_pseudo);
if (runnable) {
pe__set_action_flags(action, pe_action_runnable);
}
return action;
}
/*!
* \internal
* \brief Add the expected result to an action
*
* \param[in,out] action Action to add expected result to
* \param[in] expected_result Expected result to add
*
* \note This is more efficient than calling add_hash_param().
*/
void
pe__add_action_expected_result(pe_action_t *action, int expected_result)
{
char *name = NULL;
CRM_ASSERT((action != NULL) && (action->meta != NULL));
name = strdup(XML_ATTR_TE_TARGET_RC);
CRM_ASSERT (name != NULL);
g_hash_table_insert(action->meta, name, pcmk__itoa(expected_result));
}
diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c
index 5e0c1c8649..50910cbdc6 100644
--- a/lib/pengine/pe_notif.c
+++ b/lib/pengine/pe_notif.c
@@ -1,1002 +1,1003 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/msg_xml.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include "pe_status_private.h"
typedef struct notify_entry_s {
const pe_resource_t *rsc;
const pe_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 name) 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 names
return strcmp(entry_a->node->details->id, entry_b->node->details->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 = calloc(1, sizeof(notify_entry_t));
CRM_ASSERT(dup != NULL);
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 pe_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 pe_node_t *node = (const pe_node_t *) iter->data;
if (node->details->uname == NULL) {
continue;
}
// Always add to list of all node names
if (all_node_names != NULL) {
pcmk__add_word(all_node_names, 1024, node->details->uname);
}
// Add to host node name list if appropriate
if (host_node_names != NULL) {
if (pe__is_guest_node(node)
&& (node->details->remote_rsc->container->running_on != NULL)) {
node = pe__current_node(node->details->remote_rsc->container);
if (node->details->uname == NULL) {
continue;
}
}
pcmk__add_word(host_node_names, 1024, node->details->uname);
}
}
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->details->uname != NULL)) {
pcmk__add_word(node_names, 1024, entry->node->details->uname);
}
}
// 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)
{
pe_action_t *notify = (pe_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;
}
g_hash_table_insert(notify->meta, strdup((const char *) key),
strdup((const char *) value));
}
static void
add_notify_data_to_action_meta(const notify_data_t *n_data, pe_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;
add_hash_param(action->meta, 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 pe_action_t *
new_notify_pseudo_action(pe_resource_t *rsc, const pe_action_t *action,
const char *notif_action, const char *notif_type)
{
pe_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, pe_action_optional),
TRUE, rsc->cluster);
- pe__set_action_flags(notify, pe_action_pseudo);
+ pe__set_action_flags(notify, pcmk_action_pseudo);
add_hash_param(notify->meta, "notify_key_type", notif_type);
add_hash_param(notify->meta, "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 pe_action_t *
new_notify_action(pe_resource_t *rsc, const pe_node_t *node, pe_action_t *op,
pe_action_t *notify_done, const notify_data_t *n_data)
{
char *key = NULL;
pe_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, pe_action_runnable)) {
skip_reason = "original action not runnable";
}
if (skip_reason != NULL) {
pe_rsc_trace(rsc, "Skipping notify action for %s on %s: %s",
rsc->id, pe__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
pe_rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)",
rsc->id, pe__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, pe_action_optional),
TRUE, rsc->cluster);
// 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, pe_order_optional);
order_actions(notify_action, notify_done, pe_order_optional);
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(pe_resource_t *rsc, const pe_node_t *node,
notify_data_t *n_data)
{
pe_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 = INFINITY;
}
// Order recurring monitors after all "post-" notifications complete
if (n_data->post_done == NULL) {
return;
}
for (GList *iter = rsc->actions; iter != NULL; iter = iter->next) {
pe_action_t *mon = (pe_action_t *) iter->data;
const char *interval_ms_s = NULL;
interval_ms_s = g_hash_table_lookup(mon->meta,
XML_LRM_ATTR_INTERVAL_MS);
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, pe_order_optional);
}
}
/*!
* \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(pe_resource_t *rsc, const char *task,
pe_action_t *action, pe_action_t *complete)
{
notify_data_t *n_data = NULL;
if (!pcmk_is_set(rsc->flags, pcmk_rsc_notify)) {
return NULL;
}
n_data = calloc(1, sizeof(notify_data_t));
CRM_ASSERT(n_data != NULL);
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");
pe__set_action_flags(n_data->pre, pe_action_runnable);
add_hash_param(n_data->pre->meta, "notify_type", "pre");
add_hash_param(n_data->pre->meta, "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");
pe__set_action_flags(n_data->pre_done, pe_action_runnable);
add_hash_param(n_data->pre_done->meta, "notify_type", "pre");
add_hash_param(n_data->pre_done->meta,
"notify_operation", n_data->action);
// Order "pre-" -> "pre-" complete -> original action
order_actions(n_data->pre, n_data->pre_done, pe_order_optional);
order_actions(n_data->pre_done, action, pe_order_optional);
}
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 = INFINITY;
if (pcmk_is_set(complete->flags, pe_action_runnable)) {
pe__set_action_flags(n_data->post, pe_action_runnable);
} else {
pe__clear_action_flags(n_data->post, pe_action_runnable);
}
add_hash_param(n_data->post->meta, "notify_type", "post");
add_hash_param(n_data->post->meta, "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 = INFINITY;
if (pcmk_is_set(complete->flags, pe_action_runnable)) {
pe__set_action_flags(n_data->post_done, pe_action_runnable);
} else {
pe__clear_action_flags(n_data->post_done, pe_action_runnable);
}
add_hash_param(n_data->post_done->meta, "notify_type", "post");
add_hash_param(n_data->post_done->meta,
"notify_operation", n_data->action);
// Order original action complete -> "post-" -> "post-" complete
order_actions(complete, n_data->post, pe_order_implies_then);
order_actions(n_data->post, n_data->post_done, pe_order_implies_then);
}
// If we created both, order "pre-" complete -> "post-"
if ((action != NULL) && (complete != NULL)) {
order_actions(n_data->pre_done, n_data->post, pe_order_optional);
}
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 pe_resource_t *rsc, const pe_node_t *node)
{
notify_entry_t *entry = calloc(1, sizeof(notify_entry_t));
CRM_ASSERT(entry != NULL);
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] activity Whether to add notification entries for actions
* \param[in,out] n_data Notification data for clone
*/
static void
collect_resource_data(const pe_resource_t *rsc, bool activity,
notify_data_t *n_data)
{
const GList *iter = NULL;
notify_entry_t *entry = NULL;
const pe_node_t *node = NULL;
if (n_data == NULL) {
return;
}
if (n_data->allowed_nodes == NULL) {
n_data->allowed_nodes = rsc->allowed_nodes;
}
// If this is a clone, call recursively for each instance
if (rsc->children != NULL) {
for (iter = rsc->children; iter != NULL; iter = iter->next) {
const pe_resource_t *child = (const pe_resource_t *) iter->data;
collect_resource_data(child, activity, n_data);
}
return;
}
// This is a notification for a single clone instance
if (rsc->running_on != NULL) {
node = rsc->running_on->data; // First is sufficient
}
entry = new_notify_entry(rsc, node);
// Add notification indicating the resource state
switch (rsc->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:
crm_err("Resource %s role on %s (%s) is not supported for "
"notifications (bug?)",
rsc->id, pe__node_name(node), role2text(rsc->role));
free(entry);
break;
}
if (!activity) {
return;
}
// Add notification entries for each of the resource's actions
for (iter = rsc->actions; iter != NULL; iter = iter->next) {
const pe_action_t *op = (const pe_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pe_action_optional) && (op->node != NULL)) {
enum action_tasks task = text2task(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, pe_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 pe_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->meta, XML_RSC_ATTR_TARGET);
if (pcmk__str_eq("host", source, pcmk__str_none)) {
get_node_names(rsc->cluster->nodes, &node_list, &metal_list);
add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list);
} else {
get_node_names(rsc->cluster->nodes, &node_list, NULL);
}
add_notify_env_free_gs(n_data, "notify_all_uname", node_list);
if (required && (n_data->pre != NULL)) {
pe__clear_action_flags(n_data->pre, pe_action_optional);
pe__clear_action_flags(n_data->pre_done, pe_action_optional);
}
if (required && (n_data->post != NULL)) {
pe__clear_action_flags(n_data->post, pe_action_optional);
pe__clear_action_flags(n_data->post_done, pe_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 pe_action_t *
find_remote_start(pe_action_t *action)
{
if ((action != NULL) && (action->node != NULL)) {
pe_resource_t *remote_rsc = action->node->details->remote_rsc;
if (remote_rsc != NULL) {
return find_first_action(remote_rsc->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(pe_resource_t *rsc, notify_data_t *n_data)
{
GList *iter = NULL;
pe_action_t *stop = NULL;
pe_action_t *start = NULL;
enum action_tasks task = text2task(n_data->action);
// If this is a clone, call recursively for each instance
if (rsc->children != NULL) {
g_list_foreach(rsc->children, (GFunc) create_notify_actions, n_data);
return;
}
// Add notification meta-attributes to original actions
for (iter = rsc->actions; iter != NULL; iter = iter->next) {
pe_action_t *op = (pe_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pe_action_optional) && (op->node != NULL)) {
switch (text2task(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) {
pe_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) {
pe_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) {
pe_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;
}
pe_rsc_trace(rsc, "Creating notify actions for %s %s",
rsc->id, n_data->action);
// Create notify actions for stop or demote
if ((rsc->role != pcmk_role_stopped)
&& ((task == pcmk_action_stop) || (task == pcmk_action_demote))) {
stop = find_first_action(rsc->actions, NULL, PCMK_ACTION_STOP, NULL);
for (iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *current_node = (pe_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, pe_action_pseudo)
+ if ((stop != NULL)
+ && pcmk_is_set(stop->flags, pcmk_action_pseudo)
&& (current_node->details->unclean
|| current_node->details->remote_requires_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, pe_action_optional)) {
new_post_notify_action(rsc, current_node, n_data);
}
}
}
// Create notify actions for start or promote
if ((rsc->next_role != pcmk_role_stopped)
&& ((task == pcmk_action_start) || (task == pcmk_action_promote))) {
start = find_first_action(rsc->actions, NULL, PCMK_ACTION_START, NULL);
if (start != NULL) {
pe_action_t *remote_start = find_remote_start(start);
if ((remote_start != NULL)
&& !pcmk_is_set(remote_start->flags, pe_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->allocated_to == NULL) {
pe_proc_err("Next role '%s' but %s is not allocated",
role2text(rsc->next_role), rsc->id);
return;
}
if ((task != pcmk_action_start) || (start == NULL)
|| pcmk_is_set(start->flags, pe_action_optional)) {
new_notify_action(rsc, rsc->allocated_to, n_data->pre,
n_data->pre_done, n_data);
}
new_post_notify_action(rsc, rsc->allocated_to, 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(pe_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 pe_action_t *stop, pe_resource_t *rsc,
pe_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->details->uname);
create_notify_actions(uber_parent(rsc), n_data);
pe__free_action_notification_data(n_data);
}
}
diff --git a/tools/crm_resource_runtime.c b/tools/crm_resource_runtime.c
index 9c20eb48aa..348e7677af 100644
--- a/tools/crm_resource_runtime.c
+++ b/tools/crm_resource_runtime.c
@@ -1,2223 +1,2223 @@
/*
* Copyright 2004-2023 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm_resource.h>
#include <crm/common/ipc_attrd_internal.h>
#include <crm/common/ipc_controld.h>
#include <crm/common/lists_internal.h>
#include <crm/services_internal.h>
static GList *
build_node_info_list(const pe_resource_t *rsc)
{
GList *retval = NULL;
for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) {
const pe_resource_t *child = (const pe_resource_t *) iter->data;
for (const GList *iter2 = child->running_on;
iter2 != NULL; iter2 = iter2->next) {
const pe_node_t *node = (const pe_node_t *) iter2->data;
node_info_t *ni = calloc(1, sizeof(node_info_t));
ni->node_name = node->details->uname;
ni->promoted = pcmk_is_set(rsc->flags, pcmk_rsc_promotable) &&
child->fns->state(child, TRUE) == pcmk_role_promoted;
retval = g_list_prepend(retval, ni);
}
}
return retval;
}
GList *
cli_resource_search(pe_resource_t *rsc, const char *requested_name,
pe_working_set_t *data_set)
{
GList *retval = NULL;
const pe_resource_t *parent = pe__const_top_resource(rsc, false);
if (pe_rsc_is_clone(rsc)) {
retval = build_node_info_list(rsc);
/* The anonymous clone children's common ID is supplied */
} else if (pe_rsc_is_clone(parent)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_unique)
&& rsc->clone_name
&& pcmk__str_eq(requested_name, rsc->clone_name, pcmk__str_casei)
&& !pcmk__str_eq(requested_name, rsc->id, pcmk__str_casei)) {
retval = build_node_info_list(parent);
} else if (rsc->running_on != NULL) {
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
node_info_t *ni = calloc(1, sizeof(node_info_t));
ni->node_name = node->details->uname;
ni->promoted = (rsc->fns->state(rsc, TRUE) == pcmk_role_promoted);
retval = g_list_prepend(retval, ni);
}
}
return retval;
}
// \return Standard Pacemaker return code
static int
find_resource_attr(pcmk__output_t *out, cib_t * the_cib, const char *attr,
const char *rsc, const char *attr_set_type, const char *set_name,
const char *attr_id, const char *attr_name, char **value)
{
int rc = pcmk_rc_ok;
xmlNode *xml_search = NULL;
GString *xpath = NULL;
const char *xpath_base = NULL;
if(value) {
*value = NULL;
}
if(the_cib == NULL) {
return ENOTCONN;
}
xpath_base = pcmk_cib_xpath_for(XML_CIB_TAG_RESOURCES);
if (xpath_base == NULL) {
crm_err(XML_CIB_TAG_RESOURCES " CIB element not known (bug?)");
return ENOMSG;
}
xpath = g_string_sized_new(1024);
pcmk__g_strcat(xpath,
xpath_base, "//*[@" XML_ATTR_ID "=\"", rsc, "\"]", NULL);
if (attr_set_type != NULL) {
pcmk__g_strcat(xpath, "/", attr_set_type, NULL);
if (set_name != NULL) {
pcmk__g_strcat(xpath, "[@" XML_ATTR_ID "=\"", set_name, "\"]",
NULL);
}
}
g_string_append(xpath, "//" XML_CIB_TAG_NVPAIR "[");
if (attr_id != NULL) {
pcmk__g_strcat(xpath, "@" XML_ATTR_ID "=\"", attr_id, "\"", NULL);
}
if (attr_name != NULL) {
if (attr_id != NULL) {
g_string_append(xpath, " and ");
}
pcmk__g_strcat(xpath, "@" XML_NVPAIR_ATTR_NAME "=\"", attr_name, "\"",
NULL);
}
g_string_append_c(xpath, ']');
rc = the_cib->cmds->query(the_cib, (const char *) xpath->str, &xml_search,
cib_sync_call | cib_scope_local | cib_xpath);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
goto done;
}
crm_log_xml_debug(xml_search, "Match");
if (xml_search->children != NULL) {
xmlNode *child = NULL;
rc = ENOTUNIQ;
out->info(out, "Multiple attributes match name=%s", attr_name);
for (child = pcmk__xml_first_child(xml_search); child != NULL;
child = pcmk__xml_next(child)) {
out->info(out, " Value: %s \t(id=%s)",
crm_element_value(child, XML_NVPAIR_ATTR_VALUE), ID(child));
}
out->spacer(out);
} else if(value) {
pcmk__str_update(value, crm_element_value(xml_search, attr));
}
done:
g_string_free(xpath, TRUE);
free_xml(xml_search);
return rc;
}
/* PRIVATE. Use the find_matching_attr_resources instead. */
static void
find_matching_attr_resources_recursive(pcmk__output_t *out, GList/* <pe_resource_t*> */ ** result,
pe_resource_t * rsc, const char * rsc_id,
const char * attr_set, const char * attr_set_type,
const char * attr_id, const char * attr_name,
cib_t * cib, const char * cmd, int depth)
{
int rc = pcmk_rc_ok;
char *lookup_id = clone_strip(rsc->id);
char *local_attr_id = NULL;
/* visit the children */
for(GList *gIter = rsc->children; gIter; gIter = gIter->next) {
find_matching_attr_resources_recursive(out, result, (pe_resource_t*)gIter->data,
rsc_id, attr_set, attr_set_type,
attr_id, attr_name, cib, cmd, depth+1);
/* do it only once for clones */
if (rsc->variant == pcmk_rsc_variant_clone) {
break;
}
}
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
/* Post-order traversal.
* The root is always on the list and it is the last item. */
if((0 == depth) || (pcmk_rc_ok == rc)) {
/* push the head */
*result = g_list_append(*result, rsc);
}
free(local_attr_id);
free(lookup_id);
}
/* The result is a linearized pre-ordered tree of resources. */
static GList/*<pe_resource_t*>*/ *
find_matching_attr_resources(pcmk__output_t *out, pe_resource_t * rsc,
const char * rsc_id, const char * attr_set,
const char * attr_set_type, const char * attr_id,
const char * attr_name, cib_t * cib, const char * cmd,
gboolean force)
{
int rc = pcmk_rc_ok;
char *lookup_id = NULL;
char *local_attr_id = NULL;
GList * result = NULL;
/* If --force is used, update only the requested resource (clone or primitive).
* Otherwise, if the primitive has the attribute, use that.
* Otherwise use the clone. */
if(force == TRUE) {
return g_list_append(result, rsc);
}
if ((rsc->parent != NULL)
&& (rsc->parent->variant == pcmk_rsc_variant_clone)) {
int rc = pcmk_rc_ok;
char *local_attr_id = NULL;
rc = find_resource_attr(out, cib, XML_ATTR_ID, rsc_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
free(local_attr_id);
if(rc != pcmk_rc_ok) {
rsc = rsc->parent;
out->info(out, "Performing %s of '%s' on '%s', the parent of '%s'",
cmd, attr_name, rsc->id, rsc_id);
}
return g_list_append(result, rsc);
} else if ((rsc->parent == NULL) && (rsc->children != NULL)
&& (rsc->variant == pcmk_rsc_variant_clone)) {
pe_resource_t *child = rsc->children->data;
if (child->variant == pcmk_rsc_variant_primitive) {
lookup_id = clone_strip(child->id); /* Could be a cloned group! */
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
if(rc == pcmk_rc_ok) {
rsc = child;
out->info(out, "A value for '%s' already exists in child '%s', performing %s on that instead of '%s'",
attr_name, lookup_id, cmd, rsc_id);
}
free(local_attr_id);
free(lookup_id);
}
return g_list_append(result, rsc);
}
/* If the resource is a group ==> children inherit the attribute if defined. */
find_matching_attr_resources_recursive(out, &result, rsc, rsc_id, attr_set,
attr_set_type, attr_id, attr_name,
cib, cmd, 0);
return result;
}
// \return Standard Pacemaker return code
int
cli_resource_update_attribute(pe_resource_t *rsc, const char *requested_name,
const char *attr_set, const char *attr_set_type,
const char *attr_id, const char *attr_name,
const char *attr_value, gboolean recursive,
cib_t *cib, int cib_options, gboolean force)
{
pcmk__output_t *out = rsc->cluster->priv;
int rc = pcmk_rc_ok;
char *found_attr_id = NULL;
GList/*<pe_resource_t*>*/ *resources = NULL;
const char *top_id = pe__const_top_resource(rsc, false)->id;
if ((attr_id == NULL) && !force) {
find_resource_attr(out, cib, XML_ATTR_ID, top_id, NULL, NULL, NULL,
attr_name, NULL);
}
if (pcmk__str_eq(attr_set_type, XML_TAG_ATTR_SETS, pcmk__str_casei)) {
if (!force) {
rc = find_resource_attr(out, cib, XML_ATTR_ID, top_id,
XML_TAG_META_SETS, attr_set, attr_id,
attr_name, &found_attr_id);
if ((rc == pcmk_rc_ok) && !out->is_quiet(out)) {
out->err(out,
"WARNING: There is already a meta attribute "
"for '%s' called '%s' (id=%s)",
top_id, attr_name, found_attr_id);
out->err(out,
" Delete '%s' first or use the force option "
"to override", found_attr_id);
}
free(found_attr_id);
if (rc == pcmk_rc_ok) {
return ENOTUNIQ;
}
}
resources = g_list_append(resources, rsc);
} else if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) {
crm_xml_add(rsc->xml, attr_name, attr_value);
CRM_ASSERT(cib != NULL);
rc = cib->cmds->replace(cib, XML_CIB_TAG_RESOURCES, rsc->xml,
cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Set attribute: name=%s value=%s",
attr_name, attr_value);
}
return rc;
} else {
resources = find_matching_attr_resources(out, rsc, requested_name,
attr_set, attr_set_type,
attr_id, attr_name, cib,
"update", force);
}
/* If the user specified attr_set or attr_id, the intent is to modify a
* single resource, which will be the last item in the list.
*/
if ((attr_set != NULL) || (attr_id != NULL)) {
GList *last = g_list_last(resources);
resources = g_list_remove_link(resources, last);
g_list_free(resources);
resources = last;
}
for (GList *iter = resources; iter != NULL; iter = iter->next) {
char *lookup_id = NULL;
char *local_attr_set = NULL;
const char *rsc_attr_id = attr_id;
const char *rsc_attr_set = attr_set;
xmlNode *xml_top = NULL;
xmlNode *xml_obj = NULL;
found_attr_id = NULL;
rsc = (pe_resource_t *) iter->data;
lookup_id = clone_strip(rsc->id); /* Could be a cloned group! */
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &found_attr_id);
switch (rc) {
case pcmk_rc_ok:
crm_debug("Found a match for name=%s: id=%s",
attr_name, found_attr_id);
rsc_attr_id = found_attr_id;
break;
case ENXIO:
if (rsc_attr_set == NULL) {
local_attr_set = crm_strdup_printf("%s-%s", lookup_id,
attr_set_type);
rsc_attr_set = local_attr_set;
}
if (rsc_attr_id == NULL) {
found_attr_id = crm_strdup_printf("%s-%s",
rsc_attr_set, attr_name);
rsc_attr_id = found_attr_id;
}
xml_top = create_xml_node(NULL, (const char *) rsc->xml->name);
crm_xml_add(xml_top, XML_ATTR_ID, lookup_id);
xml_obj = create_xml_node(xml_top, attr_set_type);
crm_xml_add(xml_obj, XML_ATTR_ID, rsc_attr_set);
break;
default:
free(lookup_id);
free(found_attr_id);
g_list_free(resources);
return rc;
}
xml_obj = crm_create_nvpair_xml(xml_obj, rsc_attr_id, attr_name,
attr_value);
if (xml_top == NULL) {
xml_top = xml_obj;
}
crm_log_xml_debug(xml_top, "Update");
rc = cib->cmds->modify(cib, XML_CIB_TAG_RESOURCES, xml_top,
cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Set '%s' option: id=%s%s%s%s%s value=%s",
lookup_id, found_attr_id,
((rsc_attr_set == NULL)? "" : " set="),
pcmk__s(rsc_attr_set, ""),
((attr_name == NULL)? "" : " name="),
pcmk__s(attr_name, ""), attr_value);
}
free_xml(xml_top);
free(lookup_id);
free(found_attr_id);
free(local_attr_set);
if (recursive
&& pcmk__str_eq(attr_set_type, XML_TAG_META_SETS,
pcmk__str_casei)) {
GList *lpc = NULL;
static bool need_init = true;
if (need_init) {
need_init = false;
pcmk__unpack_constraints(rsc->cluster);
pe__clear_resource_flags_on_all(rsc->cluster,
pcmk_rsc_detect_loop);
}
/* We want to set the attribute only on resources explicitly
* colocated with this one, so we use rsc->rsc_cons_lhs directly
* rather than the with_this_colocations() method.
*/
pe__set_resource_flags(rsc, pcmk_rsc_detect_loop);
for (lpc = rsc->rsc_cons_lhs; lpc != NULL; lpc = lpc->next) {
pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data;
crm_debug("Checking %s %d", cons->id, cons->score);
if (!pcmk_is_set(cons->dependent->flags, pcmk_rsc_detect_loop)
&& (cons->score > 0)) {
crm_debug("Setting %s=%s for dependent resource %s",
attr_name, attr_value, cons->dependent->id);
cli_resource_update_attribute(cons->dependent,
cons->dependent->id, NULL,
attr_set_type, NULL,
attr_name, attr_value,
recursive, cib, cib_options,
force);
}
}
}
}
g_list_free(resources);
return rc;
}
// \return Standard Pacemaker return code
int
cli_resource_delete_attribute(pe_resource_t *rsc, const char *requested_name,
const char *attr_set, const char *attr_set_type,
const char *attr_id, const char *attr_name,
cib_t *cib, int cib_options, gboolean force)
{
pcmk__output_t *out = rsc->cluster->priv;
int rc = pcmk_rc_ok;
GList/*<pe_resource_t*>*/ *resources = NULL;
if ((attr_id == NULL) && !force) {
find_resource_attr(out, cib, XML_ATTR_ID,
pe__const_top_resource(rsc, false)->id, NULL,
NULL, NULL, attr_name, NULL);
}
if (pcmk__str_eq(attr_set_type, XML_TAG_META_SETS, pcmk__str_casei)) {
resources = find_matching_attr_resources(out, rsc, requested_name,
attr_set, attr_set_type,
attr_id, attr_name, cib,
"delete", force);
} else if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) {
xml_remove_prop(rsc->xml, attr_name);
CRM_ASSERT(cib != NULL);
rc = cib->cmds->replace(cib, XML_CIB_TAG_RESOURCES, rsc->xml,
cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Deleted attribute: %s", attr_name);
}
return rc;
} else {
resources = g_list_append(resources, rsc);
}
for (GList *iter = resources; iter != NULL; iter = iter->next) {
char *lookup_id = NULL;
xmlNode *xml_obj = NULL;
char *found_attr_id = NULL;
const char *rsc_attr_id = attr_id;
rsc = (pe_resource_t *) iter->data;
lookup_id = clone_strip(rsc->id);
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &found_attr_id);
switch (rc) {
case pcmk_rc_ok:
break;
case ENXIO:
free(lookup_id);
rc = pcmk_rc_ok;
continue;
default:
free(lookup_id);
g_list_free(resources);
return rc;
}
if (rsc_attr_id == NULL) {
rsc_attr_id = found_attr_id;
}
xml_obj = crm_create_nvpair_xml(NULL, rsc_attr_id, attr_name, NULL);
crm_log_xml_debug(xml_obj, "Delete");
CRM_ASSERT(cib);
rc = cib->cmds->remove(cib, XML_CIB_TAG_RESOURCES, xml_obj,
cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Deleted '%s' option: id=%s%s%s%s%s",
lookup_id, found_attr_id,
((attr_set == NULL)? "" : " set="),
pcmk__s(attr_set, ""),
((attr_name == NULL)? "" : " name="),
pcmk__s(attr_name, ""));
}
free(lookup_id);
free_xml(xml_obj);
free(found_attr_id);
}
g_list_free(resources);
return rc;
}
// \return Standard Pacemaker return code
static int
send_lrm_rsc_op(pcmk_ipc_api_t *controld_api, bool do_fail_resource,
const char *host_uname, const char *rsc_id, pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
const char *router_node = host_uname;
const char *rsc_api_id = NULL;
const char *rsc_long_id = NULL;
const char *rsc_class = NULL;
const char *rsc_provider = NULL;
const char *rsc_type = NULL;
bool cib_only = false;
pe_resource_t *rsc = pe_find_resource(data_set->resources, rsc_id);
if (rsc == NULL) {
out->err(out, "Resource %s not found", rsc_id);
return ENXIO;
} else if (rsc->variant != pcmk_rsc_variant_primitive) {
out->err(out, "We can only process primitive resources, not %s", rsc_id);
return EINVAL;
}
rsc_class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
rsc_provider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER),
rsc_type = crm_element_value(rsc->xml, XML_ATTR_TYPE);
if ((rsc_class == NULL) || (rsc_type == NULL)) {
out->err(out, "Resource %s does not have a class and type", rsc_id);
return EINVAL;
}
{
pe_node_t *node = pe_find_node(data_set->nodes, host_uname);
if (node == NULL) {
out->err(out, "Node %s not found", host_uname);
return pcmk_rc_node_unknown;
}
if (!(node->details->online)) {
if (do_fail_resource) {
out->err(out, "Node %s is not online", host_uname);
return ENOTCONN;
} else {
cib_only = true;
}
}
if (!cib_only && pe__is_guest_or_remote_node(node)) {
node = pe__current_node(node->details->remote_rsc);
if (node == NULL) {
out->err(out, "No cluster connection to Pacemaker Remote node %s detected",
host_uname);
return ENOTCONN;
}
router_node = node->details->uname;
}
}
if (rsc->clone_name) {
rsc_api_id = rsc->clone_name;
rsc_long_id = rsc->id;
} else {
rsc_api_id = rsc->id;
}
if (do_fail_resource) {
return pcmk_controld_api_fail(controld_api, host_uname, router_node,
rsc_api_id, rsc_long_id,
rsc_class, rsc_provider, rsc_type);
} else {
return pcmk_controld_api_refresh(controld_api, host_uname, router_node,
rsc_api_id, rsc_long_id, rsc_class,
rsc_provider, rsc_type, cib_only);
}
}
/*!
* \internal
* \brief Get resource name as used in failure-related node attributes
*
* \param[in] rsc Resource to check
*
* \return Newly allocated string containing resource's fail name
* \note The caller is responsible for freeing the result.
*/
static inline char *
rsc_fail_name(const pe_resource_t *rsc)
{
const char *name = (rsc->clone_name? rsc->clone_name : rsc->id);
if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) {
return strdup(name);
}
return clone_strip(name);
}
// \return Standard Pacemaker return code
static int
clear_rsc_history(pcmk_ipc_api_t *controld_api, const char *host_uname,
const char *rsc_id, pe_working_set_t *data_set)
{
int rc = pcmk_rc_ok;
/* Erase the resource's entire LRM history in the CIB, even if we're only
* clearing a single operation's fail count. If we erased only entries for a
* single operation, we might wind up with a wrong idea of the current
* resource state, and we might not re-probe the resource.
*/
rc = send_lrm_rsc_op(controld_api, false, host_uname, rsc_id, data_set);
if (rc != pcmk_rc_ok) {
return rc;
}
crm_trace("Processing %d mainloop inputs",
pcmk_controld_api_replies_expected(controld_api));
while (g_main_context_iteration(NULL, FALSE)) {
crm_trace("Processed mainloop input, %d still remaining",
pcmk_controld_api_replies_expected(controld_api));
}
return rc;
}
// \return Standard Pacemaker return code
static int
clear_rsc_failures(pcmk__output_t *out, pcmk_ipc_api_t *controld_api,
const char *node_name, const char *rsc_id, const char *operation,
const char *interval_spec, pe_working_set_t *data_set)
{
int rc = pcmk_rc_ok;
const char *failed_value = NULL;
const char *failed_id = NULL;
const char *interval_ms_s = NULL;
GHashTable *rscs = NULL;
GHashTableIter iter;
/* Create a hash table to use as a set of resources to clean. This lets us
* clean each resource only once (per node) regardless of how many failed
* operations it has.
*/
rscs = pcmk__strkey_table(NULL, NULL);
// Normalize interval to milliseconds for comparison to history entry
if (operation) {
interval_ms_s = crm_strdup_printf("%u",
crm_parse_interval_spec(interval_spec));
}
for (xmlNode *xml_op = pcmk__xml_first_child(data_set->failed);
xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
failed_id = crm_element_value(xml_op, XML_LRM_ATTR_RSCID);
if (failed_id == NULL) {
// Malformed history entry, should never happen
continue;
}
// No resource specified means all resources match
if (rsc_id) {
pe_resource_t *fail_rsc = NULL;
fail_rsc = pe_find_resource_with_flags(data_set->resources,
failed_id,
pcmk_rsc_match_history
|pcmk_rsc_match_anon_basename);
if (!fail_rsc || !pcmk__str_eq(rsc_id, fail_rsc->id, pcmk__str_casei)) {
continue;
}
}
// Host name should always have been provided by this point
failed_value = crm_element_value(xml_op, XML_ATTR_UNAME);
if (!pcmk__str_eq(node_name, failed_value, pcmk__str_casei)) {
continue;
}
// No operation specified means all operations match
if (operation) {
failed_value = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
if (!pcmk__str_eq(operation, failed_value, pcmk__str_casei)) {
continue;
}
// Interval (if operation was specified) defaults to 0 (not all)
failed_value = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS);
if (!pcmk__str_eq(interval_ms_s, failed_value, pcmk__str_casei)) {
continue;
}
}
g_hash_table_add(rscs, (gpointer) failed_id);
}
g_hash_table_iter_init(&iter, rscs);
while (g_hash_table_iter_next(&iter, (gpointer *) &failed_id, NULL)) {
crm_debug("Erasing failures of %s on %s", failed_id, node_name);
rc = clear_rsc_history(controld_api, node_name, failed_id, data_set);
if (rc != pcmk_rc_ok) {
return rc;
}
}
g_hash_table_destroy(rscs);
return rc;
}
// \return Standard Pacemaker return code
static int
clear_rsc_fail_attrs(const pe_resource_t *rsc, const char *operation,
const char *interval_spec, const pe_node_t *node)
{
int rc = pcmk_rc_ok;
int attr_options = pcmk__node_attr_none;
char *rsc_name = rsc_fail_name(rsc);
if (pe__is_guest_or_remote_node(node)) {
attr_options |= pcmk__node_attr_remote;
}
rc = pcmk__attrd_api_clear_failures(NULL, node->details->uname, rsc_name,
operation, interval_spec, NULL,
attr_options);
free(rsc_name);
return rc;
}
// \return Standard Pacemaker return code
int
cli_resource_delete(pcmk_ipc_api_t *controld_api, const char *host_uname,
const pe_resource_t *rsc, const char *operation,
const char *interval_spec, bool just_failures,
pe_working_set_t *data_set, gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
pe_node_t *node = NULL;
if (rsc == NULL) {
return ENXIO;
} else if (rsc->children) {
for (const GList *lpc = rsc->children; lpc != NULL; lpc = lpc->next) {
const pe_resource_t *child = (const pe_resource_t *) lpc->data;
rc = cli_resource_delete(controld_api, host_uname, child, operation,
interval_spec, just_failures, data_set, force);
if (rc != pcmk_rc_ok) {
return rc;
}
}
return pcmk_rc_ok;
} else if (host_uname == NULL) {
GList *lpc = NULL;
GList *nodes = g_hash_table_get_values(rsc->known_on);
if(nodes == NULL && force) {
nodes = pcmk__copy_node_list(data_set->nodes, false);
} else if(nodes == NULL && rsc->exclusive_discover) {
GHashTableIter iter;
pe_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void**)&node)) {
if(node->weight >= 0) {
nodes = g_list_prepend(nodes, node);
}
}
} else if(nodes == NULL) {
nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
for (lpc = nodes; lpc != NULL; lpc = lpc->next) {
node = (pe_node_t *) lpc->data;
if (node->details->online) {
rc = cli_resource_delete(controld_api, node->details->uname, rsc,
operation, interval_spec, just_failures,
data_set, force);
}
if (rc != pcmk_rc_ok) {
g_list_free(nodes);
return rc;
}
}
g_list_free(nodes);
return pcmk_rc_ok;
}
node = pe_find_node(data_set->nodes, host_uname);
if (node == NULL) {
out->err(out, "Unable to clean up %s because node %s not found",
rsc->id, host_uname);
return ENODEV;
}
if (!node->details->rsc_discovery_enabled) {
out->err(out, "Unable to clean up %s because resource discovery disabled on %s",
rsc->id, host_uname);
return EOPNOTSUPP;
}
if (controld_api == NULL) {
out->err(out, "Dry run: skipping clean-up of %s on %s due to CIB_file",
rsc->id, host_uname);
return pcmk_rc_ok;
}
rc = clear_rsc_fail_attrs(rsc, operation, interval_spec, node);
if (rc != pcmk_rc_ok) {
out->err(out, "Unable to clean up %s failures on %s: %s",
rsc->id, host_uname, pcmk_rc_str(rc));
return rc;
}
if (just_failures) {
rc = clear_rsc_failures(out, controld_api, host_uname, rsc->id, operation,
interval_spec, data_set);
} else {
rc = clear_rsc_history(controld_api, host_uname, rsc->id, data_set);
}
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned %s failures on %s, but unable to clean history: %s",
rsc->id, host_uname, pcmk_strerror(rc));
} else {
out->info(out, "Cleaned up %s on %s", rsc->id, host_uname);
}
return rc;
}
// \return Standard Pacemaker return code
int
cli_cleanup_all(pcmk_ipc_api_t *controld_api, const char *node_name,
const char *operation, const char *interval_spec,
pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
int attr_options = pcmk__node_attr_none;
const char *display_name = node_name? node_name : "all nodes";
if (controld_api == NULL) {
out->info(out, "Dry run: skipping clean-up of %s due to CIB_file",
display_name);
return rc;
}
if (node_name) {
pe_node_t *node = pe_find_node(data_set->nodes, node_name);
if (node == NULL) {
out->err(out, "Unknown node: %s", node_name);
return ENXIO;
}
if (pe__is_guest_or_remote_node(node)) {
attr_options |= pcmk__node_attr_remote;
}
}
rc = pcmk__attrd_api_clear_failures(NULL, node_name, NULL, operation,
interval_spec, NULL, attr_options);
if (rc != pcmk_rc_ok) {
out->err(out, "Unable to clean up all failures on %s: %s",
display_name, pcmk_rc_str(rc));
return rc;
}
if (node_name) {
rc = clear_rsc_failures(out, controld_api, node_name, NULL,
operation, interval_spec, data_set);
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned all resource failures on %s, but unable to clean history: %s",
node_name, pcmk_strerror(rc));
return rc;
}
} else {
for (GList *iter = data_set->nodes; iter; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
rc = clear_rsc_failures(out, controld_api, node->details->uname, NULL,
operation, interval_spec, data_set);
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned all resource failures on all nodes, but unable to clean history: %s",
pcmk_strerror(rc));
return rc;
}
}
}
out->info(out, "Cleaned up all resources on %s", display_name);
return rc;
}
static void
check_role(resource_checks_t *checks)
{
const char *role_s = g_hash_table_lookup(checks->rsc->meta,
XML_RSC_ATTR_TARGET_ROLE);
if (role_s == NULL) {
return;
}
switch (text2role(role_s)) {
case pcmk_role_stopped:
checks->flags |= rsc_remain_stopped;
break;
case pcmk_role_unpromoted:
if (pcmk_is_set(pe__const_top_resource(checks->rsc, false)->flags,
pcmk_rsc_promotable)) {
checks->flags |= rsc_unpromotable;
}
break;
default:
break;
}
}
static void
check_managed(resource_checks_t *checks)
{
const char *managed_s = g_hash_table_lookup(checks->rsc->meta,
XML_RSC_ATTR_MANAGED);
if ((managed_s != NULL) && !crm_is_true(managed_s)) {
checks->flags |= rsc_unmanaged;
}
}
static void
check_locked(resource_checks_t *checks)
{
if (checks->rsc->lock_node != NULL) {
checks->flags |= rsc_locked;
checks->lock_node = checks->rsc->lock_node->details->uname;
}
}
static bool
node_is_unhealthy(pe_node_t *node)
{
switch (pe__health_strategy(node->details->data_set)) {
case pcmk__health_strategy_none:
break;
case pcmk__health_strategy_no_red:
if (pe__node_health(node) < 0) {
return true;
}
break;
case pcmk__health_strategy_only_green:
if (pe__node_health(node) <= 0) {
return true;
}
break;
case pcmk__health_strategy_progressive:
case pcmk__health_strategy_custom:
/* @TODO These are finite scores, possibly with rules, and possibly
* combining with other scores, so attributing these as a cause is
* nontrivial.
*/
break;
}
return false;
}
static void
check_node_health(resource_checks_t *checks, pe_node_t *node)
{
if (node == NULL) {
GHashTableIter iter;
bool allowed = false;
bool all_nodes_unhealthy = true;
g_hash_table_iter_init(&iter, checks->rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
allowed = true;
if (!node_is_unhealthy(node)) {
all_nodes_unhealthy = false;
break;
}
}
if (allowed && all_nodes_unhealthy) {
checks->flags |= rsc_node_health;
}
} else if (node_is_unhealthy(node)) {
checks->flags |= rsc_node_health;
}
}
int
cli_resource_check(pcmk__output_t *out, pe_resource_t *rsc, pe_node_t *node)
{
resource_checks_t checks = { .rsc = rsc };
check_role(&checks);
check_managed(&checks);
check_locked(&checks);
check_node_health(&checks, node);
return out->message(out, "resource-check-list", &checks);
}
// \return Standard Pacemaker return code
int
cli_resource_fail(pcmk_ipc_api_t *controld_api, const char *host_uname,
const char *rsc_id, pe_working_set_t *data_set)
{
crm_notice("Failing %s on %s", rsc_id, host_uname);
return send_lrm_rsc_op(controld_api, true, host_uname, rsc_id, data_set);
}
static GHashTable *
generate_resource_params(pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set)
{
GHashTable *params = NULL;
GHashTable *meta = NULL;
GHashTable *combined = NULL;
GHashTableIter iter;
char *key = NULL;
char *value = NULL;
combined = pcmk__strkey_table(free, free);
params = pe_rsc_params(rsc, node, data_set);
if (params != NULL) {
g_hash_table_iter_init(&iter, params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
g_hash_table_insert(combined, strdup(key), strdup(value));
}
}
meta = pcmk__strkey_table(free, free);
get_meta_attributes(meta, rsc, node, data_set);
if (meta != NULL) {
g_hash_table_iter_init(&iter, meta);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
char *crm_name = crm_meta_name(key);
g_hash_table_insert(combined, crm_name, strdup(value));
}
g_hash_table_destroy(meta);
}
return combined;
}
bool resource_is_running_on(pe_resource_t *rsc, const char *host)
{
bool found = true;
GList *hIter = NULL;
GList *hosts = NULL;
if (rsc == NULL) {
return false;
}
rsc->fns->location(rsc, &hosts, TRUE);
for (hIter = hosts; host != NULL && hIter != NULL; hIter = hIter->next) {
pe_node_t *node = (pe_node_t *) hIter->data;
if (pcmk__strcase_any_of(host, node->details->uname, node->details->id, NULL)) {
crm_trace("Resource %s is running on %s\n", rsc->id, host);
goto done;
}
}
if (host != NULL) {
crm_trace("Resource %s is not running on: %s\n", rsc->id, host);
found = false;
} else if(host == NULL && hosts == NULL) {
crm_trace("Resource %s is not running\n", rsc->id);
found = false;
}
done:
g_list_free(hosts);
return found;
}
/*!
* \internal
* \brief Create a list of all resources active on host from a given list
*
* \param[in] host Name of host to check whether resources are active
* \param[in] rsc_list List of resources to check
*
* \return New list of resources from list that are active on host
*/
static GList *
get_active_resources(const char *host, GList *rsc_list)
{
GList *rIter = NULL;
GList *active = NULL;
for (rIter = rsc_list; rIter != NULL; rIter = rIter->next) {
pe_resource_t *rsc = (pe_resource_t *) rIter->data;
/* Expand groups to their members, because if we're restarting a member
* other than the first, we can't otherwise tell which resources are
* stopping and starting.
*/
if (rsc->variant == pcmk_rsc_variant_group) {
active = g_list_concat(active,
get_active_resources(host, rsc->children));
} else if (resource_is_running_on(rsc, host)) {
active = g_list_append(active, strdup(rsc->id));
}
}
return active;
}
static void dump_list(GList *items, const char *tag)
{
int lpc = 0;
GList *item = NULL;
for (item = items; item != NULL; item = item->next) {
crm_trace("%s[%d]: %s", tag, lpc, (char*)item->data);
lpc++;
}
}
static void display_list(pcmk__output_t *out, GList *items, const char *tag)
{
GList *item = NULL;
for (item = items; item != NULL; item = item->next) {
out->info(out, "%s%s", tag, (const char *)item->data);
}
}
/*!
* \internal
* \brief Upgrade XML to latest schema version and use it as working set input
*
* This also updates the working set timestamp to the current time.
*
* \param[in,out] data_set Working set instance to update
* \param[in,out] xml XML to use as input
*
* \return Standard Pacemaker return code
* \note On success, caller is responsible for freeing memory allocated for
* data_set->now.
* \todo This follows the example of other callers of cli_config_update()
* and returns ENOKEY ("Required key not available") if that fails,
* but perhaps pcmk_rc_schema_validation would be better in that case.
*/
int
update_working_set_xml(pe_working_set_t *data_set, xmlNode **xml)
{
if (cli_config_update(xml, NULL, FALSE) == FALSE) {
return ENOKEY;
}
data_set->input = *xml;
data_set->now = crm_time_new(NULL);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Update a working set's XML input based on a CIB query
*
* \param[in] data_set Data set instance to initialize
* \param[in] cib Connection to the CIB manager
*
* \return Standard Pacemaker return code
* \note On success, caller is responsible for freeing memory allocated for
* data_set->input and data_set->now.
*/
static int
update_working_set_from_cib(pcmk__output_t *out, pe_working_set_t * data_set,
cib_t *cib)
{
xmlNode *cib_xml_copy = NULL;
int rc = pcmk_rc_ok;
rc = cib->cmds->query(cib, NULL, &cib_xml_copy, cib_scope_local | cib_sync_call);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not obtain the current CIB: %s (%d)", pcmk_strerror(rc), rc);
return rc;
}
rc = update_working_set_xml(data_set, &cib_xml_copy);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not upgrade the current CIB XML");
free_xml(cib_xml_copy);
return rc;
}
return rc;
}
// \return Standard Pacemaker return code
static int
update_dataset(cib_t *cib, pe_working_set_t * data_set, bool simulate)
{
char *pid = NULL;
char *shadow_file = NULL;
cib_t *shadow_cib = NULL;
int rc = pcmk_rc_ok;
pcmk__output_t *out = data_set->priv;
pe_reset_working_set(data_set);
pe__set_working_set_flags(data_set,
pcmk_sched_no_counts|pcmk_sched_no_compat);
rc = update_working_set_from_cib(out, data_set, cib);
if (rc != pcmk_rc_ok) {
return rc;
}
if(simulate) {
bool prev_quiet = false;
pid = pcmk__getpid_s();
shadow_cib = cib_shadow_new(pid);
shadow_file = get_shadow_file(pid);
if (shadow_cib == NULL) {
out->err(out, "Could not create shadow cib: '%s'", pid);
rc = ENXIO;
goto done;
}
rc = write_xml_file(data_set->input, shadow_file, FALSE);
if (rc < 0) {
out->err(out, "Could not populate shadow cib: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
rc = shadow_cib->cmds->signon(shadow_cib, crm_system_name, cib_command);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not connect to shadow cib: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
pcmk__schedule_actions(data_set->input,
pcmk_sched_no_counts|pcmk_sched_no_compat,
data_set);
prev_quiet = out->is_quiet(out);
out->quiet = true;
pcmk__simulate_transition(data_set, shadow_cib, NULL);
out->quiet = prev_quiet;
rc = update_dataset(shadow_cib, data_set, false);
} else {
cluster_status(data_set);
}
done:
/* Do not free data_set->input here, we need rsc->xml to be valid later on */
cib_delete(shadow_cib);
free(pid);
if(shadow_file) {
unlink(shadow_file);
free(shadow_file);
}
return rc;
}
/*!
* \internal
* \brief Find the maximum stop timeout of a resource and its children (if any)
*
* \param[in,out] rsc Resource to get timeout for
*
* \return Maximum stop timeout for \p rsc (in milliseconds)
*/
static int
max_rsc_stop_timeout(pe_resource_t *rsc)
{
pe_action_t *stop = NULL;
long long result_ll;
int max_delay = 0;
if (rsc == NULL) {
return 0;
}
// If resource is collective, use maximum of its children's stop timeouts
if (rsc->children != NULL) {
for (GList *iter = rsc->children; iter; iter = iter->next) {
pe_resource_t *child = iter->data;
int delay = max_rsc_stop_timeout(child);
if (delay > max_delay) {
pe_rsc_trace(rsc,
"Maximum stop timeout for %s is now %s due to %s",
rsc->id, pcmk__readable_interval(delay), child->id);
max_delay = delay;
}
}
return max_delay;
}
/* Create a (transient) instance of the resource's stop action, to fully
* evaluate its timeout for rules, defaults, etc.
*
* @TODO This currently ignores node (which might matter for rules)
*/
stop = custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, NULL, TRUE,
FALSE, rsc->cluster);
if ((pcmk__scan_ll(g_hash_table_lookup(stop->meta, XML_ATTR_TIMEOUT),
&result_ll, -1LL) == pcmk_rc_ok)
&& (result_ll >= 0) && (result_ll <= INT_MAX)) {
max_delay = (int) result_ll;
}
pe_free_action(stop);
return max_delay;
}
/*!
* \internal
* \brief Find a reasonable waiting time for stopping any one resource in a list
*
* \param[in,out] data_set Cluster working set
* \param[in] resources List of names of resources that will be stopped
*
* \return Rough estimate of a reasonable time to wait (in seconds) to stop any
* one resource in \p resources
* \note This estimate is very rough, simply the maximum stop timeout of all
* given resources and their children, plus a small fudge factor. It does
* not account for children that must be stopped in sequence, action
* throttling, or any demotions needed. It checks the stop timeout, even
* if the resources in question are actually being started.
*/
static int
wait_time_estimate(pe_working_set_t *data_set, const GList *resources)
{
int max_delay = 0;
// Find maximum stop timeout in milliseconds
for (const GList *item = resources; item != NULL; item = item->next) {
pe_resource_t *rsc = pe_find_resource(data_set->resources,
(const char *) (item->data));
int delay = max_rsc_stop_timeout(rsc);
if (delay > max_delay) {
pe_rsc_trace(rsc,
"Wait time is now %s due to %s",
pcmk__readable_interval(delay), rsc->id);
max_delay = delay;
}
}
return (max_delay / 1000) + 5;
}
#define waiting_for_starts(d, r, h) ((d != NULL) || \
(!resource_is_running_on((r), (h))))
/*!
* \internal
* \brief Restart a resource (on a particular host if requested).
*
* \param[in,out] out Output object
* \param[in,out] rsc The resource to restart
* \param[in] node Node to restart resource on (NULL for all)
* \param[in] move_lifetime If not NULL, how long constraint should
* remain in effect (as ISO 8601 string)
* \param[in] timeout_ms Consider failed if actions do not complete
* in this time (specified in milliseconds,
* but a two-second granularity is actually
* used; if 0, it will be calculated based on
* the resource timeout)
* \param[in,out] cib Connection to the CIB manager
* \param[in] cib_options Group of enum cib_call_options flags to
* use with CIB calls
* \param[in] promoted_role_only If true, limit to promoted instances
* \param[in] force If true, apply only to requested instance
* if part of a collective resource
*
* \return Standard Pacemaker return code (exits on certain failures)
*/
int
cli_resource_restart(pcmk__output_t *out, pe_resource_t *rsc,
const pe_node_t *node, const char *move_lifetime,
int timeout_ms, cib_t *cib, int cib_options,
gboolean promoted_role_only, gboolean force)
{
int rc = pcmk_rc_ok;
int lpc = 0;
int before = 0;
int step_timeout_s = 0;
int sleep_interval = 2;
int timeout = timeout_ms / 1000;
bool stop_via_ban = false;
char *rsc_id = NULL;
char *lookup_id = NULL;
char *orig_target_role = NULL;
GList *list_delta = NULL;
GList *target_active = NULL;
GList *current_active = NULL;
GList *restart_target_active = NULL;
pe_working_set_t *data_set = NULL;
pe_resource_t *parent = uber_parent(rsc);
bool running = false;
const char *id = rsc->clone_name ? rsc->clone_name : rsc->id;
const char *host = node ? node->details->uname : NULL;
/* If the implicit resource or primitive resource of a bundle is given, operate on the
* bundle itself instead.
*/
if (pe_rsc_is_bundled(rsc)) {
rsc = parent->parent;
}
running = resource_is_running_on(rsc, host);
if (pe_rsc_is_clone(parent) && !running) {
if (pe_rsc_is_unique_clone(parent)) {
lookup_id = strdup(rsc->id);
} else {
lookup_id = clone_strip(rsc->id);
}
rsc = parent->fns->find_rsc(parent, lookup_id, node,
pcmk_rsc_match_basename
|pcmk_rsc_match_current_node);
free(lookup_id);
running = resource_is_running_on(rsc, host);
}
if (!running) {
if (host) {
out->err(out, "%s is not running on %s and so cannot be restarted", id, host);
} else {
out->err(out, "%s is not running anywhere and so cannot be restarted", id);
}
return ENXIO;
}
rsc_id = strdup(rsc->id);
if (pe_rsc_is_unique_clone(parent)) {
lookup_id = strdup(rsc->id);
} else {
lookup_id = clone_strip(rsc->id);
}
if (host) {
if (pe_rsc_is_clone(rsc) || pe_bundle_replicas(rsc)) {
stop_via_ban = true;
} else if (pe_rsc_is_clone(parent)) {
stop_via_ban = true;
free(lookup_id);
lookup_id = strdup(parent->id);
}
}
/*
grab full cib
determine originally active resources
disable or ban
poll cib and watch for affected resources to get stopped
without --timeout, calculate the stop timeout for each step and wait for that
if we hit --timeout or the service timeout, re-enable or un-ban, report failure and indicate which resources we couldn't take down
if everything stopped, re-enable or un-ban
poll cib and watch for affected resources to get started
without --timeout, calculate the start timeout for each step and wait for that
if we hit --timeout or the service timeout, report (different) failure and indicate which resources we couldn't bring back up
report success
Optimizations:
- use constraints to determine ordered list of affected resources
- Allow a --no-deps option (aka. --force-restart)
*/
data_set = pe_new_working_set();
if (data_set == NULL) {
crm_perror(LOG_ERR, "Could not allocate working set");
rc = ENOMEM;
goto done;
}
data_set->priv = out;
rc = update_dataset(cib, data_set, false);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not get new resource list: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
restart_target_active = get_active_resources(host, data_set->resources);
current_active = get_active_resources(host, data_set->resources);
dump_list(current_active, "Origin");
if (stop_via_ban) {
/* Stop the clone or bundle instance by banning it from the host */
out->quiet = true;
rc = cli_resource_ban(out, lookup_id, host, move_lifetime, NULL, cib,
cib_options, promoted_role_only);
} else {
/* Stop the resource by setting target-role to Stopped.
* Remember any existing target-role so we can restore it later
* (though it only makes any difference if it's Unpromoted).
*/
find_resource_attr(out, cib, XML_NVPAIR_ATTR_VALUE, lookup_id, NULL, NULL,
NULL, XML_RSC_ATTR_TARGET_ROLE, &orig_target_role);
rc = cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE,
PCMK_ACTION_STOPPED, FALSE, cib,
cib_options, force);
}
if(rc != pcmk_rc_ok) {
out->err(out, "Could not set target-role for %s: %s (%d)", rsc_id, pcmk_strerror(rc), rc);
if (current_active != NULL) {
g_list_free_full(current_active, free);
current_active = NULL;
}
if (restart_target_active != NULL) {
g_list_free_full(restart_target_active, free);
restart_target_active = NULL;
}
goto done;
}
rc = update_dataset(cib, data_set, true);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources would be stopped");
goto failure;
}
target_active = get_active_resources(host, data_set->resources);
dump_list(target_active, "Target");
list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp);
out->info(out, "Waiting for %d resources to stop:", g_list_length(list_delta));
display_list(out, list_delta, " * ");
step_timeout_s = timeout / sleep_interval;
while (list_delta != NULL) {
before = g_list_length(list_delta);
if(timeout_ms == 0) {
step_timeout_s = wait_time_estimate(data_set, list_delta)
/ sleep_interval;
}
/* We probably don't need the entire step timeout */
for(lpc = 0; (lpc < step_timeout_s) && (list_delta != NULL); lpc++) {
sleep(sleep_interval);
if(timeout) {
timeout -= sleep_interval;
crm_trace("%ds remaining", timeout);
}
rc = update_dataset(cib, data_set, FALSE);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources were stopped");
goto failure;
}
if (current_active != NULL) {
g_list_free_full(current_active, free);
current_active = NULL;
}
current_active = get_active_resources(host, data_set->resources);
g_list_free(list_delta);
list_delta = NULL;
list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp);
dump_list(current_active, "Current");
dump_list(list_delta, "Delta");
}
crm_trace("%d (was %d) resources remaining", g_list_length(list_delta), before);
if(before == g_list_length(list_delta)) {
/* aborted during stop phase, print the contents of list_delta */
out->err(out, "Could not complete shutdown of %s, %d resources remaining", rsc_id, g_list_length(list_delta));
display_list(out, list_delta, " * ");
rc = ETIME;
goto failure;
}
}
if (stop_via_ban) {
rc = cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force);
} else if (orig_target_role) {
rc = cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE,
orig_target_role, FALSE, cib,
cib_options, force);
free(orig_target_role);
orig_target_role = NULL;
} else {
rc = cli_resource_delete_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE, cib,
cib_options, force);
}
if(rc != pcmk_rc_ok) {
out->err(out, "Could not unset target-role for %s: %s (%d)", rsc_id, pcmk_strerror(rc), rc);
goto done;
}
if (target_active != NULL) {
g_list_free_full(target_active, free);
target_active = NULL;
}
target_active = restart_target_active;
list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp);
out->info(out, "Waiting for %d resources to start again:", g_list_length(list_delta));
display_list(out, list_delta, " * ");
step_timeout_s = timeout / sleep_interval;
while (waiting_for_starts(list_delta, rsc, host)) {
before = g_list_length(list_delta);
if(timeout_ms == 0) {
step_timeout_s = wait_time_estimate(data_set, list_delta)
/ sleep_interval;
}
/* We probably don't need the entire step timeout */
for (lpc = 0; (lpc < step_timeout_s) && waiting_for_starts(list_delta, rsc, host); lpc++) {
sleep(sleep_interval);
if(timeout) {
timeout -= sleep_interval;
crm_trace("%ds remaining", timeout);
}
rc = update_dataset(cib, data_set, false);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources were started");
goto failure;
}
if (current_active != NULL) {
g_list_free_full(current_active, free);
current_active = NULL;
}
/* It's OK if dependent resources moved to a different node,
* so we check active resources on all nodes.
*/
current_active = get_active_resources(NULL, data_set->resources);
g_list_free(list_delta);
list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp);
dump_list(current_active, "Current");
dump_list(list_delta, "Delta");
}
if(before == g_list_length(list_delta)) {
/* aborted during start phase, print the contents of list_delta */
out->err(out, "Could not complete restart of %s, %d resources remaining", rsc_id, g_list_length(list_delta));
display_list(out, list_delta, " * ");
rc = ETIME;
goto failure;
}
}
rc = pcmk_rc_ok;
goto done;
failure:
if (stop_via_ban) {
cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force);
} else if (orig_target_role) {
cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS, NULL,
XML_RSC_ATTR_TARGET_ROLE, orig_target_role,
FALSE, cib, cib_options, force);
free(orig_target_role);
} else {
cli_resource_delete_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE, cib,
cib_options, force);
}
done:
if (list_delta != NULL) {
g_list_free(list_delta);
}
if (current_active != NULL) {
g_list_free_full(current_active, free);
}
if (target_active != NULL && (target_active != restart_target_active)) {
g_list_free_full(target_active, free);
}
if (restart_target_active != NULL) {
g_list_free_full(restart_target_active, free);
}
free(rsc_id);
free(lookup_id);
pe_free_working_set(data_set);
return rc;
}
static inline bool
action_is_pending(const pe_action_t *action)
{
- if (pcmk_any_flags_set(action->flags, pe_action_optional|pe_action_pseudo)
+ if (pcmk_any_flags_set(action->flags, pe_action_optional|pcmk_action_pseudo)
|| !pcmk_is_set(action->flags, pe_action_runnable)
|| pcmk__str_eq(PCMK_ACTION_NOTIFY, action->task, pcmk__str_casei)) {
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether any actions in a list are pending
*
* \param[in] actions List of actions to check
*
* \return true if any actions in the list are pending, otherwise false
*/
static bool
actions_are_pending(const GList *actions)
{
for (const GList *action = actions; action != NULL; action = action->next) {
const pe_action_t *a = (const pe_action_t *) action->data;
if (action_is_pending(a)) {
crm_notice("Waiting for %s (flags=%#.8x)", a->uuid, a->flags);
return true;
}
}
return false;
}
static void
print_pending_actions(pcmk__output_t *out, GList *actions)
{
GList *action;
out->info(out, "Pending actions:");
for (action = actions; action != NULL; action = action->next) {
pe_action_t *a = (pe_action_t *) action->data;
if (!action_is_pending(a)) {
continue;
}
if (a->node) {
out->info(out, "\tAction %d: %s\ton %s",
a->id, a->uuid, pe__node_name(a->node));
} else {
out->info(out, "\tAction %d: %s", a->id, a->uuid);
}
}
}
/* For --wait, timeout (in seconds) to use if caller doesn't specify one */
#define WAIT_DEFAULT_TIMEOUT_S (60 * 60)
/* For --wait, how long to sleep between cluster state checks */
#define WAIT_SLEEP_S (2)
/*!
* \internal
* \brief Wait until all pending cluster actions are complete
*
* This waits until either the CIB's transition graph is idle or a timeout is
* reached.
*
* \param[in,out] out Output object
* \param[in] timeout_ms Consider failed if actions do not complete in
* this time (specified in milliseconds, but
* one-second granularity is actually used; if 0, a
* default will be used)
* \param[in,out] cib Connection to the CIB manager
*
* \return Standard Pacemaker return code
*/
int
wait_till_stable(pcmk__output_t *out, int timeout_ms, cib_t * cib)
{
pe_working_set_t *data_set = NULL;
int rc = pcmk_rc_ok;
int timeout_s = timeout_ms? ((timeout_ms + 999) / 1000) : WAIT_DEFAULT_TIMEOUT_S;
time_t expire_time = time(NULL) + timeout_s;
time_t time_diff;
bool printed_version_warning = out->is_quiet(out); // i.e. don't print if quiet
data_set = pe_new_working_set();
if (data_set == NULL) {
return ENOMEM;
}
do {
/* Abort if timeout is reached */
time_diff = expire_time - time(NULL);
if (time_diff > 0) {
crm_info("Waiting up to %lld seconds for cluster actions to complete", (long long) time_diff);
} else {
print_pending_actions(out, data_set->actions);
pe_free_working_set(data_set);
return ETIME;
}
if (rc == pcmk_rc_ok) { /* this avoids sleep on first loop iteration */
sleep(WAIT_SLEEP_S);
}
/* Get latest transition graph */
pe_reset_working_set(data_set);
rc = update_working_set_from_cib(out, data_set, cib);
if (rc != pcmk_rc_ok) {
pe_free_working_set(data_set);
return rc;
}
pcmk__schedule_actions(data_set->input,
pcmk_sched_no_counts|pcmk_sched_no_compat,
data_set);
if (!printed_version_warning) {
/* If the DC has a different version than the local node, the two
* could come to different conclusions about what actions need to be
* done. Warn the user in this case.
*
* @TODO A possible long-term solution would be to reimplement the
* wait as a new controller operation that would be forwarded to the
* DC. However, that would have potential problems of its own.
*/
const char *dc_version = g_hash_table_lookup(data_set->config_hash,
"dc-version");
if (!pcmk__str_eq(dc_version, PACEMAKER_VERSION "-" BUILD_VERSION, pcmk__str_casei)) {
out->info(out, "warning: wait option may not work properly in "
"mixed-version cluster");
printed_version_warning = true;
}
}
} while (actions_are_pending(data_set->actions));
pe_free_working_set(data_set);
return rc;
}
static const char *
get_action(const char *rsc_action) {
const char *action = NULL;
if (pcmk__str_eq(rsc_action, "validate", pcmk__str_casei)) {
action = PCMK_ACTION_VALIDATE_ALL;
} else if (pcmk__str_eq(rsc_action, "force-check", pcmk__str_casei)) {
action = PCMK_ACTION_MONITOR;
} else if (pcmk__strcase_any_of(rsc_action, "force-start", "force-stop",
"force-demote", "force-promote", NULL)) {
action = rsc_action+6;
} else {
action = rsc_action;
}
return action;
}
/*!
* \brief Set up environment variables as expected by resource agents
*
* When the cluster executes resource agents, it adds certain environment
* variables (directly or via resource meta-attributes) expected by some
* resource agents. Add the essential ones that many resource agents expect, so
* the behavior is the same for command-line execution.
*
* \param[in,out] params Resource parameters that will be passed to agent
* \param[in] timeout_ms Action timeout (in milliseconds)
* \param[in] check_level OCF check level
* \param[in] verbosity Verbosity level
*/
static void
set_agent_environment(GHashTable *params, int timeout_ms, int check_level,
int verbosity)
{
g_hash_table_insert(params, strdup("CRM_meta_timeout"),
crm_strdup_printf("%d", timeout_ms));
g_hash_table_insert(params, strdup(XML_ATTR_CRM_VERSION),
strdup(CRM_FEATURE_SET));
if (check_level >= 0) {
char *level = crm_strdup_printf("%d", check_level);
setenv("OCF_CHECK_LEVEL", level, 1);
free(level);
}
setenv("HA_debug", (verbosity > 0)? "1" : "0", 1);
if (verbosity > 1) {
setenv("OCF_TRACE_RA", "1", 1);
}
/* A resource agent using the standard ocf-shellfuncs library will not print
* messages to stderr if it doesn't have a controlling terminal (e.g. if
* crm_resource is called via script or ssh). This forces it to do so.
*/
setenv("OCF_TRACE_FILE", "/dev/stderr", 0);
}
/*!
* \internal
* \brief Apply command-line overrides to resource parameters
*
* \param[in,out] params Parameters to be passed to agent
* \param[in] overrides Parameters to override (or NULL if none)
*/
static void
apply_overrides(GHashTable *params, GHashTable *overrides)
{
if (overrides != NULL) {
GHashTableIter iter;
char *name = NULL;
char *value = NULL;
g_hash_table_iter_init(&iter, overrides);
while (g_hash_table_iter_next(&iter, (gpointer *) &name,
(gpointer *) &value)) {
g_hash_table_replace(params, strdup(name), strdup(value));
}
}
}
crm_exit_t
cli_resource_execute_from_params(pcmk__output_t *out, const char *rsc_name,
const char *rsc_class, const char *rsc_prov,
const char *rsc_type, const char *rsc_action,
GHashTable *params, GHashTable *override_hash,
int timeout_ms, int resource_verbose, gboolean force,
int check_level)
{
const char *class = rsc_class;
const char *action = get_action(rsc_action);
crm_exit_t exit_code = CRM_EX_OK;
svc_action_t *op = NULL;
// If no timeout was provided, use the same default as the cluster
if (timeout_ms == 0) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
set_agent_environment(params, timeout_ms, check_level, resource_verbose);
apply_overrides(params, override_hash);
op = services__create_resource_action(rsc_name? rsc_name : "test",
rsc_class, rsc_prov, rsc_type, action,
0, timeout_ms, params, 0);
if (op == NULL) {
out->err(out, "Could not execute %s using %s%s%s:%s: %s",
action, rsc_class, (rsc_prov? ":" : ""),
(rsc_prov? rsc_prov : ""), rsc_type, strerror(ENOMEM));
g_hash_table_destroy(params);
return CRM_EX_OSERR;
}
if (pcmk__str_eq(rsc_class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) {
class = resources_find_service_class(rsc_type);
}
if (!pcmk__strcase_any_of(class, PCMK_RESOURCE_CLASS_OCF,
PCMK_RESOURCE_CLASS_LSB, NULL)) {
services__format_result(op, CRM_EX_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR,
"Manual execution of the %s standard is "
"unsupported", pcmk__s(class, "unspecified"));
}
if (op->rc != PCMK_OCF_UNKNOWN) {
exit_code = op->rc;
goto done;
}
services_action_sync(op);
// Map results to OCF codes for consistent reporting to user
{
enum ocf_exitcode ocf_code = services_result2ocf(class, action, op->rc);
// Cast variable instead of function return to keep compilers happy
exit_code = (crm_exit_t) ocf_code;
}
done:
out->message(out, "resource-agent-action", resource_verbose, rsc_class,
rsc_prov, rsc_type, rsc_name, rsc_action, override_hash,
exit_code, op->status, services__exit_reason(op),
op->stdout_data, op->stderr_data);
services_action_free(op);
return exit_code;
}
crm_exit_t
cli_resource_execute(pe_resource_t *rsc, const char *requested_name,
const char *rsc_action, GHashTable *override_hash,
int timeout_ms, cib_t * cib, pe_working_set_t *data_set,
int resource_verbose, gboolean force, int check_level)
{
pcmk__output_t *out = data_set->priv;
crm_exit_t exit_code = CRM_EX_OK;
const char *rid = NULL;
const char *rtype = NULL;
const char *rprov = NULL;
const char *rclass = NULL;
GHashTable *params = NULL;
if (pcmk__strcase_any_of(rsc_action, "force-start", "force-demote",
"force-promote", NULL)) {
if(pe_rsc_is_clone(rsc)) {
GList *nodes = cli_resource_search(rsc, requested_name, data_set);
if(nodes != NULL && force == FALSE) {
out->err(out, "It is not safe to %s %s here: the cluster claims it is already active",
rsc_action, rsc->id);
out->err(out, "Try setting target-role=Stopped first or specifying "
"the force option");
return CRM_EX_UNSAFE;
}
g_list_free_full(nodes, free);
}
}
if(pe_rsc_is_clone(rsc)) {
/* Grab the first child resource in the hope it's not a group */
rsc = rsc->children->data;
}
if (rsc->variant == pcmk_rsc_variant_group) {
out->err(out, "Sorry, the %s option doesn't support group resources", rsc_action);
return CRM_EX_UNIMPLEMENT_FEATURE;
} else if (pe_rsc_is_bundled(rsc)) {
out->err(out, "Sorry, the %s option doesn't support bundled resources", rsc_action);
return CRM_EX_UNIMPLEMENT_FEATURE;
}
rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
rprov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE);
params = generate_resource_params(rsc, NULL /* @TODO use local node */,
data_set);
if (timeout_ms == 0) {
timeout_ms = pe_get_configured_timeout(rsc, get_action(rsc_action), data_set);
}
rid = pe_rsc_is_anon_clone(rsc->parent)? requested_name : rsc->id;
exit_code = cli_resource_execute_from_params(out, rid, rclass, rprov, rtype, rsc_action,
params, override_hash, timeout_ms,
resource_verbose, force, check_level);
return exit_code;
}
// \return Standard Pacemaker return code
int
cli_resource_move(const pe_resource_t *rsc, const char *rsc_id,
const char *host_name, const char *move_lifetime, cib_t *cib,
int cib_options, pe_working_set_t *data_set,
gboolean promoted_role_only, gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
unsigned int count = 0;
pe_node_t *current = NULL;
pe_node_t *dest = pe_find_node(data_set->nodes, host_name);
bool cur_is_dest = false;
if (dest == NULL) {
return pcmk_rc_node_unknown;
}
if (promoted_role_only
&& !pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) {
const pe_resource_t *p = pe__const_top_resource(rsc, false);
if (pcmk_is_set(p->flags, pcmk_rsc_promotable)) {
out->info(out, "Using parent '%s' for move instead of '%s'.", rsc->id, rsc_id);
rsc_id = p->id;
rsc = p;
} else {
out->info(out, "Ignoring --promoted option: %s is not promotable",
rsc_id);
promoted_role_only = FALSE;
}
}
current = pe__find_active_requires(rsc, &count);
if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) {
unsigned int promoted_count = 0;
pe_node_t *promoted_node = NULL;
for (const GList *iter = rsc->children; iter; iter = iter->next) {
const pe_resource_t *child = (const pe_resource_t *) iter->data;
enum rsc_role_e child_role = child->fns->state(child, TRUE);
if (child_role == pcmk_role_promoted) {
rsc = child;
promoted_node = pe__current_node(child);
promoted_count++;
}
}
if (promoted_role_only || (promoted_count != 0)) {
count = promoted_count;
current = promoted_node;
}
}
if (count > 1) {
if (pe_rsc_is_clone(rsc)) {
current = NULL;
} else {
return pcmk_rc_multiple;
}
}
if (current && (current->details == dest->details)) {
cur_is_dest = true;
if (force) {
crm_info("%s is already %s on %s, reinforcing placement with location constraint.",
rsc_id, promoted_role_only?"promoted":"active",
pe__node_name(dest));
} else {
return pcmk_rc_already;
}
}
/* Clear any previous prefer constraints across all nodes. */
cli_resource_clear(rsc_id, NULL, data_set->nodes, cib, cib_options, false, force);
/* Clear any previous ban constraints on 'dest'. */
cli_resource_clear(rsc_id, dest->details->uname, data_set->nodes, cib,
cib_options, TRUE, force);
/* Record an explicit preference for 'dest' */
rc = cli_resource_prefer(out, rsc_id, dest->details->uname, move_lifetime,
cib, cib_options, promoted_role_only);
crm_trace("%s%s now prefers %s%s",
rsc->id, (promoted_role_only? " (promoted)" : ""),
pe__node_name(dest), force?"(forced)":"");
/* only ban the previous location if current location != destination location.
* it is possible to use -M to enforce a location without regard of where the
* resource is currently located */
if (force && !cur_is_dest) {
/* Ban the original location if possible */
if(current) {
(void)cli_resource_ban(out, rsc_id, current->details->uname, move_lifetime,
NULL, cib, cib_options, promoted_role_only);
} else if(count > 1) {
out->info(out, "Resource '%s' is currently %s in %d locations. "
"One may now move to %s",
rsc_id, (promoted_role_only? "promoted" : "active"),
count, pe__node_name(dest));
out->info(out, "To prevent '%s' from being %s at a specific location, "
"specify a node.",
rsc_id, (promoted_role_only? "promoted" : "active"));
} else {
crm_trace("Not banning %s from its current location: not active", rsc_id);
}
}
return rc;
}