Page MenuHomeClusterLabs Projects
Files F5733957

No OneTemporary
Actions

View Options

This file is larger than 256 KB, so syntax highlighting was skipped.
diff --git a/lib/pengine/failcounts.c b/lib/pengine/failcounts.c
index e7ce2e2e6f..ecc633e1e5 100644
--- a/lib/pengine/failcounts.c
+++ b/lib/pengine/failcounts.c
@@ -1,481 +1,481 @@
/*
* Copyright 2008-2025 the Pacemaker project contributors
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/types.h>
#include <regex.h>
#include <glib.h>
#include <libxml/xpath.h> // xmlXPathObject, etc.
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/util.h>
#include <crm/pengine/internal.h>
static gboolean
is_matched_failure(const char *rsc_id, const xmlNode *conf_op_xml,
const xmlNode *lrm_op_xml)
{
gboolean matched = FALSE;
const char *conf_op_name = NULL;
const char *lrm_op_task = NULL;
const char *conf_op_interval_spec = NULL;
guint conf_op_interval_ms = 0;
guint lrm_op_interval_ms = 0;
const char *lrm_op_id = NULL;
char *last_failure_key = NULL;
if (rsc_id == NULL || conf_op_xml == NULL || lrm_op_xml == NULL) {
return FALSE;
}
// Get name and interval from configured op
conf_op_name = pcmk__xe_get(conf_op_xml, PCMK_XA_NAME);
conf_op_interval_spec = pcmk__xe_get(conf_op_xml, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(conf_op_interval_spec, &conf_op_interval_ms);
// Get name and interval from op history entry
lrm_op_task = pcmk__xe_get(lrm_op_xml, PCMK_XA_OPERATION);
pcmk__xe_get_guint(lrm_op_xml, PCMK_META_INTERVAL, &lrm_op_interval_ms);
if ((conf_op_interval_ms != lrm_op_interval_ms)
|| !pcmk__str_eq(conf_op_name, lrm_op_task, pcmk__str_casei)) {
return FALSE;
}
lrm_op_id = pcmk__xe_id(lrm_op_xml);
last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0);
if (pcmk__str_eq(last_failure_key, lrm_op_id, pcmk__str_casei)) {
matched = TRUE;
} else {
char *expected_op_key = pcmk__op_key(rsc_id, conf_op_name,
conf_op_interval_ms);
if (pcmk__str_eq(expected_op_key, lrm_op_id, pcmk__str_casei)) {
int rc = 0;
int target_rc = pe__target_rc_from_xml(lrm_op_xml);
pcmk__xe_get_int(lrm_op_xml, PCMK__XA_RC_CODE, &rc);
if (rc != target_rc) {
matched = TRUE;
}
}
free(expected_op_key);
}
free(last_failure_key);
return matched;
}
static gboolean
block_failure(const pcmk_node_t *node, pcmk_resource_t *rsc,
const xmlNode *xml_op)
{
char *xml_name = clone_strip(rsc->id);
/* @TODO This xpath search occurs after template expansion, but it is unable
* to properly detect on-fail in id-ref, operation meta-attributes, or
* op_defaults, or evaluate rules.
*
* Also, PCMK_META_ON_FAIL defaults to PCMK_VALUE_BLOCK (in
- * unpack_operation()) for stop actions when stonith is disabled.
+ * unpack_operation()) for stop actions when fencing is disabled.
*
* Ideally, we'd unpack the operation before this point, and pass in a
* meta-attributes table that takes all that into consideration.
*/
char *xpath = crm_strdup_printf("//" PCMK_XE_PRIMITIVE
"[@" PCMK_XA_ID "='%s']"
"//" PCMK_XE_OP
"[@" PCMK_META_ON_FAIL
"='" PCMK_VALUE_BLOCK "']",
xml_name);
xmlXPathObject *xpathObj = pcmk__xpath_search(rsc->priv->xml->doc, xpath);
gboolean should_block = FALSE;
free(xpath);
if (xpathObj) {
int max = pcmk__xpath_num_results(xpathObj);
int lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *pref = pcmk__xpath_result(xpathObj, lpc);
if (xml_op) {
should_block = is_matched_failure(xml_name, pref, xml_op);
if (should_block) {
break;
}
} else {
const char *conf_op_name = NULL;
const char *conf_op_interval_spec = NULL;
guint conf_op_interval_ms = 0;
pcmk_scheduler_t *scheduler = rsc->priv->scheduler;
char *lrm_op_xpath = NULL;
xmlXPathObject *lrm_op_xpathObj = NULL;
// Get name and interval from configured op
conf_op_name = pcmk__xe_get(pref, PCMK_XA_NAME);
conf_op_interval_spec = pcmk__xe_get(pref, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(conf_op_interval_spec,
&conf_op_interval_ms);
#define XPATH_FMT "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \
"//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \
"/" PCMK__XE_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='%s']" \
"[@" PCMK_META_INTERVAL "='%u']"
lrm_op_xpath = crm_strdup_printf(XPATH_FMT,
node->priv->name, xml_name,
conf_op_name,
conf_op_interval_ms);
lrm_op_xpathObj = pcmk__xpath_search(scheduler->input->doc,
lrm_op_xpath);
free(lrm_op_xpath);
if (lrm_op_xpathObj) {
int max2 = pcmk__xpath_num_results(lrm_op_xpathObj);
int lpc2 = 0;
for (lpc2 = 0; lpc2 < max2; lpc2++) {
xmlNode *lrm_op_xml = NULL;
lrm_op_xml = pcmk__xpath_result(lrm_op_xpathObj, lpc2);
should_block = is_matched_failure(xml_name, pref,
lrm_op_xml);
if (should_block) {
break;
}
}
}
xmlXPathFreeObject(lrm_op_xpathObj);
if (should_block) {
break;
}
}
}
}
free(xml_name);
xmlXPathFreeObject(xpathObj);
return should_block;
}
/*!
* \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 pcmk_resource_t *rsc)
{
const char *name = pcmk__s(rsc->priv->history_id, rsc->id);
return pcmk_is_set(rsc->flags, pcmk__rsc_unique)? strdup(name) : clone_strip(name);
}
/*!
* \internal
* \brief Compile regular expression to match a failure-related node attribute
*
* \param[in] prefix Attribute prefix to match
* \param[in] rsc_name Resource name to match as used in failure attributes
* \param[in] is_unique Whether the resource is a globally unique clone
* \param[out] re Where to store resulting regular expression
*
* \return Standard Pacemaker return code
* \note Fail attributes are named like PREFIX-RESOURCE#OP_INTERVAL.
* The caller is responsible for freeing re with regfree().
*/
static int
generate_fail_regex(const char *prefix, const char *rsc_name, bool is_unique,
regex_t *re)
{
char *pattern = NULL;
const char *op_pattern = "#.+_[0-9]+";
/* Ignore instance numbers for anything other than globally unique clones.
* Anonymous clone fail counts could contain an instance number if the
* clone was initially unique, failed, then was converted to anonymous.
*/
const char *instance_pattern = (is_unique? "" : "(:[0-9]+)?");
pattern = crm_strdup_printf("^%s-%s%s%s$", prefix, rsc_name,
instance_pattern, op_pattern);
if (regcomp(re, pattern, REG_EXTENDED|REG_NOSUB) != 0) {
free(pattern);
return EINVAL;
}
free(pattern);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Compile regular expressions to match failure-related node attributes
*
* \param[in] rsc Resource being checked for failures
* \param[out] failcount_re Storage for regular expression for fail count
* \param[out] lastfailure_re Storage for regular expression for last failure
*
* \return Standard Pacemaker return code
* \note On success, the caller is responsible for freeing the expressions with
* regfree().
*/
static int
generate_fail_regexes(const pcmk_resource_t *rsc, regex_t *failcount_re,
regex_t *lastfailure_re)
{
int rc = pcmk_rc_ok;
char *rsc_name = rsc_fail_name(rsc);
if (generate_fail_regex(PCMK__FAIL_COUNT_PREFIX, rsc_name,
pcmk_is_set(rsc->flags, pcmk__rsc_unique),
failcount_re) != pcmk_rc_ok) {
rc = EINVAL;
} else if (generate_fail_regex(PCMK__LAST_FAILURE_PREFIX, rsc_name,
pcmk_is_set(rsc->flags, pcmk__rsc_unique),
lastfailure_re) != pcmk_rc_ok) {
rc = EINVAL;
regfree(failcount_re);
}
free(rsc_name);
return rc;
}
// Data for fail-count-related iterators
struct failcount_data {
const pcmk_node_t *node;// Node to check for fail count
pcmk_resource_t *rsc; // Resource to check for fail count
uint32_t flags; // Fail count flags
const xmlNode *xml_op; // History entry for expiration purposes (or NULL)
regex_t failcount_re; // Fail count regular expression to match
regex_t lastfailure_re; // Last failure regular expression to match
int failcount; // Fail count so far
time_t last_failure; // Time of most recent failure so far
};
/*!
* \internal
* \brief Update fail count and last failure appropriately for a node attribute
*
* \param[in] key Node attribute name
* \param[in] value Node attribute value
* \param[in] user_data Fail count data to update
*/
static void
update_failcount_for_attr(gpointer key, gpointer value, gpointer user_data)
{
struct failcount_data *fc_data = user_data;
// If this is a matching fail count attribute, update fail count
if (regexec(&(fc_data->failcount_re), (const char *) key, 0, NULL, 0) == 0) {
int score = 0;
int rc = pcmk_parse_score(value, &score, 0);
if (rc != pcmk_rc_ok) {
crm_warn("Ignoring %s for %s "
"because '%s' is not a valid fail count: %s",
(const char *) key, pcmk__node_name(fc_data->node),
value, pcmk_rc_str(rc));
return;
}
fc_data->failcount = pcmk__add_scores(fc_data->failcount, score);
pcmk__rsc_trace(fc_data->rsc, "Added %s (%s) to %s fail count (now %s)",
(const char *) key, (const char *) value,
fc_data->rsc->id,
pcmk_readable_score(fc_data->failcount));
return;
}
// If this is a matching last failure attribute, update last failure
if (regexec(&(fc_data->lastfailure_re), (const char *) key, 0, NULL,
0) == 0) {
long long last_ll;
int rc = pcmk__scan_ll(value, &last_ll, 0LL);
if (rc != pcmk_rc_ok) {
crm_info("Ignoring invalid value '%s' for %s: %s",
(const char *) value, (const char *) key, pcmk_rc_str(rc));
return;
}
fc_data->last_failure = (time_t) QB_MAX(fc_data->last_failure, last_ll);
}
}
/*!
* \internal
* \brief Update fail count and last failure appropriately for launched resource
*
* \param[in] data Launched resource
* \param[in] user_data Fail count data to update
*/
static void
update_launched_failcount(gpointer data, gpointer user_data)
{
pcmk_resource_t *launched = data;
struct failcount_data *fc_data = user_data;
time_t launched_last_failure = 0;
fc_data->failcount += pe_get_failcount(fc_data->node, launched,
&launched_last_failure,
fc_data->flags, fc_data->xml_op);
fc_data->last_failure = QB_MAX(fc_data->last_failure, launched_last_failure);
}
#define readable_expiration(rsc) \
pcmk__readable_interval((rsc)->priv->failure_expiration_ms)
/*!
* \internal
* \brief Get a resource's fail count on a node
*
* \param[in] node Node to check
* \param[in,out] rsc Resource to check
* \param[out] last_failure If not NULL, where to set time of most recent
* failure of \p rsc on \p node
* \param[in] flags Group of enum pcmk__fc_flags
* \param[in] xml_op If not NULL, consider only the action in this
* history entry when determining whether on-fail
* is configured as "blocked", otherwise consider
* all actions configured for \p rsc
*
* \return Fail count for \p rsc on \p node according to \p flags
*/
int
pe_get_failcount(const pcmk_node_t *node, pcmk_resource_t *rsc,
time_t *last_failure, uint32_t flags, const xmlNode *xml_op)
{
struct failcount_data fc_data = {
.node = node,
.rsc = rsc,
.flags = flags,
.xml_op = xml_op,
.failcount = 0,
.last_failure = (time_t) 0,
};
// Calculate resource failcount as sum of all matching operation failcounts
CRM_CHECK(generate_fail_regexes(rsc, &fc_data.failcount_re,
&fc_data.lastfailure_re) == pcmk_rc_ok,
return 0);
g_hash_table_foreach(node->priv->attrs, update_failcount_for_attr,
&fc_data);
regfree(&(fc_data.failcount_re));
regfree(&(fc_data.lastfailure_re));
// If failure blocks the resource, disregard any failure timeout
if ((fc_data.failcount > 0) && (rsc->priv->failure_expiration_ms > 0)
&& block_failure(node, rsc, xml_op)) {
pcmk__config_warn("Ignoring failure timeout (%s) for %s "
"because it conflicts with "
PCMK_META_ON_FAIL "=" PCMK_VALUE_BLOCK,
readable_expiration(rsc), rsc->id);
rsc->priv->failure_expiration_ms = 0;
}
// If all failures have expired, ignore fail count
if (pcmk_is_set(flags, pcmk__fc_effective) && (fc_data.failcount > 0)
&& (fc_data.last_failure > 0)
&& (rsc->priv->failure_expiration_ms > 0)) {
time_t now = pcmk__scheduler_epoch_time(rsc->priv->scheduler);
const guint expiration = pcmk__timeout_ms2s(rsc->priv->failure_expiration_ms);
if (now > (fc_data.last_failure + expiration)) {
pcmk__rsc_debug(rsc, "Failcount for %s on %s expired after %s",
rsc->id, pcmk__node_name(node),
readable_expiration(rsc));
fc_data.failcount = 0;
}
}
/* Add the fail count of any launched resources, except that we never want
* the fail counts of a bundle container's launched resources to count
* towards the container's fail count.
*
* Most importantly, a Pacemaker Remote connection to a bundle container
* is launched by the container, but can reside on a different node than the
* container itself. Counting its fail count on its node towards the
* container's fail count on that node could lead to attempting to stop the
* container on the wrong node.
*/
if (pcmk_is_set(flags, pcmk__fc_launched)
&& (rsc->priv->launched != NULL) && !pcmk__is_bundled(rsc)) {
g_list_foreach(rsc->priv->launched, update_launched_failcount,
&fc_data);
if (fc_data.failcount > 0) {
pcmk__rsc_info(rsc,
"Container %s and the resources within it "
"have failed %s time%s on %s",
rsc->id, pcmk_readable_score(fc_data.failcount),
pcmk__plural_s(fc_data.failcount),
pcmk__node_name(node));
}
} else if (fc_data.failcount > 0) {
pcmk__rsc_info(rsc, "%s has failed %s time%s on %s",
rsc->id, pcmk_readable_score(fc_data.failcount),
pcmk__plural_s(fc_data.failcount),
pcmk__node_name(node));
}
if (last_failure != NULL) {
if ((fc_data.failcount > 0) && (fc_data.last_failure > 0)) {
*last_failure = fc_data.last_failure;
} else {
*last_failure = 0;
}
}
return fc_data.failcount;
}
/*!
* \brief Schedule a controller operation to clear a fail count
*
* \param[in,out] rsc Resource with failure
* \param[in] node Node failure occurred on
* \param[in] reason Readable description why needed (for logging)
* \param[in,out] scheduler Scheduler data cluster
*
* \return Scheduled action
*/
pcmk_action_t *
pe__clear_failcount(pcmk_resource_t *rsc, const pcmk_node_t *node,
const char *reason, pcmk_scheduler_t *scheduler)
{
char *key = NULL;
pcmk_action_t *clear = NULL;
CRM_CHECK(rsc && node && reason && scheduler, return NULL);
key = pcmk__op_key(rsc->id, PCMK_ACTION_CLEAR_FAILCOUNT, 0);
clear = custom_action(rsc, key, PCMK_ACTION_CLEAR_FAILCOUNT, node, FALSE,
scheduler);
pcmk__insert_meta(clear, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE);
crm_notice("Clearing failure of %s on %s because %s " QB_XS " %s",
rsc->id, pcmk__node_name(node), reason, clear->uuid);
return clear;
}
diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c
index 0db4290175..6be34b23be 100644
--- a/lib/pengine/pe_actions.c
+++ b/lib/pengine/pe_actions.c
@@ -1,1750 +1,1749 @@
/*
* Copyright 2004-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/scheduler_internal.h>
#include <crm/pengine/internal.h>
#include <crm/common/xml_internal.h>
#include "pe_status_private.h"
static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj,
guint interval_ms);
static void
add_singleton(pcmk_scheduler_t *scheduler, pcmk_action_t *action)
{
if (scheduler->priv->singletons == NULL) {
scheduler->priv->singletons = pcmk__strkey_table(NULL, NULL);
}
g_hash_table_insert(scheduler->priv->singletons, action->uuid, action);
}
static pcmk_action_t *
lookup_singleton(pcmk_scheduler_t *scheduler, const char *action_uuid)
{
/* @TODO This is the only use of the pcmk_scheduler_t:singletons hash table.
* Compare the performance of this approach to keeping the
* pcmk_scheduler_t:actions list sorted by action key and just searching
* that instead.
*/
if (scheduler->priv->singletons == NULL) {
return NULL;
}
return g_hash_table_lookup(scheduler->priv->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] scheduler Scheduler data
*
* \return Existing action that matches arguments (or NULL if none)
*/
static pcmk_action_t *
find_existing_action(const char *key, const pcmk_resource_t *rsc,
const pcmk_node_t *node, const pcmk_scheduler_t *scheduler)
{
/* When rsc is NULL, it would be quicker to check
* scheduler->priv->singletons, but checking all scheduler->priv->actions
* takes the node into account.
*/
GList *actions = (rsc == NULL)? scheduler->priv->actions : rsc->priv->actions;
GList *matches = find_actions(actions, key, node);
pcmk_action_t *action = NULL;
if (matches == NULL) {
return NULL;
}
CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches));
action = matches->data;
g_list_free(matches);
return action;
}
/*!
* \internal
* \brief Find the XML configuration corresponding to a specific action key
*
* \param[in] rsc Resource to find action configuration for
* \param[in] key "RSC_ACTION_INTERVAL" of action to find
* \param[in] include_disabled If false, do not return disabled actions
*
* \return XML configuration of desired action if any, otherwise NULL
*/
static xmlNode *
find_exact_action_config(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, bool include_disabled)
{
for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
bool enabled = false;
const char *config_name = NULL;
const char *interval_spec = NULL;
guint tmp_ms = 0U;
// @TODO This does not consider meta-attributes, rules, defaults, etc.
if (!include_disabled
&& (pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
interval_spec = pcmk__xe_get(operation, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &tmp_ms);
if (tmp_ms != interval_ms) {
continue;
}
config_name = pcmk__xe_get(operation, PCMK_XA_NAME);
if (pcmk__str_eq(action_name, config_name, pcmk__str_none)) {
return operation;
}
}
return NULL;
}
/*!
* \internal
* \brief Find the XML configuration of a resource action
*
* \param[in] rsc Resource to find action configuration for
* \param[in] action_name Action name to search for
* \param[in] interval_ms Action interval (in milliseconds) to search for
* \param[in] include_disabled If false, do not return disabled actions
*
* \return XML configuration of desired action if any, otherwise NULL
*/
xmlNode *
pcmk__find_action_config(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, bool include_disabled)
{
xmlNode *action_config = NULL;
// Try requested action first
action_config = find_exact_action_config(rsc, action_name, interval_ms,
include_disabled);
// For migrate_to and migrate_from actions, retry with "migrate"
// @TODO This should be either documented or deprecated
if ((action_config == NULL)
&& pcmk__str_any_of(action_name, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
action_config = find_exact_action_config(rsc, "migrate", 0,
include_disabled);
}
return action_config;
}
/*!
* \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,out] scheduler Scheduler data
*
* \return Newly allocated action
* \note This function takes ownership of \p key. It is the caller's
* responsibility to free the return value using pcmk__free_action().
*/
static pcmk_action_t *
new_action(char *key, const char *task, pcmk_resource_t *rsc,
const pcmk_node_t *node, bool optional, pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = pcmk__assert_alloc(1, sizeof(pcmk_action_t));
action->rsc = rsc;
action->task = pcmk__str_copy(task);
action->uuid = key;
action->scheduler = scheduler;
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
pcmk__set_action_flags(action, pcmk__action_on_dc);
}
pcmk__set_action_flags(action, pcmk__action_runnable);
if (optional) {
pcmk__set_action_flags(action, pcmk__action_optional);
} else {
pcmk__clear_action_flags(action, pcmk__action_optional);
}
if (rsc == NULL) {
action->meta = pcmk__strkey_table(free, free);
} else {
guint interval_ms = 0;
parse_op_key(key, NULL, NULL, &interval_ms);
action->op_entry = pcmk__find_action_config(rsc, task, interval_ms,
true);
/* If the given key is for one of the many notification pseudo-actions
* (pre_notify_promote, etc.), the actual action name is "notify"
*/
if ((action->op_entry == NULL) && (strstr(key, "_notify_") != NULL)) {
action->op_entry = find_exact_action_config(rsc, PCMK_ACTION_NOTIFY,
0, true);
}
unpack_operation(action, action->op_entry, interval_ms);
}
pcmk__rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s",
(optional? "optional" : "required"),
scheduler->priv->next_action_id, key, task,
((rsc == NULL)? "no resource" : rsc->id),
pcmk__node_name(node));
action->id = scheduler->priv->next_action_id++;
scheduler->priv->actions = g_list_prepend(scheduler->priv->actions, action);
if (rsc == NULL) {
add_singleton(scheduler, action);
} else {
rsc->priv->actions = g_list_prepend(rsc->priv->actions, action);
}
return action;
}
/*!
* \internal
* \brief Unpack a resource's action-specific instance parameters
*
* \param[in] action_xml XML of action's configuration in CIB (if any)
* \param[in,out] node_attrs Table of node attributes (for rule evaluation)
* \param[in,out] scheduler Scheduler data (for rule evaluation)
*
* \return Newly allocated hash table of action-specific instance parameters
*/
GHashTable *
pcmk__unpack_action_rsc_params(const xmlNode *action_xml,
GHashTable *node_attrs,
pcmk_scheduler_t *scheduler)
{
GHashTable *params = pcmk__strkey_table(free, free);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
.node_attrs = node_attrs,
};
pe__unpack_dataset_nvpairs(action_xml, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, params, NULL, scheduler);
return params;
}
/*!
* \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(pcmk_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, pcmk__action_pseudo)
&& !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed)
&& (g_hash_table_lookup(action->meta, PCMK_META_INTERVAL) == NULL)) {
pcmk__rsc_debug(action->rsc,
"%s on %s is optional (%s is unmanaged)",
action->uuid, pcmk__node_name(action->node),
action->rsc->id);
pcmk__set_action_flags(action, pcmk__action_optional);
// We shouldn't clear runnable here because ... something
// Otherwise require the action if requested
} else if (!optional) {
pcmk__clear_action_flags(action, pcmk__action_optional);
}
}
static enum pe_quorum_policy
effective_quorum_policy(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
enum pe_quorum_policy policy = scheduler->no_quorum_policy;
if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
policy = pcmk_no_quorum_ignore;
} else if (scheduler->no_quorum_policy == pcmk_no_quorum_demote) {
switch (rsc->priv->orig_role) {
case pcmk_role_promoted:
case pcmk_role_unpromoted:
if (rsc->priv->next_role > pcmk_role_unpromoted) {
pe__set_next_role(rsc, pcmk_role_unpromoted,
PCMK_OPT_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,out] scheduler Scheduler data
*
* \note This may also schedule fencing if a stop is unrunnable.
*/
static void
update_resource_action_runnable(pcmk_action_t *action,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = action->rsc;
if (pcmk_is_set(action->flags, pcmk__action_pseudo)) {
return;
}
if (action->node == NULL) {
pcmk__rsc_trace(rsc, "%s is unrunnable (unallocated)", action->uuid);
pcmk__clear_action_flags(action, pcmk__action_runnable);
} else if (!pcmk_is_set(action->flags, pcmk__action_on_dc)
&& !(action->node->details->online)
&& (!pcmk__is_guest_or_bundle_node(action->node)
|| pcmk_is_set(action->node->priv->flags,
pcmk__node_remote_reset))) {
pcmk__clear_action_flags(action, pcmk__action_runnable);
do_crm_log(LOG_WARNING, "%s on %s is unrunnable (node is offline)",
action->uuid, pcmk__node_name(action->node));
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)
&& pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei)
&& !(action->node->details->unclean)) {
pe_fence_node(scheduler, action->node, "stop is unrunnable", false);
}
} else if (!pcmk_is_set(action->flags, pcmk__action_on_dc)
&& action->node->details->pending) {
pcmk__clear_action_flags(action, pcmk__action_runnable);
do_crm_log(LOG_WARNING,
"Action %s on %s is unrunnable (node is pending)",
action->uuid, pcmk__node_name(action->node));
} else if (action->needs == pcmk__requires_nothing) {
pe_action_set_reason(action, NULL, TRUE);
if (pcmk__is_guest_or_bundle_node(action->node)
&& !pe_can_fence(scheduler, 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.
*/
pcmk__rsc_debug(rsc,
"%s on %s is unrunnable "
"(node's host cannot be fenced)",
action->uuid, pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
} else {
pcmk__rsc_trace(rsc,
"%s on %s does not require fencing or quorum",
action->uuid, pcmk__node_name(action->node));
pcmk__set_action_flags(action, pcmk__action_runnable);
}
} else {
switch (effective_quorum_policy(rsc, scheduler)) {
case pcmk_no_quorum_stop:
pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)",
action->uuid, pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
pe_action_set_reason(action, "no quorum", true);
break;
case pcmk_no_quorum_freeze:
if (!rsc->priv->fns->active(rsc, true)
|| (rsc->priv->next_role > rsc->priv->orig_role)) {
pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)",
action->uuid,
pcmk__node_name(action->node));
pcmk__clear_action_flags(action, pcmk__action_runnable);
pe_action_set_reason(action, "quorum freeze", true);
}
break;
default:
//pe_action_set_reason(action, NULL, TRUE);
pcmk__set_action_flags(action, pcmk__action_runnable);
break;
}
}
}
static bool
valid_stop_on_fail(const char *value)
{
return !pcmk__strcase_any_of(value,
PCMK_VALUE_STANDBY, PCMK_VALUE_DEMOTE,
PCMK_VALUE_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
*/
static void
validate_on_fail(const pcmk_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, PCMK_META_ON_FAIL);
guint interval_ms = 0U;
// 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 '" PCMK_META_ON_FAIL "' for %s stop "
"action to default value because '%s' is not "
"allowed for stop", rsc->id, value);
g_hash_table_remove(meta, PCMK_META_ON_FAIL);
return;
}
/* 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 = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL;
operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
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 = pcmk__xe_get(operation, PCMK_META_ON_FAIL);
if (promote_on_fail == NULL) {
continue;
}
// We only care about recurring monitors for the promoted role
name = pcmk__xe_get(operation, PCMK_XA_NAME);
role = pcmk__xe_get(operation, PCMK_XA_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 = pcmk__xe_get(operation, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (interval_ms == 0U) {
continue;
}
// We only care about enabled monitors
if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
/* Demote actions can't default to
* PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE
*/
if (pcmk__str_eq(promote_on_fail, PCMK_VALUE_DEMOTE,
pcmk__str_casei)) {
continue;
}
// Use value from first applicable promote action found
pcmk__insert_dup(meta, PCMK_META_ON_FAIL, promote_on_fail);
}
return;
}
if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none)
&& !pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) {
pcmk__insert_dup(meta, PCMK_META_ON_FAIL, PCMK_VALUE_IGNORE);
return;
}
// PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE is allowed only for certain actions
if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
name = pcmk__xe_get(action_config, PCMK_XA_NAME);
role = pcmk__xe_get(action_config, PCMK_XA_ROLE);
interval_spec = pcmk__xe_get(action_config, PCMK_META_INTERVAL);
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none)
&& ((interval_ms == 0U)
|| !pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none)
|| !pcmk__strcase_any_of(role, PCMK_ROLE_PROMOTED,
PCMK__ROLE_PROMOTED_LEGACY, NULL))) {
pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s %s "
"action to default value because 'demote' is not "
"allowed for it", rsc->id, name);
g_hash_table_remove(meta, PCMK_META_ON_FAIL);
return;
}
}
}
static int
unpack_timeout(const char *value)
{
long long timeout_ms = 0;
if ((value == NULL) || (pcmk__parse_ms(value, &timeout_ms) != pcmk_rc_ok)
|| (timeout_ms <= 0)) {
timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS;
}
return (int) QB_MIN(timeout_ms, INT_MAX);
}
// 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 = pcmk__timeout_ms2s(interval_ms);
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 '" PCMK_META_INTERVAL_ORIGIN "' for "
"operation '%s' because '%s' is not valid",
pcmk__s(pcmk__xe_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, pcmk__s(pcmk__xe_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)
{
long long start_delay_ms = 0;
if (value == NULL) {
return 0;
}
if (pcmk__parse_ms(value, &start_delay_ms) == pcmk_rc_ok) {
start_delay_ms = QB_MAX(start_delay_ms, 0);
start_delay_ms = QB_MIN(start_delay_ms, INT_MAX);
}
if (meta != NULL) {
g_hash_table_replace(meta, strdup(PCMK_META_START_DELAY),
pcmk__itoa((int) start_delay_ms));
}
return (int) start_delay_ms;
}
/*!
* \internal
* \brief Find a resource's most frequent recurring monitor
*
* \param[in] rsc Resource to check
*
* \return Operation XML configured for most frequent recurring monitor for
* \p rsc (if any)
*/
static xmlNode *
most_frequent_monitor(const pcmk_resource_t *rsc)
{
guint min_interval_ms = G_MAXUINT;
xmlNode *op = NULL;
for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml,
PCMK_XE_OP, NULL, NULL);
operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) {
bool enabled = false;
guint interval_ms = 0U;
const char *interval_spec = pcmk__xe_get(operation, PCMK_META_INTERVAL);
// We only care about enabled recurring monitors
if (!pcmk__str_eq(pcmk__xe_get(operation, PCMK_XA_NAME),
PCMK_ACTION_MONITOR, pcmk__str_none)) {
continue;
}
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (interval_ms == 0U) {
continue;
}
// @TODO This does not consider meta-attributes, rules, defaults, etc.
if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED,
&enabled) == pcmk_rc_ok) && !enabled) {
continue;
}
if (interval_ms < min_interval_ms) {
min_interval_ms = interval_ms;
op = operation;
}
}
return op;
}
/*!
* \internal
* \brief Unpack action meta-attributes
*
* \param[in,out] rsc Resource that action is for
* \param[in] node Node that action is on
* \param[in] action_name Action name
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in] action_config Action XML configuration from CIB (if any)
*
* Unpack a resource action's meta-attributes (normalizing the interval,
* timeout, and start delay values as integer milliseconds) from its CIB XML
* configuration (including defaults).
*
* \return Newly allocated hash table with normalized action meta-attributes
*/
GHashTable *
pcmk__unpack_action_meta(pcmk_resource_t *rsc, const pcmk_node_t *node,
const char *action_name, guint interval_ms,
const xmlNode *action_config)
{
GHashTable *meta = NULL;
const char *timeout_spec = NULL;
const char *str = NULL;
const pcmk_rule_input_t rule_input = {
/* Node attributes are not set because node expressions are not allowed
* for meta-attributes
*/
.now = rsc->priv->scheduler->priv->now,
.rsc_standard = pcmk__xe_get(rsc->priv->xml, PCMK_XA_CLASS),
.rsc_provider = pcmk__xe_get(rsc->priv->xml, PCMK_XA_PROVIDER),
.rsc_agent = pcmk__xe_get(rsc->priv->xml, PCMK_XA_TYPE),
.op_name = action_name,
.op_interval_ms = interval_ms,
};
meta = pcmk__strkey_table(free, free);
if (action_config != NULL) {
// <op> <meta_attributes> take precedence over defaults
pe__unpack_dataset_nvpairs(action_config, PCMK_XE_META_ATTRIBUTES,
&rule_input, meta, NULL,
rsc->priv->scheduler);
/* Anything set as an <op> XML property has highest precedence.
* This ensures we use the name and interval from the <op> tag.
* (See below for the only exception, fence device start/probe timeout.)
*/
for (xmlAttrPtr attr = action_config->properties;
attr != NULL; attr = attr->next) {
pcmk__insert_dup(meta, (const char *) attr->name,
pcmk__xml_attr_value(attr));
}
}
// Derive default timeout for probes from recurring monitor timeouts
if (pcmk_is_probe(action_name, interval_ms)
&& (g_hash_table_lookup(meta, PCMK_META_TIMEOUT) == NULL)) {
xmlNode *min_interval_mon = most_frequent_monitor(rsc);
if (min_interval_mon != NULL) {
/* @TODO This does not consider timeouts set in
* PCMK_XE_META_ATTRIBUTES blocks (which may also have rules that
* need to be evaluated).
*/
timeout_spec = pcmk__xe_get(min_interval_mon, PCMK_META_TIMEOUT);
if (timeout_spec != NULL) {
pcmk__rsc_trace(rsc,
"Setting default timeout for %s probe to "
"most frequent monitor's timeout '%s'",
rsc->id, timeout_spec);
pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec);
}
}
}
// Cluster-wide <op_defaults> <meta_attributes>
pe__unpack_dataset_nvpairs(rsc->priv->scheduler->priv->op_defaults,
PCMK_XE_META_ATTRIBUTES, &rule_input, meta, NULL,
rsc->priv->scheduler);
g_hash_table_remove(meta, PCMK_XA_ID);
// Normalize interval to milliseconds
if (interval_ms > 0) {
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_INTERVAL),
crm_strdup_printf("%u", interval_ms));
} else {
g_hash_table_remove(meta, PCMK_META_INTERVAL);
}
/* Timeout order of precedence (highest to lowest):
* 1. pcmk_monitor_timeout resource parameter (only for starts and probes
* when rsc has pcmk_ra_cap_fence_params; this gets used for recurring
* monitors via the executor instead)
* 2. timeout configured in <op> (with <op timeout> taking precedence over
* <op> <meta_attributes>)
* 3. timeout configured in <op_defaults> <meta_attributes>
* 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS
*/
// Check for pcmk_monitor_timeout
if (pcmk_is_set(pcmk_get_ra_caps(rule_input.rsc_standard),
pcmk_ra_cap_fence_params)
&& (pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)
|| pcmk_is_probe(action_name, interval_ms))) {
GHashTable *params = pe_rsc_params(rsc, node, rsc->priv->scheduler);
timeout_spec = g_hash_table_lookup(params, "pcmk_monitor_timeout");
if (timeout_spec != NULL) {
pcmk__rsc_trace(rsc,
"Setting timeout for %s %s to "
"pcmk_monitor_timeout (%s)",
rsc->id, action_name, timeout_spec);
pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec);
}
}
// Normalize timeout to positive milliseconds
timeout_spec = g_hash_table_lookup(meta, PCMK_META_TIMEOUT);
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_TIMEOUT),
pcmk__itoa(unpack_timeout(timeout_spec)));
// Ensure on-fail has a valid value
validate_on_fail(rsc, action_name, action_config, meta);
// Normalize PCMK_META_START_DELAY
str = g_hash_table_lookup(meta, PCMK_META_START_DELAY);
if (str != NULL) {
unpack_start_delay(str, meta);
} else {
long long start_delay = 0;
str = g_hash_table_lookup(meta, PCMK_META_INTERVAL_ORIGIN);
if (unpack_interval_origin(str, action_config, interval_ms,
rsc->priv->scheduler->priv->now,
&start_delay)) {
g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_START_DELAY),
crm_strdup_printf("%lld", start_delay));
}
}
return meta;
}
/*!
* \internal
* \brief Determine an action's quorum and fencing dependency
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Name of action being unpacked
*
* \return Quorum and fencing dependency appropriate to action
*/
enum pcmk__requires
pcmk__action_requires(const pcmk_resource_t *rsc, const char *action_name)
{
const char *value = NULL;
enum pcmk__requires requires = pcmk__requires_nothing;
CRM_CHECK((rsc != NULL) && (action_name != NULL), return requires);
if (!pcmk__strcase_any_of(action_name, PCMK_ACTION_START,
PCMK_ACTION_PROMOTE, NULL)) {
value = "nothing (not start or promote)";
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) {
requires = pcmk__requires_fencing;
value = "fencing";
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_quorum)) {
requires = pcmk__requires_quorum;
value = "quorum";
} else {
value = "nothing";
}
pcmk__rsc_trace(rsc, "%s of %s requires %s", action_name, rsc->id, value);
return requires;
}
/*!
* \internal
* \brief Parse action failure response from a user-provided string
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Name of action
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in] value User-provided configuration value for on-fail
*
* \return Action failure response parsed from \p text
*/
enum pcmk__on_fail
pcmk__parse_on_fail(const pcmk_resource_t *rsc, const char *action_name,
guint interval_ms, const char *value)
{
const char *desc = NULL;
bool needs_remote_reset = false;
enum pcmk__on_fail on_fail = pcmk__on_fail_ignore;
const pcmk_scheduler_t *scheduler = NULL;
// There's no enum value for unknown or invalid, so assert
pcmk__assert((rsc != NULL) && (action_name != NULL));
scheduler = rsc->priv->scheduler;
if (value == NULL) {
// Use default
} else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) {
on_fail = pcmk__on_fail_block;
desc = "block";
} else if (pcmk__str_eq(value, PCMK_VALUE_FENCE, pcmk__str_casei)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
on_fail = pcmk__on_fail_fence_node;
desc = "node fencing";
} else {
pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for "
"%s of %s to 'stop' because 'fence' is not "
"valid when fencing is disabled",
action_name, rsc->id);
/* @TODO This should probably do
g_hash_table_remove(meta, PCMK_META_ON_FAIL);
like the other "Resetting" spots, to avoid repeating the message
*/
on_fail = pcmk__on_fail_stop;
desc = "stop resource";
}
} else if (pcmk__str_eq(value, PCMK_VALUE_STANDBY, pcmk__str_casei)) {
on_fail = pcmk__on_fail_standby_node;
desc = "node standby";
} else if (pcmk__strcase_any_of(value,
PCMK_VALUE_IGNORE, PCMK_VALUE_NOTHING,
NULL)) {
desc = "ignore";
} else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) {
on_fail = pcmk__on_fail_ban;
desc = "force migration";
} else if (pcmk__str_eq(value, PCMK_VALUE_STOP, pcmk__str_casei)) {
on_fail = pcmk__on_fail_stop;
desc = "stop resource";
} else if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) {
on_fail = pcmk__on_fail_restart;
desc = "restart (and possibly migrate)";
} else if (pcmk__str_eq(value, PCMK_VALUE_RESTART_CONTAINER,
pcmk__str_casei)) {
if (rsc->priv->launcher == NULL) {
pcmk__rsc_debug(rsc,
"Using default " PCMK_META_ON_FAIL " for %s "
"of %s because it does not have a launcher",
action_name, rsc->id);
} else {
on_fail = pcmk__on_fail_restart_container;
desc = "restart container (and possibly migrate)";
}
} else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
on_fail = pcmk__on_fail_demote;
desc = "demote instance";
} else {
pcmk__config_err("Using default '" PCMK_META_ON_FAIL "' for "
"%s of %s because '%s' is not valid",
action_name, rsc->id, value);
}
/* Remote node connections are handled specially. Failures that result
* in dropping an active connection must result in fencing. The only
* failures that don't are probes and starts. The user can explicitly set
* PCMK_META_ON_FAIL=PCMK_VALUE_FENCE to fence after start failures.
*/
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)
&& pcmk__is_remote_node(pcmk_find_node(scheduler, rsc->id))
&& !pcmk_is_probe(action_name, interval_ms)
&& !pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)) {
needs_remote_reset = true;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
desc = NULL; // Force default for unmanaged connections
}
}
if (desc != NULL) {
// Explicit value used, default not needed
} else if (rsc->priv->launcher != NULL) {
on_fail = pcmk__on_fail_restart_container;
desc = "restart container (and possibly migrate) (default)";
} else if (needs_remote_reset) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
desc = "fence remote node (default)";
} else {
desc = "recover remote node connection (default)";
}
on_fail = pcmk__on_fail_reset_remote;
} else {
on_fail = pcmk__on_fail_stop;
desc = "stop unmanaged remote node (enforcing default)";
}
} else if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
on_fail = pcmk__on_fail_fence_node;
desc = "resource fence (default)";
} else {
on_fail = pcmk__on_fail_block;
desc = "resource block (default)";
}
} else {
on_fail = pcmk__on_fail_restart;
desc = "restart (and possibly migrate) (default)";
}
pcmk__rsc_trace(rsc, "Failure handling for %s-interval %s of %s: %s",
pcmk__readable_interval(interval_ms), action_name,
rsc->id, desc);
return on_fail;
}
/*!
* \internal
* \brief Determine a resource's role after failure of an action
*
* \param[in] rsc Resource that action is for
* \param[in] action_name Action name
* \param[in] on_fail Failure handling for action
* \param[in] meta Unpacked action meta-attributes
*
* \return Resource role that results from failure of action
*/
enum rsc_role_e
pcmk__role_after_failure(const pcmk_resource_t *rsc, const char *action_name,
enum pcmk__on_fail on_fail, GHashTable *meta)
{
enum rsc_role_e role = pcmk_role_unknown;
// Set default for role after failure specially in certain circumstances
switch (on_fail) {
case pcmk__on_fail_stop:
role = pcmk_role_stopped;
break;
case pcmk__on_fail_reset_remote:
if (rsc->priv->remote_reconnect_ms != 0U) {
role = pcmk_role_stopped;
}
break;
default:
break;
}
if (role == pcmk_role_unknown) {
// Use default
if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
role = pcmk_role_unpromoted;
} else {
role = pcmk_role_started;
}
}
pcmk__rsc_trace(rsc, "Role after %s %s failure is: %s",
rsc->id, action_name, pcmk_role_text(role));
return role;
}
/*!
* \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] interval_ms How frequently to perform the operation
*/
static void
unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj,
guint interval_ms)
{
const char *value = NULL;
action->meta = pcmk__unpack_action_meta(action->rsc, action->node,
action->task, interval_ms, xml_obj);
action->needs = pcmk__action_requires(action->rsc, action->task);
value = g_hash_table_lookup(action->meta, PCMK_META_ON_FAIL);
action->on_fail = pcmk__parse_on_fail(action->rsc, action->task,
interval_ms, value);
action->fail_role = pcmk__role_after_failure(action->rsc, action->task,
action->on_fail, action->meta);
}
/*!
* \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,out] scheduler Scheduler data
*
* \return Action object corresponding to arguments (guaranteed not to be
* \c NULL)
* \note This function takes ownership of (and might free) \p key, and
* \p scheduler takes ownership of the returned action (the caller should
* not free it).
*/
pcmk_action_t *
custom_action(pcmk_resource_t *rsc, char *key, const char *task,
const pcmk_node_t *on_node, gboolean optional,
pcmk_scheduler_t *scheduler)
{
pcmk_action_t *action = NULL;
pcmk__assert((key != NULL) && (task != NULL) && (scheduler != NULL));
action = find_existing_action(key, rsc, on_node, scheduler);
if (action == NULL) {
action = new_action(key, task, rsc, on_node, optional, scheduler);
} else {
free(key);
}
update_action_optional(action, optional);
if (rsc != NULL) {
/* An action can be initially created with a NULL node, and later have
* the node added via find_existing_action() (above) -> find_actions().
* That is why the extra parameters are unpacked here rather than in
* new_action().
*/
if ((action->node != NULL) && (action->op_entry != NULL)
&& !pcmk_is_set(action->flags, pcmk__action_attrs_evaluated)) {
GHashTable *attrs = action->node->priv->attrs;
if (action->extra != NULL) {
g_hash_table_destroy(action->extra);
}
action->extra = pcmk__unpack_action_rsc_params(action->op_entry,
attrs, scheduler);
pcmk__set_action_flags(action, pcmk__action_attrs_evaluated);
}
update_resource_action_runnable(action, scheduler);
}
if (action->extra == NULL) {
action->extra = pcmk__strkey_table(free, free);
}
return action;
}
pcmk_action_t *
get_pseudo_op(const char *name, pcmk_scheduler_t *scheduler)
{
pcmk_action_t *op = lookup_singleton(scheduler, name);
if (op == NULL) {
op = custom_action(NULL, strdup(name), name, NULL, TRUE, scheduler);
pcmk__set_action_flags(op, pcmk__action_pseudo|pcmk__action_runnable);
}
return op;
}
static GList *
find_unfencing_devices(GList *candidates, GList *matches)
{
for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *candidate = gIter->data;
if (candidate->priv->children != NULL) {
matches = find_unfencing_devices(candidate->priv->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->priv->meta,
PCMK_FENCING_PROVIDES),
PCMK_VALUE_UNFENCING, pcmk__str_casei)) {
matches = g_list_prepend(matches, candidate);
}
}
return matches;
}
static int
node_priority_fencing_delay(const pcmk_node_t *node,
const pcmk_scheduler_t *scheduler)
{
int member_count = 0;
int online_count = 0;
int top_priority = 0;
int lowest_priority = 0;
GList *gIter = NULL;
// PCMK_OPT_PRIORITY_FENCING_DELAY is disabled
if (scheduler->priv->priority_fencing_ms == 0U) {
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->priv->variant != 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 = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *n = gIter->data;
if (n->priv->variant != pcmk__node_variant_cluster) {
continue;
}
member_count ++;
if (n->details->online) {
online_count++;
}
if (member_count == 1
|| n->priv->priority > top_priority) {
top_priority = n->priv->priority;
}
if (member_count == 1
|| n->priv->priority < lowest_priority) {
lowest_priority = n->priv->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->priv->priority < top_priority) {
return 0;
}
return pcmk__timeout_ms2s(scheduler->priv->priority_fencing_ms);
}
pcmk_action_t *
pe_fence_op(pcmk_node_t *node, const char *op, bool optional,
const char *reason, bool priority_delay,
pcmk_scheduler_t *scheduler)
{
char *op_key = NULL;
- pcmk_action_t *stonith_op = NULL;
+ pcmk_action_t *fencing_op = NULL;
if(op == NULL) {
op = scheduler->priv->fence_action;
}
op_key = crm_strdup_printf("%s-%s-%s",
PCMK_ACTION_STONITH, node->priv->name, op);
- stonith_op = lookup_singleton(scheduler, op_key);
- if(stonith_op == NULL) {
- stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node,
+ fencing_op = lookup_singleton(scheduler, op_key);
+ if (fencing_op == NULL) {
+ fencing_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node,
TRUE, scheduler);
- pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE, node->priv->name);
- pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE_UUID,
- node->priv->id);
- pcmk__insert_meta(stonith_op, PCMK__META_STONITH_ACTION, op);
+ pcmk__insert_meta(fencing_op, PCMK__META_ON_NODE, node->priv->name);
+ pcmk__insert_meta(fencing_op, PCMK__META_ON_NODE_UUID, node->priv->id);
+ pcmk__insert_meta(fencing_op, PCMK__META_STONITH_ACTION, op);
if (pcmk_is_set(scheduler->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(scheduler->priv->resources,
NULL);
for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *match = gIter->data;
const char *agent = g_hash_table_lookup(match->priv->meta,
PCMK_XA_TYPE);
pcmk__op_digest_t *data = NULL;
data = pe__compare_fencing_digest(match, agent, node,
scheduler);
if (data->rc == pcmk__digest_mismatch) {
optional = FALSE;
crm_notice("Unfencing node %s because the definition of "
"%s changed", pcmk__node_name(node), match->id);
if (!pcmk__is_daemon && (scheduler->priv->out != NULL)) {
pcmk__output_t *out = scheduler->priv->out;
out->info(out,
"notice: Unfencing node %s because the "
"definition of %s changed",
pcmk__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);
}
- pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_ALL,
+ pcmk__insert_dup(fencing_op->meta, PCMK__META_DIGESTS_ALL,
digests_all->str);
g_string_free(digests_all, TRUE);
- pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_SECURE,
+ pcmk__insert_dup(fencing_op->meta, PCMK__META_DIGESTS_SECURE,
digests_secure->str);
g_string_free(digests_secure, TRUE);
g_list_free(matches);
}
} else {
free(op_key);
}
if ((scheduler->priv->priority_fencing_ms > 0U)
/* It's a suitable case where PCMK_OPT_PRIORITY_FENCING_DELAY
* applies. At least add PCMK_OPT_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,
+ || g_hash_table_lookup(fencing_op->meta,
PCMK_OPT_PRIORITY_FENCING_DELAY) != NULL)) {
/* Add PCMK_OPT_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,
scheduler));
- g_hash_table_insert(stonith_op->meta,
+ g_hash_table_insert(fencing_op->meta,
strdup(PCMK_OPT_PRIORITY_FENCING_DELAY),
delay_s);
}
if(optional == FALSE && pe_can_fence(scheduler, node)) {
- pcmk__clear_action_flags(stonith_op, pcmk__action_optional);
- pe_action_set_reason(stonith_op, reason, false);
+ pcmk__clear_action_flags(fencing_op, pcmk__action_optional);
+ pe_action_set_reason(fencing_op, reason, false);
- } else if(reason && stonith_op->reason == NULL) {
- stonith_op->reason = strdup(reason);
+ } else if ((reason != NULL) && (fencing_op->reason == NULL)) {
+ fencing_op->reason = strdup(reason);
}
- return stonith_op;
+ return fencing_op;
}
enum pcmk__action_type
get_complex_task(const pcmk_resource_t *rsc, const char *name)
{
enum pcmk__action_type task = pcmk__parse_action(name);
if (pcmk__is_primitive(rsc)) {
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
*/
pcmk_action_t *
find_first_action(const GList *input, const char *uuid, const char *task,
const pcmk_node_t *on_node)
{
CRM_CHECK(uuid || task, return NULL);
for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *action = (pcmk_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 (pcmk__same_node(on_node, action->node)) {
return action;
}
}
return NULL;
}
GList *
find_actions(GList *input, const char *key, const pcmk_node_t *on_node)
{
GList *gIter = input;
GList *result = NULL;
CRM_CHECK(key != NULL, return NULL);
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *action = (pcmk_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, pcmk__node_name(on_node));
action->node = pe__copy_node(on_node);
result = g_list_prepend(result, action);
} else if (pcmk__same_node(on_node, action->node)) {
crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node));
result = g_list_prepend(result, action);
}
}
return result;
}
GList *
find_actions_exact(GList *input, const char *key, const pcmk_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) {
pcmk_action_t *action = (pcmk_action_t *) gIter->data;
if ((action->node != NULL)
&& pcmk__str_eq(key, action->uuid, pcmk__str_casei)
&& pcmk__same_node(on_node, action->node)) {
crm_trace("Action %s on %s matches", key, pcmk__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 pcmk_resource_t *rsc, const pcmk_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->priv->actions, key, node);
} else {
result = find_actions(rsc->priv->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 pcmk_action_t *action, enum pcmk__action_flags flag)
{
const char *change = NULL;
switch (flag) {
case pcmk__action_runnable:
change = "unrunnable";
break;
case pcmk__action_migratable:
change = "unmigrateable";
break;
case pcmk__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(pcmk_action_t *action, const char *reason,
bool overwrite)
{
if (action->reason != NULL && overwrite) {
pcmk__rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'",
action->uuid, action->reason,
pcmk__s(reason, "(none)"));
} else if (action->reason == NULL) {
pcmk__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
*/
void
pe__clear_resource_history(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__assert((rsc != NULL) && (node != NULL));
custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
PCMK_ACTION_LRM_DELETE, node, FALSE, rsc->priv->scheduler);
}
#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)
{
int a_call_id = -1;
int b_call_id = -1;
char *a_uuid = NULL;
char *b_uuid = NULL;
const char *a_xml_id = pcmk__xe_get(xml_a, PCMK_XA_ID);
const char *b_xml_id = pcmk__xe_get(xml_b, PCMK_XA_ID);
const char *a_node = pcmk__xe_get(xml_a, PCMK__META_ON_NODE);
const char *b_node = pcmk__xe_get(xml_b, PCMK__META_ON_NODE);
bool 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 PCMK__XE_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.
*/
pcmk__config_err("Duplicate " PCMK__XE_LRM_RSC_OP " entries named %s",
a_xml_id);
sort_return(0, "duplicate");
}
pcmk__xe_get_int(xml_a, PCMK__XA_CALL_ID, &a_call_id);
pcmk__xe_get_int(xml_b, PCMK__XA_CALL_ID, &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
* PCMK_XA_LAST_RC_CHANGE.
*/
time_t last_a = -1;
time_t last_b = -1;
pcmk__xe_get_time(xml_a, PCMK_XA_LAST_RC_CHANGE, &last_a);
pcmk__xe_get_time(xml_b, PCMK_XA_LAST_RC_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 PCMK__XA_TRANSITION_MAGIC to determine its age relative
* to the other.
*/
int a_id = -1;
int b_id = -1;
const char *a_magic = pcmk__xe_get(xml_a, PCMK__XA_TRANSITION_MAGIC);
const char *b_magic = pcmk__xe_get(xml_b, PCMK__XA_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)
{
return pe__is_newer_op((const xmlNode *) a, (const xmlNode *) b);
}
/*!
* \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
*/
pcmk_action_t *
pe__new_rsc_pseudo_action(pcmk_resource_t *rsc, const char *task, bool optional,
bool runnable)
{
pcmk_action_t *action = NULL;
pcmk__assert((rsc != NULL) && (task != NULL));
action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL,
optional, rsc->priv->scheduler);
pcmk__set_action_flags(action, pcmk__action_pseudo);
if (runnable) {
pcmk__set_action_flags(action, pcmk__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 pcmk__insert_meta().
*/
void
pe__add_action_expected_result(pcmk_action_t *action, int expected_result)
{
pcmk__assert((action != NULL) && (action->meta != NULL));
g_hash_table_insert(action->meta, pcmk__str_copy(PCMK__META_OP_TARGET_RC),
pcmk__itoa(expected_result));
}
diff --git a/lib/pengine/pe_digest.c b/lib/pengine/pe_digest.c
index 5a369fd051..8cc0c8633a 100644
--- a/lib/pengine/pe_digest.c
+++ b/lib/pengine/pe_digest.c
@@ -1,613 +1,613 @@
/*
* Copyright 2004-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
/*!
* \internal
* \brief Free an operation digest cache entry
*
* \param[in,out] ptr Pointer to cache entry to free
*
* \note The argument is a gpointer so this can be used as a hash table
* free function.
*/
void
pe__free_digests(gpointer ptr)
{
pcmk__op_digest_t *data = ptr;
if (data != NULL) {
pcmk__xml_free(data->params_all);
pcmk__xml_free(data->params_secure);
pcmk__xml_free(data->params_restart);
free(data->digest_all_calc);
free(data->digest_restart_calc);
free(data->digest_secure_calc);
free(data);
}
}
// Return true if XML attribute name is not substring of a given string
static bool
attr_not_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) == NULL) {
crm_trace("Filtering %s (not found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
// Return true if XML attribute name is substring of a given string
static bool
attr_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) != NULL) {
crm_trace("Filtering %s (found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
/*!
* \internal
* \brief Add digest of all parameters to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in,out] rsc Resource that action was for
* \param[in] node Node action was performed on
* \param[in] params Resource parameters evaluated for node
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] xml_op Unused
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value table to override resource parameters
* \param[in,out] scheduler Scheduler data
*/
static void
calculate_main_digest(pcmk__op_digest_t *data, pcmk_resource_t *rsc,
const pcmk_node_t *node, GHashTable *params,
const char *task, guint *interval_ms,
const xmlNode *xml_op, const char *op_version,
GHashTable *overrides, pcmk_scheduler_t *scheduler)
{
xmlNode *action_config = NULL;
data->params_all = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
/* REMOTE_CONTAINER_HACK: Allow Pacemaker Remote nodes to run containers
* that themselves are Pacemaker Remote nodes
*/
(void) pe__add_bundle_remote_name(rsc, data->params_all,
PCMK_REMOTE_RA_ADDR);
if (overrides != NULL) {
// If interval was overridden, reset it
const char *meta_name = CRM_META "_" PCMK_META_INTERVAL;
const char *interval_s = g_hash_table_lookup(overrides, meta_name);
if (interval_s != NULL) {
long long value_ll;
if ((pcmk__scan_ll(interval_s, &value_ll, 0LL) == pcmk_rc_ok)
&& (value_ll >= 0) && (value_ll <= G_MAXUINT)) {
*interval_ms = (guint) value_ll;
}
}
// Add overrides to list of all parameters
g_hash_table_foreach(overrides, hash2field, data->params_all);
}
// Add provided instance parameters
g_hash_table_foreach(params, hash2field, data->params_all);
// Find action configuration XML in CIB
action_config = pcmk__find_action_config(rsc, task, *interval_ms, true);
/* Add action-specific resource instance attributes to the digest list.
*
* If this is a one-time action with action-specific instance attributes,
* enforce a restart instead of reload-agent in case the main digest doesn't
* match, even if the restart digest does. This ensures any changes of the
* action-specific parameters get applied for this specific action, and
* digests calculated for the resulting history will be correct. Default the
* result to RSC_DIGEST_RESTART for the case where the main digest doesn't
* match.
*/
params = pcmk__unpack_action_rsc_params(action_config, node->priv->attrs,
scheduler);
if ((*interval_ms == 0) && (g_hash_table_size(params) > 0)) {
data->rc = pcmk__digest_restart;
}
g_hash_table_foreach(params, hash2field, data->params_all);
g_hash_table_destroy(params);
// Add action meta-attributes
params = pcmk__unpack_action_meta(rsc, node, task, *interval_ms,
action_config);
g_hash_table_foreach(params, hash2metafield, data->params_all);
g_hash_table_destroy(params);
pcmk__filter_op_for_digest(data->params_all);
data->digest_all_calc = pcmk__digest_op_params(data->params_all);
}
// Return true if XML attribute name is a Pacemaker-defined fencing parameter
static bool
is_fence_param(xmlAttrPtr attr, void *user_data)
{
return pcmk_stonith_param((const char *) attr->name);
}
/*!
* \internal
* \brief Add secure digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] rsc Resource that action was for
* \param[in] params Resource parameters evaluated for node
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value hash table to override resource parameters
*/
static void
calculate_secure_digest(pcmk__op_digest_t *data, const pcmk_resource_t *rsc,
GHashTable *params, const xmlNode *xml_op,
const char *op_version, GHashTable *overrides)
{
/* @COMPAT CRM_FEATURE_SET was bumped to 3.16.0 in Pacemaker 2.1.5. When we
* no longer support rolling upgrades from 2.1.4 and below, we can drop the
* old_version code.
*/
const char *class = pcmk__xe_get(rsc->priv->xml, PCMK_XA_CLASS);
const char *secure_list = NULL;
bool old_version = (compare_version(op_version, "3.16.0") < 0);
if (xml_op == NULL) {
secure_list = " passwd password user ";
} else {
secure_list = pcmk__xe_get(xml_op, PCMK__XA_OP_SECURE_PARAMS);
}
if (old_version) {
data->params_secure = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
if (overrides != NULL) {
g_hash_table_foreach(overrides, hash2field, data->params_secure);
}
g_hash_table_foreach(params, hash2field, data->params_secure);
} else {
// Start with a copy of all parameters
data->params_secure = pcmk__xml_copy(NULL, data->params_all);
}
if (secure_list != NULL) {
pcmk__xe_remove_matching_attrs(data->params_secure, false,
attr_in_string, (void *) secure_list);
}
if (old_version
&& pcmk_is_set(pcmk_get_ra_caps(class),
pcmk_ra_cap_fence_params)) {
- /* For stonith resources, Pacemaker adds special parameters,
- * but these are not listed in fence agent meta-data, so with older
- * versions of DC, the controller will not hash them. That means we have
- * to filter them out before calculating our hash for comparison.
+ /* For fencing resources, Pacemaker adds special parameters, but these
+ * are not listed in fence agent meta-data, so with older versions of
+ * DC, the controller will not hash them. That means we have to filter
+ * them out before calculating our hash for comparison.
*/
pcmk__xe_remove_matching_attrs(data->params_secure, false,
is_fence_param, NULL);
}
pcmk__filter_op_for_digest(data->params_secure);
/* CRM_meta_timeout *should* be part of a digest for recurring operations.
* However, with older versions of DC, the controller does not add timeout
* to secure digests, because it only includes parameters declared by the
* resource agent.
* Remove any timeout that made it this far, to match.
*/
if (old_version) {
pcmk__xe_remove_attr(data->params_secure,
CRM_META "_" PCMK_META_TIMEOUT);
}
data->digest_secure_calc = pcmk__digest_op_params(data->params_secure);
}
/*!
* \internal
* \brief Add restart digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
*
* \note This function doesn't need to handle overrides because it starts with
* data->params_all, which already has overrides applied.
*/
static void
calculate_restart_digest(pcmk__op_digest_t *data, const xmlNode *xml_op,
const char *op_version)
{
const char *value = NULL;
// We must have XML of resource operation history
if (xml_op == NULL) {
return;
}
// And the history must have a restart digest to compare against
if (pcmk__xe_get(xml_op, PCMK__XA_OP_RESTART_DIGEST) == NULL) {
return;
}
// Start with a copy of all parameters
data->params_restart = pcmk__xml_copy(NULL, data->params_all);
// Then filter out reloadable parameters, if any
value = pcmk__xe_get(xml_op, PCMK__XA_OP_FORCE_RESTART);
if (value != NULL) {
pcmk__xe_remove_matching_attrs(data->params_restart, false,
attr_not_in_string, (void *) value);
}
data->digest_restart_calc = pcmk__digest_op_params(data->params_restart);
}
/*!
* \internal
* \brief Create a new digest cache entry with calculated digests
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] overrides Key/value table to override resource parameters
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to new digest cache entry (or NULL on memory error)
* \note It is the caller's responsibility to free the result using
* pe__free_digests().
*/
pcmk__op_digest_t *
pe__calculate_digests(pcmk_resource_t *rsc, const char *task,
guint *interval_ms, const pcmk_node_t *node,
const xmlNode *xml_op, GHashTable *overrides,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
const char *op_version = NULL;
GHashTable *params = NULL;
CRM_CHECK(scheduler != NULL, return NULL);
data = calloc(1, sizeof(pcmk__op_digest_t));
if (data == NULL) {
pcmk__sched_err(scheduler,
"Could not allocate memory for operation digest");
return NULL;
}
data->rc = pcmk__digest_match;
if (xml_op != NULL) {
op_version = pcmk__xe_get(xml_op, PCMK_XA_CRM_FEATURE_SET);
}
if ((op_version == NULL) && (scheduler->input != NULL)) {
op_version = pcmk__xe_get(scheduler->input, PCMK_XA_CRM_FEATURE_SET);
}
if (op_version == NULL) {
op_version = CRM_FEATURE_SET;
}
params = pe_rsc_params(rsc, node, scheduler);
calculate_main_digest(data, rsc, node, params, task, interval_ms, xml_op,
op_version, overrides, scheduler);
if (calc_secure) {
calculate_secure_digest(data, rsc, params, xml_op, op_version,
overrides);
}
calculate_restart_digest(data, xml_op, op_version);
return data;
}
/*!
* \internal
* \brief Calculate action digests and store in node's digest cache
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in] interval_ms Action's interval
* \param[in,out] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry
*/
static pcmk__op_digest_t *
rsc_action_digest(pcmk_resource_t *rsc, const char *task, guint interval_ms,
pcmk_node_t *node, const xmlNode *xml_op,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
char *key = pcmk__op_key(rsc->id, task, interval_ms);
data = g_hash_table_lookup(node->priv->digest_cache, key);
if (data == NULL) {
data = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op,
NULL, calc_secure, scheduler);
pcmk__assert(data != NULL);
g_hash_table_insert(node->priv->digest_cache, strdup(key), data);
}
free(key);
return data;
}
/*!
* \internal
* \brief Calculate operation digests and compare against an XML history entry
*
* \param[in,out] rsc Resource to check
* \param[in] xml_op Resource history XML
* \param[in,out] node Node to use for digest calculation
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry, with comparison result set
*/
pcmk__op_digest_t *
rsc_action_digest_cmp(pcmk_resource_t *rsc, const xmlNode *xml_op,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
guint interval_ms = 0;
const char *op_version;
const char *task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION);
const char *digest_all;
const char *digest_restart;
pcmk__assert(node != NULL);
op_version = pcmk__xe_get(xml_op, PCMK_XA_CRM_FEATURE_SET);
digest_all = pcmk__xe_get(xml_op, PCMK__XA_OP_DIGEST);
digest_restart = pcmk__xe_get(xml_op, PCMK__XA_OP_RESTART_DIGEST);
pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &interval_ms);
data = rsc_action_digest(rsc, task, interval_ms, node, xml_op,
pcmk_is_set(scheduler->flags,
pcmk__sched_sanitized),
scheduler);
if (!pcmk__str_eq(data->digest_restart_calc, digest_restart,
pcmk__str_null_matches)) {
pcmk__rsc_info(rsc,
"Parameter change for %s-interval %s of %s on %s "
"requires restart (hash now %s vs. %s "
"with op feature set %s for transition %s)",
pcmk__readable_interval(interval_ms), task, rsc->id,
pcmk__node_name(node), data->digest_restart_calc,
pcmk__s(digest_restart, "missing"), op_version,
pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_restart;
} else if (digest_all == NULL) {
/* it is unknown what the previous op digest was */
data->rc = pcmk__digest_unknown;
} else if (strcmp(digest_all, data->digest_all_calc) != 0) {
/* Given a non-recurring operation with extra parameters configured,
* in case that the main digest doesn't match, even if the restart
* digest matches, enforce a restart rather than a reload-agent anyway.
* So that it ensures any changes of the extra parameters get applied
* for this specific operation, and the digests calculated for the
* resulting PCMK__XE_LRM_RSC_OP will be correct.
* Preserve the implied rc pcmk__digest_restart for the case that the
* main digest doesn't match.
*/
if ((interval_ms == 0) && (data->rc == pcmk__digest_restart)) {
pcmk__rsc_info(rsc,
"Parameters containing extra ones to %ums-interval"
" %s action for %s on %s "
"changed: hash was %s vs. now %s (restart:%s) %s",
interval_ms, task, rsc->id, pcmk__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc, op_version,
pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC));
} else {
pcmk__rsc_info(rsc,
"Parameters to %ums-interval %s action for %s on %s "
"changed: hash was %s vs. now %s (%s:%s) %s",
interval_ms, task, rsc->id, pcmk__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc,
(interval_ms > 0)? "reschedule" : "reload",
op_version,
pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_mismatch;
}
} else {
data->rc = pcmk__digest_match;
}
return data;
}
/*!
* \internal
* \brief Create an unfencing summary for use in special node attribute
*
* Create a string combining a fence device's resource ID, agent type, and
* parameter digest (whether for all parameters or just non-private parameters).
* This can be stored in a special node attribute, allowing us to detect changes
* in either the agent type or parameters, to know whether unfencing must be
* redone or can be safely skipped when the device's history is cleaned.
*
* \param[in] rsc_id Fence device resource ID
* \param[in] agent_type Fence device agent
* \param[in] param_digest Fence device parameter digest
*
* \return Newly allocated string with unfencing digest
* \note The caller is responsible for freeing the result.
*/
static inline char *
create_unfencing_summary(const char *rsc_id, const char *agent_type,
const char *param_digest)
{
return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest);
}
/*!
* \internal
* \brief Check whether a node can skip unfencing
*
* Check whether a fence device's current definition matches a node's
* stored summary of when it was last unfenced by the device.
*
* \param[in] rsc_id Fence device's resource ID
* \param[in] agent Fence device's agent type
* \param[in] digest_calc Fence device's current parameter digest
* \param[in] node_summary Value of node's special unfencing node attribute
* (a comma-separated list of unfencing summaries for
* all devices that have unfenced this node)
*
* \return TRUE if digest matches, FALSE otherwise
*/
static bool
unfencing_digest_matches(const char *rsc_id, const char *agent,
const char *digest_calc, const char *node_summary)
{
bool matches = FALSE;
if (rsc_id && agent && digest_calc && node_summary) {
char *search_secure = create_unfencing_summary(rsc_id, agent,
digest_calc);
/* The digest was calculated including the device ID and agent,
* so there is no risk of collision using strstr().
*/
matches = (strstr(node_summary, search_secure) != NULL);
crm_trace("Calculated unfencing digest '%s' %sfound in '%s'",
search_secure, matches? "" : "not ", node_summary);
free(search_secure);
}
return matches;
}
/* Magic string to use as action name for digest cache entries used for
* unfencing checks. This is not a real action name (i.e. "on"), so
* pcmk__check_action_config() won't confuse these entries with real actions.
*/
#define STONITH_DIGEST_TASK "stonith-on"
/*!
* \internal
* \brief Calculate fence device digests and digest comparison result
*
* \param[in,out] rsc Fence device resource
* \param[in] agent Fence device's agent type
* \param[in,out] node Node with digest cache to use
* \param[in,out] scheduler Scheduler data
*
* \return Node's digest cache entry
*/
pcmk__op_digest_t *
pe__compare_fencing_digest(pcmk_resource_t *rsc, const char *agent,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
const char *node_summary = NULL;
// Calculate device's current parameter digests
pcmk__op_digest_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, 0U,
node, NULL, TRUE, scheduler);
// Check whether node has special unfencing summary node attribute
node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_ALL, NULL,
pcmk__rsc_node_current);
if (node_summary == NULL) {
data->rc = pcmk__digest_unknown;
return data;
}
// Check whether full parameter digest matches
if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc,
node_summary)) {
data->rc = pcmk__digest_match;
return data;
}
// Check whether secure parameter digest matches
node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_SECURE, NULL,
pcmk__rsc_node_current);
if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc,
node_summary)) {
data->rc = pcmk__digest_match;
if (!pcmk__is_daemon && (scheduler->priv->out != NULL)) {
pcmk__output_t *out = scheduler->priv->out;
out->info(out, "Only 'private' parameters to %s "
"for unfencing %s changed", rsc->id,
pcmk__node_name(node));
}
return data;
}
// Parameters don't match
data->rc = pcmk__digest_mismatch;
if (pcmk_is_set(scheduler->flags, pcmk__sched_sanitized)
&& (data->digest_secure_calc != NULL)) {
if (scheduler->priv->out != NULL) {
pcmk__output_t *out = scheduler->priv->out;
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
out->info(out, "Parameters to %s for unfencing "
"%s changed, try '%s'", rsc->id,
pcmk__node_name(node), digest);
free(digest);
} else if (!pcmk__is_daemon) {
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
printf("Parameters to %s for unfencing %s changed, try '%s'\n",
rsc->id, pcmk__node_name(node), digest);
free(digest);
}
}
return data;
}
diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c
index 89d678a712..241dcd47e1 100644
--- a/lib/pengine/pe_notif.c
+++ b/lib/pengine/pe_notif.c
@@ -1,1027 +1,1026 @@
/*
- * Copyright 2004-2024 the Pacemaker project contributors
+ * Copyright 2004-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/common/xml.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include "pe_status_private.h"
typedef struct notify_entry_s {
const pcmk_resource_t *rsc;
const pcmk_node_t *node;
} notify_entry_t;
/*!
* \internal
* \brief Compare two notification entries
*
* Compare two notification entries, where the one with the alphabetically first
* resource name (or if equal, node ID) sorts as first, with NULL sorting as
* less than non-NULL.
*
* \param[in] a First notification entry to compare
* \param[in] b Second notification entry to compare
*
* \return -1 if \p a sorts before \p b, 0 if they are equal, otherwise 1
*/
static gint
compare_notify_entries(gconstpointer a, gconstpointer b)
{
int tmp;
const notify_entry_t *entry_a = a;
const notify_entry_t *entry_b = b;
// NULL a or b is not actually possible
if ((entry_a == NULL) && (entry_b == NULL)) {
return 0;
}
if (entry_a == NULL) {
return 1;
}
if (entry_b == NULL) {
return -1;
}
// NULL resources sort first
if ((entry_a->rsc == NULL) && (entry_b->rsc == NULL)) {
return 0;
}
if (entry_a->rsc == NULL) {
return 1;
}
if (entry_b->rsc == NULL) {
return -1;
}
// Compare resource names
tmp = strcmp(entry_a->rsc->id, entry_b->rsc->id);
if (tmp != 0) {
return tmp;
}
// Otherwise NULL nodes sort first
if ((entry_a->node == NULL) && (entry_b->node == NULL)) {
return 0;
}
if (entry_a->node == NULL) {
return 1;
}
if (entry_b->node == NULL) {
return -1;
}
// Finally, compare node IDs
return strcmp(entry_a->node->priv->id, entry_b->node->priv->id);
}
/*!
* \internal
* \brief Duplicate a notification entry
*
* \param[in] entry Entry to duplicate
*
* \return Newly allocated duplicate of \p entry
* \note It is the caller's responsibility to free the return value.
*/
static notify_entry_t *
dup_notify_entry(const notify_entry_t *entry)
{
notify_entry_t *dup = pcmk__assert_alloc(1, sizeof(notify_entry_t));
dup->rsc = entry->rsc;
dup->node = entry->node;
return dup;
}
/*!
* \internal
* \brief Given a list of nodes, create strings with node names
*
* \param[in] list List of nodes (as pcmk_node_t *)
* \param[out] all_node_names If not NULL, will be set to space-separated list
* of the names of all nodes in \p list
* \param[out] host_node_names Same as \p all_node_names, except active
* guest nodes will list the name of their host
*
* \note The caller is responsible for freeing the output argument values using
* \p g_string_free().
*/
static void
get_node_names(const GList *list, GString **all_node_names,
GString **host_node_names)
{
if (all_node_names != NULL) {
*all_node_names = NULL;
}
if (host_node_names != NULL) {
*host_node_names = NULL;
}
for (const GList *iter = list; iter != NULL; iter = iter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) iter->data;
if (node->priv->name == NULL) {
/* @TODO This breaks the comparability of the various notification
* variables and thus any agent relying on it. Maybe add "UNKNOWN"
* or something like that.
*/
continue;
}
// Always add to list of all node names
if (all_node_names != NULL) {
pcmk__add_word(all_node_names, 1024, node->priv->name);
}
// Add to host node name list if appropriate
if (host_node_names != NULL) {
if (pcmk__is_guest_or_bundle_node(node)) {
const pcmk_resource_t *launcher = NULL;
launcher = node->priv->remote->priv->launcher;
if (launcher->priv->active_nodes != NULL) {
node = pcmk__current_node(launcher);
if (node->priv->name == NULL) {
continue;
}
}
}
pcmk__add_word(host_node_names, 1024, node->priv->name);
}
}
if ((all_node_names != NULL) && (*all_node_names == NULL)) {
*all_node_names = g_string_new(" ");
}
if ((host_node_names != NULL) && (*host_node_names == NULL)) {
*host_node_names = g_string_new(" ");
}
}
/*!
* \internal
* \brief Create strings of instance and node names from notification entries
*
* \param[in,out] list List of notification entries (will be sorted here)
* \param[out] rsc_names If not NULL, will be set to space-separated list
* of clone instances from \p list
* \param[out] node_names If not NULL, will be set to space-separated list
* of node names from \p list
*
* \return (Possibly new) head of sorted \p list
* \note The caller is responsible for freeing the output argument values using
* \p g_list_free_full() and \p g_string_free().
*/
static GList *
notify_entries_to_strings(GList *list, GString **rsc_names,
GString **node_names)
{
const char *last_rsc_id = NULL;
// Initialize output lists to NULL
if (rsc_names != NULL) {
*rsc_names = NULL;
}
if (node_names != NULL) {
*node_names = NULL;
}
// Sort input list for user-friendliness (and ease of filtering duplicates)
list = g_list_sort(list, compare_notify_entries);
for (GList *gIter = list; gIter != NULL; gIter = gIter->next) {
notify_entry_t *entry = (notify_entry_t *) gIter->data;
// Entry must have a resource (with ID)
CRM_LOG_ASSERT((entry != NULL) && (entry->rsc != NULL)
&& (entry->rsc->id != NULL));
if ((entry == NULL) || (entry->rsc == NULL)
|| (entry->rsc->id == NULL)) {
continue;
}
// Entry must have a node unless listing inactive resources
CRM_LOG_ASSERT((node_names == NULL) || (entry->node != NULL));
if ((node_names != NULL) && (entry->node == NULL)) {
continue;
}
// Don't add duplicates of a particular clone instance
if (pcmk__str_eq(entry->rsc->id, last_rsc_id, pcmk__str_none)) {
continue;
}
last_rsc_id = entry->rsc->id;
if (rsc_names != NULL) {
pcmk__add_word(rsc_names, 1024, entry->rsc->id);
}
if ((node_names != NULL) && (entry->node->priv->name != NULL)) {
pcmk__add_word(node_names, 1024, entry->node->priv->name);
}
}
// If there are no entries, return "empty" lists
if ((rsc_names != NULL) && (*rsc_names == NULL)) {
*rsc_names = g_string_new(" ");
}
if ((node_names != NULL) && (*node_names == NULL)) {
*node_names = g_string_new(" ");
}
return list;
}
/*!
* \internal
* \brief Copy a meta-attribute into a notify action
*
* \param[in] key Name of meta-attribute to copy
* \param[in] value Value of meta-attribute to copy
* \param[in,out] user_data Notify action to copy into
*/
static void
copy_meta_to_notify(gpointer key, gpointer value, gpointer user_data)
{
pcmk_action_t *notify = (pcmk_action_t *) user_data;
/* Any existing meta-attributes (for example, the action timeout) are for
* the notify action itself, so don't override those.
*/
if (g_hash_table_lookup(notify->meta, (const char *) key) != NULL) {
return;
}
pcmk__insert_dup(notify->meta, (const char *) key, (const char *) value);
}
static void
add_notify_data_to_action_meta(const notify_data_t *n_data,
pcmk_action_t *action)
{
for (const GSList *item = n_data->keys; item; item = item->next) {
const pcmk_nvpair_t *nvpair = (const pcmk_nvpair_t *) item->data;
pcmk__insert_meta(action, nvpair->name, nvpair->value);
}
}
/*!
* \internal
* \brief Create a new notify pseudo-action for a clone resource
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in] action Action to use in notify action key
* \param[in] notif_action PCMK_ACTION_NOTIFY or PCMK_ACTION_NOTIFIED
* \param[in] notif_type "pre", "post", "confirmed-pre", "confirmed-post"
*
* \return Newly created notify pseudo-action
*/
static pcmk_action_t *
new_notify_pseudo_action(pcmk_resource_t *rsc, const pcmk_action_t *action,
const char *notif_action, const char *notif_type)
{
pcmk_action_t *notify = NULL;
notify = custom_action(rsc,
pcmk__notify_key(rsc->id, notif_type, action->task),
notif_action, NULL,
pcmk_is_set(action->flags, pcmk__action_optional),
rsc->priv->scheduler);
pcmk__set_action_flags(notify, pcmk__action_pseudo);
pcmk__insert_meta(notify, "notify_key_type", notif_type);
pcmk__insert_meta(notify, "notify_key_operation", action->task);
return notify;
}
/*!
* \internal
* \brief Create a new notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] op Action that notification is for
* \param[in,out] notify_done Parent pseudo-action for notifications complete
* \param[in] n_data Notification values to add to action meta-data
*
* \return Newly created notify action
*/
static pcmk_action_t *
new_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
pcmk_action_t *op, pcmk_action_t *notify_done,
const notify_data_t *n_data)
{
char *key = NULL;
pcmk_action_t *notify_action = NULL;
const char *value = NULL;
const char *task = NULL;
const char *skip_reason = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL), return NULL);
// Ensure we have all the info we need
if (op == NULL) {
skip_reason = "no action";
} else if (notify_done == NULL) {
skip_reason = "no parent notification";
} else if (!node->details->online) {
skip_reason = "node offline";
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
skip_reason = "original action not runnable";
}
if (skip_reason != NULL) {
pcmk__rsc_trace(rsc, "Skipping notify action for %s on %s: %s",
rsc->id, pcmk__node_name(node), skip_reason);
return NULL;
}
value = g_hash_table_lookup(op->meta, "notify_type"); // "pre" or "post"
task = g_hash_table_lookup(op->meta, "notify_operation"); // original action
pcmk__rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)",
rsc->id, pcmk__node_name(node), value, task);
// Create the notify action
key = pcmk__notify_key(rsc->id, value, task);
notify_action = custom_action(rsc, key, op->task, node,
pcmk_is_set(op->flags, pcmk__action_optional),
rsc->priv->scheduler);
// Add meta-data to notify action
g_hash_table_foreach(op->meta, copy_meta_to_notify, notify_action);
add_notify_data_to_action_meta(n_data, notify_action);
// Order notify after original action and before parent notification
order_actions(op, notify_action, pcmk__ar_ordered);
order_actions(notify_action, notify_done, pcmk__ar_ordered);
return notify_action;
}
/*!
* \internal
* \brief Create a new "post-" notify action for a clone instance
*
* \param[in,out] rsc Clone instance that notification is for
* \param[in] node Node that notification is for
* \param[in,out] n_data Notification values to add to action meta-data
*/
static void
new_post_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node,
notify_data_t *n_data)
{
pcmk_action_t *notify = NULL;
pcmk__assert(n_data != NULL);
// Create the "post-" notify action for specified instance
notify = new_notify_action(rsc, node, n_data->post, n_data->post_done,
n_data);
if (notify != NULL) {
notify->priority = PCMK_SCORE_INFINITY;
}
// Order recurring monitors after all "post-" notifications complete
if (n_data->post_done == NULL) {
return;
}
for (GList *iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *mon = (pcmk_action_t *) iter->data;
const char *interval_ms_s = NULL;
interval_ms_s = g_hash_table_lookup(mon->meta, PCMK_META_INTERVAL);
if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)
|| pcmk__str_eq(mon->task, PCMK_ACTION_CANCEL, pcmk__str_none)) {
continue; // Not a recurring monitor
}
order_actions(n_data->post_done, mon, pcmk__ar_ordered);
}
}
/*!
* \internal
* \brief Create and order notification pseudo-actions for a clone action
*
* In addition to the actual notify actions needed for each clone instance,
* clone notifications also require pseudo-actions to provide ordering points
* in the notification process. This creates the notification data, along with
* appropriate pseudo-actions and their orderings.
*
* For example, the ordering sequence for starting a clone is:
*
* "pre-" notify pseudo-action for clone
* -> "pre-" notify actions for each clone instance
* -> "pre-" notifications complete pseudo-action for clone
* -> start actions for each clone instance
* -> "started" pseudo-action for clone
* -> "post-" notify pseudo-action for clone
* -> "post-" notify actions for each clone instance
* -> "post-" notifications complete pseudo-action for clone
*
* \param[in,out] rsc Clone that notifications are for
* \param[in] task Name of action that notifications are for
* \param[in,out] action If not NULL, create a "pre-" pseudo-action ordered
* before a "pre-" complete pseudo-action, ordered
* before this action
* \param[in,out] complete If not NULL, create a "post-" pseudo-action ordered
* after this action, and a "post-" complete
* pseudo-action ordered after that
*
* \return Newly created notification data
*/
notify_data_t *
pe__action_notif_pseudo_ops(pcmk_resource_t *rsc, const char *task,
pcmk_action_t *action, pcmk_action_t *complete)
{
notify_data_t *n_data = NULL;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_notify)) {
return NULL;
}
n_data = pcmk__assert_alloc(1, sizeof(notify_data_t));
n_data->action = task;
if (action != NULL) { // Need "pre-" pseudo-actions
// Create "pre-" notify pseudo-action for clone
n_data->pre = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFY,
"pre");
pcmk__set_action_flags(n_data->pre, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre, "notify_type", "pre");
pcmk__insert_meta(n_data->pre, "notify_operation", n_data->action);
// Create "pre-" notifications complete pseudo-action for clone
n_data->pre_done = new_notify_pseudo_action(rsc, action,
PCMK_ACTION_NOTIFIED,
"confirmed-pre");
pcmk__set_action_flags(n_data->pre_done, pcmk__action_runnable);
pcmk__insert_meta(n_data->pre_done, "notify_type", "pre");
pcmk__insert_meta(n_data->pre_done, "notify_operation", n_data->action);
// Order "pre-" -> "pre-" complete -> original action
order_actions(n_data->pre, n_data->pre_done, pcmk__ar_ordered);
order_actions(n_data->pre_done, action, pcmk__ar_ordered);
}
if (complete != NULL) { // Need "post-" pseudo-actions
// Create "post-" notify pseudo-action for clone
n_data->post = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFY, "post");
n_data->post->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post, "notify_type", "post");
pcmk__insert_meta(n_data->post, "notify_operation", n_data->action);
// Create "post-" notifications complete pseudo-action for clone
n_data->post_done = new_notify_pseudo_action(rsc, complete,
PCMK_ACTION_NOTIFIED,
"confirmed-post");
n_data->post_done->priority = PCMK_SCORE_INFINITY;
if (pcmk_is_set(complete->flags, pcmk__action_runnable)) {
pcmk__set_action_flags(n_data->post_done, pcmk__action_runnable);
} else {
pcmk__clear_action_flags(n_data->post_done, pcmk__action_runnable);
}
pcmk__insert_meta(n_data->post_done, "notify_type", "post");
pcmk__insert_meta(n_data->post_done,
"notify_operation", n_data->action);
/* Order original action complete -> "post-" -> "post-" complete
*
* @TODO Should we add |pcmk__ar_unrunnable_first_blocks to these?
* Otherwise we might get an invalid transition due to unresolved
* dependencies when "complete" is a fencing op (which can happen at
* least for bundles) but that op is unrunnable (due to lack of quorum,
* for example).
*/
order_actions(complete, n_data->post, pcmk__ar_first_implies_then);
order_actions(n_data->post, n_data->post_done,
pcmk__ar_first_implies_then);
}
// If we created both, order "pre-" complete -> "post-"
if ((action != NULL) && (complete != NULL)) {
order_actions(n_data->pre_done, n_data->post, pcmk__ar_ordered);
}
return n_data;
}
/*!
* \internal
* \brief Create a new notification entry
*
* \param[in] rsc Resource for notification
* \param[in] node Node for notification
*
* \return Newly allocated notification entry
* \note The caller is responsible for freeing the return value.
*/
static notify_entry_t *
new_notify_entry(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
notify_entry_t *entry = pcmk__assert_alloc(1, sizeof(notify_entry_t));
entry->rsc = rsc;
entry->node = node;
return entry;
}
/*!
* \internal
* \brief Add notification data for resource state and optionally actions
*
* \param[in,out] rsc Clone or clone instance being notified
* \param[in] activity Whether to add notification entries for actions
* \param[in,out] n_data Notification data for clone
*/
static void
collect_resource_data(pcmk_resource_t *rsc, bool activity,
notify_data_t *n_data)
{
const GList *iter = NULL;
notify_entry_t *entry = NULL;
const pcmk_node_t *node = NULL;
if (n_data == NULL) {
return;
}
if (n_data->allowed_nodes == NULL) {
n_data->allowed_nodes = rsc->priv->allowed_nodes;
}
// If this is a clone, call recursively for each instance
if (rsc->priv->children != NULL) {
for (iter = rsc->priv->children; iter != NULL; iter = iter->next) {
pcmk_resource_t *child = iter->data;
collect_resource_data(child, activity, n_data);
}
return;
}
// This is a notification for a single clone instance
if (rsc->priv->active_nodes != NULL) {
node = rsc->priv->active_nodes->data; // First is sufficient
}
entry = new_notify_entry(rsc, node);
// Add notification indicating the resource state
switch (rsc->priv->orig_role) {
case pcmk_role_stopped:
n_data->inactive = g_list_prepend(n_data->inactive, entry);
break;
case pcmk_role_started:
n_data->active = g_list_prepend(n_data->active, entry);
break;
case pcmk_role_unpromoted:
n_data->unpromoted = g_list_prepend(n_data->unpromoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
case pcmk_role_promoted:
n_data->promoted = g_list_prepend(n_data->promoted, entry);
n_data->active = g_list_prepend(n_data->active,
dup_notify_entry(entry));
break;
default:
pcmk__sched_err(rsc->priv->scheduler,
"Resource %s role on %s (%s) is not supported for "
"notifications (bug?)",
rsc->id, pcmk__node_name(node),
pcmk_role_text(rsc->priv->orig_role));
free(entry);
break;
}
if (!activity) {
return;
}
// Add notification entries for each of the resource's actions
for (iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
const pcmk_action_t *op = (const pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
enum pcmk__action_type task = pcmk__parse_action(op->task);
if ((task == pcmk__action_stop) && op->node->details->unclean) {
// Create anyway (additional noise if node can't be fenced)
} else if (!pcmk_is_set(op->flags, pcmk__action_runnable)) {
continue;
}
entry = new_notify_entry(rsc, op->node);
switch (task) {
case pcmk__action_start:
n_data->start = g_list_prepend(n_data->start, entry);
break;
case pcmk__action_stop:
n_data->stop = g_list_prepend(n_data->stop, entry);
break;
case pcmk__action_promote:
n_data->promote = g_list_prepend(n_data->promote, entry);
break;
case pcmk__action_demote:
n_data->demote = g_list_prepend(n_data->demote, entry);
break;
default:
free(entry);
break;
}
}
}
}
// For (char *) value
#define add_notify_env(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, value); \
} while (0)
// For (GString *) value
#define add_notify_env_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
} while (0)
// For (GString *) value
#define add_notify_env_free_gs(n_data, key, value) do { \
n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \
(const char *) value->str); \
g_string_free(value, TRUE); value = NULL; \
} while (0)
/*!
* \internal
* \brief Create notification name/value pairs from structured data
*
* \param[in] rsc Resource that notification is for
* \param[in,out] n_data Notification data
*/
static void
add_notif_keys(const pcmk_resource_t *rsc, notify_data_t *n_data)
{
bool required = false; // Whether to make notify actions required
GString *rsc_list = NULL;
GString *node_list = NULL;
GString *metal_list = NULL;
const char *source = NULL;
GList *nodes = NULL;
n_data->stop = notify_entries_to_strings(n_data->stop,
&rsc_list, &node_list);
if ((strcmp(" ", (const char *) rsc_list->str) != 0)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_STOP, pcmk__str_none)) {
required = true;
}
add_notify_env_free_gs(n_data, "notify_stop_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_stop_uname", node_list);
if ((n_data->start != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_START, pcmk__str_none)) {
required = true;
}
n_data->start = notify_entries_to_strings(n_data->start,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_start_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_start_uname", node_list);
if ((n_data->demote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
required = true;
}
n_data->demote = notify_entries_to_strings(n_data->demote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_demote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_demote_uname", node_list);
if ((n_data->promote != NULL)
&& pcmk__str_eq(n_data->action, PCMK_ACTION_PROMOTE, pcmk__str_none)) {
required = true;
}
n_data->promote = notify_entries_to_strings(n_data->promote,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_promote_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_promote_uname", node_list);
n_data->active = notify_entries_to_strings(n_data->active,
&rsc_list, &node_list);
add_notify_env_free_gs(n_data, "notify_active_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_active_uname", node_list);
n_data->unpromoted = notify_entries_to_strings(n_data->unpromoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_unpromoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_unpromoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_slave_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_slave_uname", node_list);
n_data->promoted = notify_entries_to_strings(n_data->promoted,
&rsc_list, &node_list);
add_notify_env_gs(n_data, "notify_promoted_resource", rsc_list);
add_notify_env_gs(n_data, "notify_promoted_uname", node_list);
// Deprecated: kept for backward compatibility with older resource agents
add_notify_env_free_gs(n_data, "notify_master_resource", rsc_list);
add_notify_env_free_gs(n_data, "notify_master_uname", node_list);
n_data->inactive = notify_entries_to_strings(n_data->inactive,
&rsc_list, NULL);
add_notify_env_free_gs(n_data, "notify_inactive_resource", rsc_list);
nodes = g_hash_table_get_values(n_data->allowed_nodes);
if (!pcmk__is_daemon) {
/* For display purposes, sort the node list, for consistent
* regression test output (while avoiding the performance hit
* for the live cluster).
*/
nodes = g_list_sort(nodes, pe__cmp_node_name);
}
get_node_names(nodes, &node_list, NULL);
add_notify_env_free_gs(n_data, "notify_available_uname", node_list);
g_list_free(nodes);
source = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_CONTAINER_ATTRIBUTE_TARGET);
if (pcmk__str_eq(PCMK_VALUE_HOST, source, pcmk__str_none)) {
get_node_names(rsc->priv->scheduler->nodes, &node_list, &metal_list);
add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list);
} else {
get_node_names(rsc->priv->scheduler->nodes, &node_list, NULL);
}
add_notify_env_free_gs(n_data, "notify_all_uname", node_list);
if (required && (n_data->pre != NULL)) {
pcmk__clear_action_flags(n_data->pre, pcmk__action_optional);
pcmk__clear_action_flags(n_data->pre_done, pcmk__action_optional);
}
if (required && (n_data->post != NULL)) {
pcmk__clear_action_flags(n_data->post, pcmk__action_optional);
pcmk__clear_action_flags(n_data->post_done, pcmk__action_optional);
}
}
/*
* \internal
* \brief Find any remote connection start relevant to an action
*
* \param[in] action Action to check
*
* \return If action is behind a remote connection, connection's start
*/
static pcmk_action_t *
find_remote_start(pcmk_action_t *action)
{
if ((action != NULL) && (action->node != NULL)) {
pcmk_resource_t *remote_rsc = action->node->priv->remote;
if (remote_rsc != NULL) {
return find_first_action(remote_rsc->priv->actions, NULL,
PCMK_ACTION_START,
NULL);
}
}
return NULL;
}
/*!
* \internal
* \brief Create notify actions, and add notify data to original actions
*
* \param[in,out] rsc Clone or clone instance that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
static void
create_notify_actions(pcmk_resource_t *rsc, notify_data_t *n_data)
{
GList *iter = NULL;
pcmk_action_t *stop = NULL;
pcmk_action_t *start = NULL;
enum pcmk__action_type task = pcmk__parse_action(n_data->action);
// If this is a clone, call recursively for each instance
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, (GFunc) create_notify_actions,
n_data);
return;
}
// Add notification meta-attributes to original actions
for (iter = rsc->priv->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *op = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(op->flags, pcmk__action_optional)
&& (op->node != NULL)) {
switch (pcmk__parse_action(op->task)) {
case pcmk__action_start:
case pcmk__action_stop:
case pcmk__action_promote:
case pcmk__action_demote:
add_notify_data_to_action_meta(n_data, op);
break;
default:
break;
}
}
}
// Skip notify action itself if original action was not needed
switch (task) {
case pcmk__action_start:
if (n_data->start == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_promote:
if (n_data->promote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
case pcmk__action_demote:
if (n_data->demote == NULL) {
pcmk__rsc_trace(rsc, "No notify action needed for %s %s",
rsc->id, n_data->action);
return;
}
break;
default:
// We cannot do same for stop because it might be implied by fencing
break;
}
pcmk__rsc_trace(rsc, "Creating notify actions for %s %s",
rsc->id, n_data->action);
// Create notify actions for stop or demote
if ((rsc->priv->orig_role != pcmk_role_stopped)
&& ((task == pcmk__action_stop) || (task == pcmk__action_demote))) {
stop = find_first_action(rsc->priv->actions, NULL, PCMK_ACTION_STOP,
NULL);
for (iter = rsc->priv->active_nodes;
iter != NULL; iter = iter->next) {
pcmk_node_t *current_node = (pcmk_node_t *) iter->data;
/* If a stop is a pseudo-action implied by fencing, don't try to
* notify the node getting fenced.
*/
if ((stop != NULL)
&& pcmk_is_set(stop->flags, pcmk__action_pseudo)
&& (current_node->details->unclean
|| pcmk_is_set(current_node->priv->flags,
pcmk__node_remote_reset))) {
continue;
}
new_notify_action(rsc, current_node, n_data->pre,
n_data->pre_done, n_data);
if ((task == pcmk__action_demote) || (stop == NULL)
|| pcmk_is_set(stop->flags, pcmk__action_optional)) {
new_post_notify_action(rsc, current_node, n_data);
}
}
}
// Create notify actions for start or promote
if ((rsc->priv->next_role != pcmk_role_stopped)
&& ((task == pcmk__action_start) || (task == pcmk__action_promote))) {
start = find_first_action(rsc->priv->actions, NULL,
PCMK_ACTION_START, NULL);
if (start != NULL) {
pcmk_action_t *remote_start = find_remote_start(start);
if ((remote_start != NULL)
&& !pcmk_is_set(remote_start->flags, pcmk__action_runnable)) {
/* Start and promote actions for a clone instance behind
* a Pacemaker Remote connection happen after the
* connection starts. If the connection start is blocked, do
* not schedule notifications for these actions.
*/
return;
}
}
if (rsc->priv->assigned_node == NULL) {
pcmk__sched_err(rsc->priv->scheduler,
"Next role '%s' but %s is not allocated",
pcmk_role_text(rsc->priv->next_role), rsc->id);
return;
}
if ((task != pcmk__action_start) || (start == NULL)
|| pcmk_is_set(start->flags, pcmk__action_optional)) {
new_notify_action(rsc, rsc->priv->assigned_node, n_data->pre,
n_data->pre_done, n_data);
}
new_post_notify_action(rsc, rsc->priv->assigned_node, n_data);
}
}
/*!
* \internal
* \brief Create notification data and actions for one clone action
*
* \param[in,out] rsc Clone resource that notification is for
* \param[in,out] n_data Clone notification data for some action
*/
void
pe__create_action_notifications(pcmk_resource_t *rsc, notify_data_t *n_data)
{
if ((rsc == NULL) || (n_data == NULL)) {
return;
}
collect_resource_data(rsc, true, n_data);
add_notif_keys(rsc, n_data);
create_notify_actions(rsc, n_data);
}
/*!
* \internal
* \brief Free notification data for one action
*
* \param[in,out] n_data Notification data to free
*/
void
pe__free_action_notification_data(notify_data_t *n_data)
{
if (n_data == NULL) {
return;
}
g_list_free_full(n_data->stop, free);
g_list_free_full(n_data->start, free);
g_list_free_full(n_data->demote, free);
g_list_free_full(n_data->promote, free);
g_list_free_full(n_data->promoted, free);
g_list_free_full(n_data->unpromoted, free);
g_list_free_full(n_data->active, free);
g_list_free_full(n_data->inactive, free);
pcmk_free_nvpairs(n_data->keys);
free(n_data);
}
/*!
* \internal
* \brief Order clone "notifications complete" pseudo-action after fencing
*
* If a stop action is implied by fencing, the usual notification pseudo-actions
* will not be sufficient to order things properly, or even create all needed
* notifications if the clone is also stopping on another node, and another
* clone is ordered after it. This function creates new notification
* pseudo-actions relative to the fencing to ensure everything works properly.
*
- * \param[in] stop Stop action implied by fencing
- * \param[in,out] rsc Clone resource that notification is for
- * \param[in,out] stonith_op Fencing action that implies \p stop
+ * \param[in] stop Stop action implied by fencing
+ * \param[in,out] rsc Clone resource that notification is for
+ * \param[in,out] fencing Fencing action that implies \p stop
*/
void
pe__order_notifs_after_fencing(const pcmk_action_t *stop, pcmk_resource_t *rsc,
- pcmk_action_t *stonith_op)
+ pcmk_action_t *fencing)
{
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);
+ n_data = pe__action_notif_pseudo_ops(rsc, PCMK_ACTION_STOP, NULL, fencing);
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->priv->name);
create_notify_actions(uber_parent(rsc), n_data);
pe__free_action_notification_data(n_data);
}
}
diff --git a/lib/pengine/status.c b/lib/pengine/status.c
index c4fa0ebfd6..bc70a48087 100644
--- a/lib/pengine/status.c
+++ b/lib/pengine/status.c
@@ -1,369 +1,369 @@
/*
* Copyright 2004-2025 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 <stdint.h> // uint32_t
#include <sys/param.h>
#include <glib.h>
#include <libxml/tree.h> // xmlNode
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/cib_internal.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
#define XPATH_DEPRECATED_RULES \
"//" PCMK_XE_OP_DEFAULTS "//" PCMK_XE_EXPRESSION \
"|//" PCMK_XE_OP "//" PCMK_XE_EXPRESSION
/*!
* \internal
* \brief Log a warning for deprecated rule syntax in operations
*
* \param[in] scheduler Scheduler data
*/
static void
check_for_deprecated_rules(pcmk_scheduler_t *scheduler)
{
// @COMPAT Drop this function when support for the syntax is dropped
xmlNode *deprecated = pcmk__xpath_find_one(scheduler->input->doc,
XPATH_DEPRECATED_RULES,
LOG_NEVER);
if (deprecated != NULL) {
pcmk__warn_once(pcmk__wo_op_attr_expr,
"Support for rules with node attribute expressions in "
PCMK_XE_OP " or " PCMK_XE_OP_DEFAULTS " is deprecated "
"and will be dropped in a future release");
}
}
/*
* Unpack everything
* At the end you'll have:
* - A list of nodes
* - A list of resources (each with any dependencies on other resources)
* - A list of constraints between resources and nodes
* - A list of constraints between start/stop actions
- * - A list of nodes that need to be stonith'd
+ * - A list of nodes that need to be fenced
* - A list of nodes that need to be shutdown
* - A list of the possible stop/start actions (without dependencies)
*
* @TODO Currently this function can modify scheduler->input by creating
* primitive elements for guest nodes. Commit 5dbc819 aimed to make this
* function idempotent. It seems to be idempotent now, except that if
* scheduler->input is reused, it may gain duplicate primitive elements for
* guest nodes each time.
*
* Investigate whether we can leave scheduler->input unmodified without a lot of
* unnecessary XML copying. Otherwise, just document that scheduler->input may
* be modified.
*/
gboolean
cluster_status(pcmk_scheduler_t * scheduler)
{
// @TODO Deprecate, replacing with a safer public alternative if necessary
const char *new_version = NULL;
xmlNode *section = NULL;
if ((scheduler == NULL) || (scheduler->input == NULL)) {
return FALSE;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_have_status)) {
/* cluster_status() has already been called since the last time the
* scheduler was reset. Unpacking the input CIB again would cause
* duplication within the scheduler object's data structures.
*
* The correct return code here is not obvious. Nothing internal checks
* the code, however.
*/
return TRUE;
}
new_version = pcmk__xe_get(scheduler->input, PCMK_XA_CRM_FEATURE_SET);
if (pcmk__check_feature_set(new_version) != pcmk_rc_ok) {
pcmk__config_err("Can't process CIB with feature set '%s' greater than our own '%s'",
new_version, CRM_FEATURE_SET);
return FALSE;
}
crm_trace("Beginning unpack");
pcmk__xml_free(scheduler->priv->failed);
scheduler->priv->failed = pcmk__xe_create(NULL, "failed-ops");
if (scheduler->priv->now == NULL) {
scheduler->priv->now = crm_time_new(NULL);
}
if (pcmk__xe_attr_is_true(scheduler->input, PCMK_XA_HAVE_QUORUM)) {
pcmk__set_scheduler_flags(scheduler, pcmk__sched_quorate);
} else {
pcmk__clear_scheduler_flags(scheduler, pcmk__sched_quorate);
}
scheduler->priv->op_defaults =
pcmk__xpath_find_one(scheduler->input->doc, "//" PCMK_XE_OP_DEFAULTS,
LOG_NEVER);
check_for_deprecated_rules(scheduler);
scheduler->priv->rsc_defaults =
pcmk__xpath_find_one(scheduler->input->doc, "//" PCMK_XE_RSC_DEFAULTS,
LOG_NEVER);
section = pcmk__xpath_find_one(scheduler->input->doc,
"//" PCMK_XE_CRM_CONFIG, LOG_TRACE);
unpack_config(section, scheduler);
if (!pcmk_any_flags_set(scheduler->flags,
pcmk__sched_location_only|pcmk__sched_quorate)
&& (scheduler->no_quorum_policy != pcmk_no_quorum_ignore)) {
pcmk__sched_warn(scheduler,
"Fencing and resource management disabled "
"due to lack of quorum");
}
section = pcmk__xpath_find_one(scheduler->input->doc, "//" PCMK_XE_NODES,
LOG_TRACE);
unpack_nodes(section, scheduler);
section = pcmk__xpath_find_one(scheduler->input->doc,
"//" PCMK_XE_RESOURCES, LOG_TRACE);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
unpack_remote_nodes(section, scheduler);
}
unpack_resources(section, scheduler);
section = pcmk__xpath_find_one(scheduler->input->doc,
"//" PCMK_XE_FENCING_TOPOLOGY, LOG_TRACE);
pcmk__validate_fencing_topology(section);
section = pcmk__xpath_find_one(scheduler->input->doc, "//" PCMK_XE_TAGS,
LOG_NEVER);
unpack_tags(section, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
section = pcmk__xpath_find_one(scheduler->input->doc,
"//" PCMK_XE_STATUS, LOG_TRACE);
unpack_status(section, scheduler);
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_no_counts)) {
for (GList *item = scheduler->priv->resources;
item != NULL; item = item->next) {
pcmk_resource_t *rsc = item->data;
rsc->priv->fns->count(item->data);
}
crm_trace("Cluster resource count: %d (%d disabled, %d blocked)",
scheduler->priv->ninstances,
scheduler->priv->disabled_resources,
scheduler->priv->blocked_resources);
}
if ((scheduler->priv->local_node_name != NULL)
&& (pcmk_find_node(scheduler,
scheduler->priv->local_node_name) == NULL)) {
crm_info("Creating a fake local node for %s",
scheduler->priv->local_node_name);
pe_create_node(scheduler->priv->local_node_name,
scheduler->priv->local_node_name, NULL, 0, scheduler);
}
pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_status);
return TRUE;
}
pcmk_resource_t *
pe_find_resource(GList *rsc_list, const char *id)
{
return pe_find_resource_with_flags(rsc_list, id, pcmk_rsc_match_history);
}
pcmk_resource_t *
pe_find_resource_with_flags(GList *rsc_list, const char *id, enum pe_find flags)
{
GList *rIter = NULL;
for (rIter = rsc_list; id && rIter; rIter = rIter->next) {
pcmk_resource_t *parent = rIter->data;
pcmk_resource_t *match = parent->priv->fns->find_rsc(parent, id, NULL,
(uint32_t) flags);
if (match != NULL) {
return match;
}
}
crm_trace("No match for %s", id);
return NULL;
}
/*!
* \brief Find a node by name or ID in a list of nodes
*
* \param[in] nodes List of nodes (as pcmk_node_t*)
* \param[in] id If not NULL, ID of node to find
* \param[in] node_name If not NULL, name of node to find
*
* \return Node from \p nodes that matches \p id if any,
* otherwise node from \p nodes that matches \p uname if any,
* otherwise NULL
*/
pcmk_node_t *
pe_find_node_any(const GList *nodes, const char *id, const char *uname)
{
pcmk_node_t *match = NULL;
if (id != NULL) {
match = pe_find_node_id(nodes, id);
}
if ((match == NULL) && (uname != NULL)) {
match = pcmk__find_node_in_list(nodes, uname);
}
return match;
}
/*!
* \brief Find a node by ID in a list of nodes
*
* \param[in] nodes List of nodes (as pcmk_node_t*)
* \param[in] id ID of node to find
*
* \return Node from \p nodes that matches \p id if any, otherwise NULL
*/
pcmk_node_t *
pe_find_node_id(const GList *nodes, const char *id)
{
for (const GList *iter = nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
/* @TODO Whether node IDs should be considered case-sensitive should
* probably depend on the node type, so functionizing the comparison
* would be worthwhile
*/
if (pcmk__str_eq(node->priv->id, id, pcmk__str_casei)) {
return node;
}
}
return NULL;
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/pengine/status_compat.h>
pcmk_scheduler_t *
pe_new_working_set(void)
{
return pcmk_new_scheduler();
}
void
pe_reset_working_set(pcmk_scheduler_t *scheduler)
{
if (scheduler == NULL) {
return;
}
pcmk_reset_scheduler(scheduler);
}
void
cleanup_calculations(pcmk_scheduler_t *scheduler)
{
if (scheduler == NULL) {
return;
}
pcmk__clear_scheduler_flags(scheduler, pcmk__sched_have_status);
if (scheduler->priv->options != NULL) {
g_hash_table_destroy(scheduler->priv->options);
}
if (scheduler->priv->singletons != NULL) {
g_hash_table_destroy(scheduler->priv->singletons);
}
if (scheduler->priv->ticket_constraints != NULL) {
g_hash_table_destroy(scheduler->priv->ticket_constraints);
}
if (scheduler->priv->templates != NULL) {
g_hash_table_destroy(scheduler->priv->templates);
}
if (scheduler->priv->tags != NULL) {
g_hash_table_destroy(scheduler->priv->tags);
}
crm_trace("deleting resources");
g_list_free_full(scheduler->priv->resources, pcmk__free_resource);
crm_trace("deleting actions");
g_list_free_full(scheduler->priv->actions, pcmk__free_action);
crm_trace("deleting nodes");
g_list_free_full(scheduler->nodes, pcmk__free_node);
scheduler->nodes = NULL;
pcmk__free_param_checks(scheduler);
g_list_free(scheduler->priv->stop_needed);
crm_time_free(scheduler->priv->now);
pcmk__xml_free(scheduler->input);
pcmk__xml_free(scheduler->priv->failed);
pcmk__xml_free(scheduler->priv->graph);
set_working_set_defaults(scheduler);
CRM_LOG_ASSERT((scheduler->priv->location_constraints == NULL)
&& (scheduler->priv->ordering_constraints == NULL));
}
void
set_working_set_defaults(pcmk_scheduler_t *scheduler)
{
// These members must be preserved
pcmk__scheduler_private_t *priv = scheduler->priv;
pcmk__output_t *out = priv->out;
char *local_node_name = scheduler->priv->local_node_name;
// Wipe the main structs (any other members must have previously been freed)
memset(scheduler, 0, sizeof(pcmk_scheduler_t));
memset(priv, 0, sizeof(pcmk__scheduler_private_t));
// Restore the members to preserve
scheduler->priv = priv;
scheduler->priv->out = out;
scheduler->priv->local_node_name = local_node_name;
// Set defaults for everything else
pcmk__set_scheduler_defaults(scheduler);
}
void
pe_free_working_set(pcmk_scheduler_t *scheduler)
{
pcmk_free_scheduler(scheduler);
}
pcmk_node_t *
pe_find_node(const GList *nodes, const char *node_name)
{
return pcmk__find_node_in_list(nodes, node_name);
}
// LCOV_EXCL_STOP
// End deprecated API
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index e84853dc00..7bf7147d11 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5131 +1,5132 @@
/*
* Copyright 2004-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <time.h>
#include <glib.h>
#include <libxml/tree.h> // xmlNode
#include <libxml/xpath.h> // xmlXPathObject, etc.
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
// A (parsed) resource action history entry
struct action_history {
pcmk_resource_t *rsc; // Resource that history is for
pcmk_node_t *node; // Node that history is for
xmlNode *xml; // History entry XML
// Parsed from entry XML
const char *id; // XML ID of history entry
const char *key; // Operation key of action
const char *task; // Action name
const char *exit_reason; // Exit reason given for result
guint interval_ms; // Action interval
int call_id; // Call ID of action
int expected_exit_status; // Expected exit status of action
int exit_status; // Actual exit status of action
int execution_status; // Execution status of action
};
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(scheduler, option, flag) do { \
GHashTable *config_hash = (scheduler)->priv->options; \
const char *scf_value = pcmk__cluster_option(config_hash, (option)); \
\
if (scf_value != NULL) { \
if (pcmk__is_true(scf_value)) { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} else { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node,
xmlNode *xml_op, xmlNode **last_failure,
enum pcmk__on_fail *failed);
static void determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node);
static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node,
bool overwrite, pcmk_scheduler_t *scheduler);
static void determine_online_status(const xmlNode *node_state,
pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler);
static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler);
/*!
* \internal
* \brief Check whether a node is a dangling guest node
*
* \param[in] node Node to check
*
* \return true if \p node had a Pacemaker Remote connection resource with a
* launcher that was removed from the CIB, otherwise false.
*/
static bool
is_dangling_guest_node(pcmk_node_t *node)
{
return pcmk__is_pacemaker_remote_node(node)
&& (node->priv->remote != NULL)
&& (node->priv->remote->priv->launcher == NULL)
&& pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_removed_launched);
}
/*!
* \brief Schedule a fence action for a node
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] node Node to fence
* \param[in] reason Text description of why fencing is needed
* \param[in] priority_delay Whether to consider
* \c PCMK_OPT_PRIORITY_FENCING_DELAY
*/
void
pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node,
const char *reason, bool priority_delay)
{
CRM_CHECK(node, return);
if (pcmk__is_guest_or_bundle_node(node)) {
// Fence a guest or bundle node by marking its launcher as failed
pcmk_resource_t *rsc = node->priv->remote->priv->launcher;
if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing guest node %s "
"(otherwise would because %s): "
"its guest resource %s is unmanaged",
pcmk__node_name(node), reason, rsc->id);
} else {
pcmk__sched_warn(scheduler,
"Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pcmk__node_name(node), rsc->id, reason);
/* We don't mark the node as unclean because that would prevent the
* node from running resources. We want to allow it to run resources
* in this transition if the recovery succeeds.
*/
pcmk__set_node_flags(node, pcmk__node_remote_reset);
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
}
} else if (is_dangling_guest_node(node)) {
crm_info("Cleaning up dangling connection for guest node %s: "
"fencing was already done because %s, "
"and guest resource no longer exists",
pcmk__node_name(node), reason);
pcmk__set_rsc_flags(node->priv->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
} else if (pcmk__is_remote_node(node)) {
pcmk_resource_t *rsc = node->priv->remote;
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Not fencing remote node %s "
"(otherwise would because %s): connection is unmanaged",
pcmk__node_name(node), reason);
} else if (!pcmk_is_set(node->priv->flags, pcmk__node_remote_reset)) {
pcmk__set_node_flags(node, pcmk__node_remote_reset);
pcmk__sched_warn(scheduler, "Remote node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
pcmk__sched_warn(scheduler, "Cluster node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \
"[(@" PCMK_XA_NAME "='" PCMK_FENCING_PROVIDES "'" \
"or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \
"and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \
"/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
xmlXPathObject *result = NULL;
if (!pcmk_is_set(scheduler->flags, flag)) {
result = pcmk__xpath_search(scheduler->input->doc, xpath);
if (pcmk__xpath_num_results(result) > 0) {
pcmk__set_scheduler_flags(scheduler, flag);
}
xmlXPathFreeObject(result);
}
}
gboolean
unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
GHashTable *config_hash = pcmk__strkey_table(free, free);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
};
scheduler->priv->options = config_hash;
pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET,
&rule_input, config_hash,
PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, scheduler);
pcmk__validate_cluster_options(config_hash);
set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES,
pcmk__sched_probe_resources);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_probe_resources)) {
crm_info("Startup probes: disabled (dangerous)");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG);
if (pcmk__is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and " PCMK_OPT_FENCING_WATCHDOG_TIMEOUT
" is nonzero");
pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_fencing);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pcmk__sched_enable_unfencing, XPATH_ENABLE_UNFENCING,
scheduler);
value = pcmk__cluster_option(config_hash, PCMK_OPT_FENCING_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->priv->fence_timeout_ms));
crm_debug("Default fencing action timeout: %s",
pcmk__readable_interval(scheduler->priv->fence_timeout_ms));
set_config_flag(scheduler, PCMK_OPT_FENCING_ENABLED,
pcmk__sched_fencing_enabled);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
crm_debug("Fencing of failed nodes is enabled");
} else {
crm_debug("Fencing of failed nodes is disabled");
}
scheduler->priv->fence_action =
pcmk__cluster_option(config_hash, PCMK_OPT_FENCING_ACTION);
crm_trace("Fencing will %s nodes", scheduler->priv->fence_action);
set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING,
pcmk__sched_concurrent_fencing);
if (pcmk_is_set(scheduler->flags, pcmk__sched_concurrent_fencing)) {
crm_debug("Concurrent fencing is enabled");
} else {
crm_debug("Concurrent fencing is disabled");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY);
if (value) {
pcmk_parse_interval_spec(value,
&(scheduler->priv->priority_fencing_ms));
crm_trace("Priority fencing delay is %s",
pcmk__readable_interval(scheduler->priv->priority_fencing_ms));
}
set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES,
pcmk__sched_stop_all);
crm_debug("Stop all active resources: %s",
pcmk__flag_text(scheduler->flags, pcmk__sched_stop_all));
set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER,
pcmk__sched_symmetric_cluster);
if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY);
if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__strcase_any_of(value, PCMK_VALUE_FENCE,
PCMK_VALUE_FENCE_LEGACY, NULL)) {
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
int do_panic = 0;
pcmk__xe_get_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC,
&do_panic);
if (do_panic
|| pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
scheduler->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop': cluster has never had quorum");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop' because fencing is disabled");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
crm_debug("On loss of quorum: "
"Freeze resources that require quorum");
break;
case pcmk_no_quorum_stop:
crm_debug("On loss of quorum: "
"Stop resources that require quorum");
break;
case pcmk_no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case pcmk_no_quorum_fence:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case pcmk_no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(scheduler, PCMK_OPT_STOP_REMOVED_RESOURCES,
pcmk__sched_stop_removed_resources);
if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) {
crm_trace("Removed resources are stopped");
} else {
crm_trace("Removed resources are ignored");
}
set_config_flag(scheduler, PCMK_OPT_STOP_REMOVED_ACTIONS,
pcmk__sched_cancel_removed_actions);
if (pcmk_is_set(scheduler->flags, pcmk__sched_cancel_removed_actions)) {
crm_trace("Removed resource actions are stopped");
} else {
crm_trace("Removed resource actions are ignored");
}
set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE,
pcmk__sched_in_maintenance);
crm_trace("Maintenance mode: %s",
pcmk__flag_text(scheduler->flags, pcmk__sched_in_maintenance));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk__sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk__sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk__sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->priv->placement_strategy =
pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY);
crm_trace("Placement strategy: %s", scheduler->priv->placement_strategy);
set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK,
pcmk__sched_shutdown_lock);
if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT);
pcmk_parse_interval_spec(value, &(scheduler->priv->shutdown_lock_ms));
crm_trace("Resources will be locked to nodes that were cleanly "
"shut down (locks expire after %s)",
pcmk__readable_interval(scheduler->priv->shutdown_lock_ms));
} else {
crm_trace("Resources will not be locked to nodes that were cleanly "
"shut down");
}
value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->priv->node_pending_ms));
if (scheduler->priv->node_pending_ms == 0U) {
crm_trace("Do not fence pending nodes");
} else {
crm_trace("Fence pending nodes after %s",
pcmk__readable_interval(scheduler->priv->node_pending_ms));
}
set_config_flag(scheduler, PCMK_OPT_FENCE_REMOTE_WITHOUT_QUORUM,
pcmk__sched_fence_remote_no_quorum);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fence_remote_no_quorum)) {
crm_trace("Pacemaker Remote nodes may be fenced without quorum");
} else {
crm_trace("Pacemaker Remote nodes require quorum to be fenced");
}
return TRUE;
}
/*!
* \internal
* \brief Create a new node object in scheduler data
*
* \param[in] id ID of new node
* \param[in] uname Name of new node
* \param[in] type Type of new node
* \param[in] score Score of new node
* \param[in,out] scheduler Scheduler data
*
* \return Newly created node object
* \note The returned object is part of the scheduler data and should not be
* freed separately.
*/
pcmk_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
int score, pcmk_scheduler_t *scheduler)
{
enum pcmk__node_variant variant = pcmk__node_variant_cluster;
pcmk_node_t *new_node = NULL;
if (pcmk_find_node(scheduler, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
if (pcmk__str_eq(type, PCMK_VALUE_MEMBER,
pcmk__str_null_matches|pcmk__str_casei)) {
variant = pcmk__node_variant_cluster;
} else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) {
variant = pcmk__node_variant_remote;
} else {
pcmk__config_err("Ignoring node %s with unrecognized type '%s'",
pcmk__s(uname, "without name"), type);
return NULL;
}
new_node = calloc(1, sizeof(pcmk_node_t));
if (new_node == NULL) {
pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
uname);
return NULL;
}
new_node->assign = calloc(1, sizeof(struct pcmk__node_assignment));
new_node->details = calloc(1, sizeof(struct pcmk__node_details));
new_node->priv = calloc(1, sizeof(pcmk__node_private_t));
if ((new_node->assign == NULL) || (new_node->details == NULL)
|| (new_node->priv == NULL)) {
free(new_node->assign);
free(new_node->details);
free(new_node->priv);
free(new_node);
pcmk__sched_err(scheduler, "Could not allocate memory for node %s",
uname);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->assign->score = score;
new_node->priv->id = id;
new_node->priv->name = uname;
new_node->priv->flags = pcmk__node_probes_allowed;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->running_rsc = NULL;
new_node->priv->scheduler = scheduler;
new_node->priv->variant = variant;
new_node->priv->attrs = pcmk__strkey_table(free, free);
new_node->priv->utilization = pcmk__strkey_table(free, free);
new_node->priv->digest_cache = pcmk__strkey_table(free, pe__free_digests);
if (pcmk__is_pacemaker_remote_node(new_node)) {
pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "remote");
pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_remote_nodes);
} else {
pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "cluster");
}
scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = pcmk__xe_id(xml_obj);
const char *remote_name = NULL;
const char *remote_server = NULL;
const char *remote_port = NULL;
const char *connect_timeout = "60s";
const char *remote_allow_migrate=NULL;
const char *is_managed = NULL;
// @TODO This doesn't handle rules or id-ref
for (attr_set = pcmk__xe_first_child(xml_obj, PCMK_XE_META_ATTRIBUTES,
NULL, NULL);
attr_set != NULL;
attr_set = pcmk__xe_next(attr_set, PCMK_XE_META_ATTRIBUTES)) {
for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL);
attr != NULL; attr = pcmk__xe_next(attr, NULL)) {
const char *value = pcmk__xe_get(attr, PCMK_XA_VALUE);
const char *name = pcmk__xe_get(attr, PCMK_XA_NAME);
if (name == NULL) { // Sanity
continue;
}
if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) {
remote_name = value;
} else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) {
remote_server = value;
} else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) {
remote_port = value;
} else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) {
connect_timeout = value;
} else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) {
remote_allow_migrate = value;
} else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->priv->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node)
{
if ((new_node->priv->variant == pcmk__node_variant_remote)
&& (new_node->priv->remote == NULL)) {
/* Ignore fencing for remote nodes that don't have a connection resource
* associated with them. This happens when remote node entries get left
* in the nodes section after the connection resource is removed.
*/
return;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) {
// All nodes are unclean until we've seen their status entry
new_node->details->unclean = TRUE;
} else {
// Blind faith ...
new_node->details->unclean = FALSE;
}
}
gboolean
unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
pcmk_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
for (xml_obj = pcmk__xe_first_child(xml_nodes, PCMK_XE_NODE, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, PCMK_XE_NODE)) {
int score = 0;
int rc = pcmk__xe_get_score(xml_obj, PCMK_XA_SCORE, &score, 0);
new_node = NULL;
id = pcmk__xe_get(xml_obj, PCMK_XA_ID);
uname = pcmk__xe_get(xml_obj, PCMK_XA_UNAME);
type = pcmk__xe_get(xml_obj, PCMK_XA_TYPE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" PCMK_XE_NODE
"> entry in configuration without id");
continue;
}
if (rc != pcmk_rc_ok) {
// Not possible with schema validation enabled
pcmk__config_warn("Using 0 as score for node %s "
"because '%s' is not a valid score: %s",
pcmk__s(uname, "without name"),
pcmk__xe_get(xml_obj, PCMK_XA_SCORE),
pcmk_rc_str(rc));
}
new_node = pe_create_node(id, uname, type, score, scheduler);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(scheduler, new_node);
add_node_attrs(xml_obj, new_node, FALSE, scheduler);
crm_trace("Done with node %s", pcmk__xe_get(xml_obj, PCMK_XA_UNAME));
}
return TRUE;
}
static void
unpack_launcher(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *launcher_id = NULL;
if (rsc->priv->children != NULL) {
g_list_foreach(rsc->priv->children, (GFunc) unpack_launcher,
scheduler);
return;
}
launcher_id = g_hash_table_lookup(rsc->priv->meta, PCMK__META_CONTAINER);
if ((launcher_id != NULL)
&& !pcmk__str_eq(launcher_id, rsc->id, pcmk__str_none)) {
pcmk_resource_t *launcher = pe_find_resource(scheduler->priv->resources,
launcher_id);
if (launcher != NULL) {
rsc->priv->launcher = launcher;
launcher->priv->launched =
g_list_append(launcher->priv->launched, rsc);
pcmk__rsc_trace(rsc, "Resource %s's launcher is %s",
rsc->id, launcher_id);
} else {
pcmk__config_err("Resource %s: Unknown " PCMK__META_CONTAINER " %s",
rsc->id, launcher_id);
}
}
}
gboolean
unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
/* Create remote nodes and guest nodes from the resource configuration
* before unpacking resources.
*/
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) {
const char *new_node_id = NULL;
/* Check for remote nodes, which are defined by ocf:pacemaker:remote
* primitives.
*/
if (xml_contains_remote_node(xml_obj)) {
new_node_id = pcmk__xe_id(xml_obj);
/* The pcmk_find_node() check ensures we don't iterate over an
* expanded node that has already been added to the node list
*/
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) {
/* This will add an ocf:pacemaker:remote primitive to the
* configuration for the guest node's connection, to be unpacked
* later.
*/
new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s",
new_node_id, pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL);
xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2, NULL)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources,
scheduler);
if (new_node_id
&& (pcmk_find_node(scheduler, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, pcmk__xe_id(xml_obj2),
pcmk__xe_id(xml_obj));
pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE,
0, scheduler);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc)
{
pcmk_node_t *remote_node = NULL;
if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) {
return;
}
if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
/* remote_nodes and remote_resources are not linked in quick location calculations */
return;
}
remote_node = pcmk_find_node(scheduler, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s",
new_rsc->id, pcmk__node_name(remote_node));
remote_node->priv->remote = new_rsc;
if (new_rsc->priv->launcher == NULL) {
/* Handle start-up fencing for remote nodes (as opposed to guest nodes)
* the same as is done for cluster nodes.
*/
handle_startup_fencing(scheduler, remote_node);
} else {
/* pe_create_node() marks the new node as "remote" or "cluster"; now
* that we know the node is a guest node, update it correctly.
*/
pcmk__insert_dup(remote_node->priv->attrs,
CRM_ATTR_KIND, "container");
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] scheduler Scheduler data
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when pe__unpack_resource() calls resource_location()
*/
gboolean
unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
scheduler->priv->templates = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL);
xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) {
pcmk_resource_t *new_rsc = NULL;
const char *id = pcmk__xe_id(xml_obj);
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring <%s> resource without ID",
xml_obj->name);
continue;
}
if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) {
if (g_hash_table_lookup_extended(scheduler->priv->templates, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
pcmk__insert_dup(scheduler->priv->templates, id, NULL);
}
continue;
}
crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id);
if (pe__unpack_resource(xml_obj, &new_rsc, NULL,
scheduler) == pcmk_rc_ok) {
scheduler->priv->resources =
g_list_append(scheduler->priv->resources, new_rsc);
pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
xml_obj->name, id);
}
}
for (gIter = scheduler->priv->resources;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
unpack_launcher(rsc, scheduler);
link_rsc2remotenode(scheduler, rsc);
}
scheduler->priv->resources = g_list_sort(scheduler->priv->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) {
/* Ignore */
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) {
pcmk__config_err("Resource start-up disabled since no fencing "
"resources have been defined. Either configure some "
"or disable fencing with the "
PCMK_OPT_FENCING_ENABLED " option. NOTE: Clusters "
"with shared data need fencing to ensure data "
"integrity.");
}
return TRUE;
}
/*!
* \internal
* \brief Validate the levels in a fencing topology
*
* \param[in] xml \c PCMK_XE_FENCING_TOPOLOGY element
*/
void
pcmk__validate_fencing_topology(const xmlNode *xml)
{
if (xml == NULL) {
return;
}
CRM_CHECK(pcmk__xe_is(xml, PCMK_XE_FENCING_TOPOLOGY), return);
for (const xmlNode *level = pcmk__xe_first_child(xml, PCMK_XE_FENCING_LEVEL,
NULL, NULL);
level != NULL; level = pcmk__xe_next(level, PCMK_XE_FENCING_LEVEL)) {
const char *id = pcmk__xe_id(level);
int index = 0;
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring fencing level without ID");
continue;
}
if (pcmk__xe_get_int(level, PCMK_XA_INDEX, &index) != pcmk_rc_ok) {
pcmk__config_err("Ignoring fencing level %s with invalid index",
id);
continue;
}
if ((index < ST__LEVEL_MIN) || (index > ST__LEVEL_MAX)) {
pcmk__config_err("Ignoring fencing level %s with out-of-range "
"index %d",
id, index);
}
}
}
gboolean
unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_tag = NULL;
scheduler->priv->tags = pcmk__strkey_table(free, pcmk__free_idref);
for (xml_tag = pcmk__xe_first_child(xml_tags, PCMK_XE_TAG, NULL, NULL);
xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag, PCMK_XE_TAG)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = pcmk__xe_id(xml_tag);
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID,
(const char *) xml_tag->name);
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag, PCMK_XE_OBJ_REF,
NULL, NULL);
xml_obj_ref != NULL;
xml_obj_ref = pcmk__xe_next(xml_obj_ref, PCMK_XE_OBJ_REF)) {
const char *obj_ref = pcmk__xe_id(xml_obj_ref);
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID,
xml_obj_ref->name, tag_id);
continue;
}
pcmk__add_idref(scheduler->priv->tags, tag_id, obj_ref);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack a ticket state entry
*
* \param[in] xml_ticket XML ticket state to unpack
* \param[in,out] userdata Scheduler data
*
* \return pcmk_rc_ok (to always continue unpacking further entries)
*/
static int
unpack_ticket_state(xmlNode *xml_ticket, void *userdata)
{
pcmk_scheduler_t *scheduler = userdata;
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pcmk__ticket_t *ticket = NULL;
ticket_id = pcmk__xe_id(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return pcmk_rc_ok;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(scheduler->priv->ticket_constraints,
ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, scheduler);
if (ticket == NULL) {
return pcmk_rc_ok;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) {
continue;
}
pcmk__insert_dup(ticket->state, prop_name, prop_value);
}
granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED);
if (pcmk__is_true(granted)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We have ticket '%s'", ticket->id);
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_granted);
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED);
if (last_granted) {
long long last_granted_ll = 0LL;
int rc = pcmk__scan_ll(last_granted, &last_granted_ll, 0LL);
if (rc != pcmk_rc_ok) {
crm_warn("Using %lld instead of invalid " PCMK_XA_LAST_GRANTED
" value '%s' in state for ticket %s: %s",
last_granted_ll, last_granted, ticket->id,
pcmk_rc_str(rc));
}
ticket->last_granted = (time_t) last_granted_ll;
}
standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY);
if (pcmk__is_true(standby)) {
pcmk__set_ticket_flags(ticket, pcmk__ticket_standby);
if (pcmk_is_set(ticket->flags, pcmk__ticket_granted)) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
pcmk__clear_ticket_flags(ticket, pcmk__ticket_standby);
}
crm_trace("Done with ticket state for %s", ticket_id);
return pcmk_rc_ok;
}
static void
unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = NULL;
pcmk_resource_t *rsc = NULL;
int maint = 0;
if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
return;
}
if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) {
return;
}
crm_trace("Processing Pacemaker Remote node %s",
pcmk__node_name(this_node));
pcmk__scan_min_int(pcmk__xe_get(state, PCMK__XA_NODE_IN_MAINTENANCE),
&maint, 0);
if (maint) {
pcmk__set_node_flags(this_node, pcmk__node_remote_maint);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_maint);
}
rsc = this_node->priv->remote;
if (!pcmk_is_set(this_node->priv->flags, pcmk__node_remote_reset)) {
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
}
attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL,
NULL);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (pcmk__is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
}
if (pcmk__is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE,
NULL, pcmk__rsc_node_current))
|| ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) {
crm_info("%s is in maintenance mode", pcmk__node_name(this_node));
this_node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(this_node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !pcmk__is_true(discovery)) {
pcmk__warn_once(pcmk__wo_rdisc_enabled,
"Support for the "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" node attribute is deprecated and will be removed"
" (and behave as 'true') in a future release.");
if (pcmk__is_remote_node(this_node)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute on Pacemaker Remote node %s"
" because fencing is disabled",
pcmk__node_name(this_node));
} else {
/* This is either a remote node with fencing enabled, or a guest
* node. We don't care whether fencing is enabled when fencing guest
* nodes, because they are "fenced" by recovering their containing
* resource.
*/
crm_info("%s has resource discovery disabled",
pcmk__node_name(this_node));
pcmk__clear_node_flags(this_node, pcmk__node_probes_allowed);
}
}
}
/*!
* \internal
* \brief Unpack a cluster node's transient attributes
*
* \param[in] state CIB node state XML
* \param[in,out] node Cluster node whose attributes are being unpacked
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
const char *discovery = NULL;
const xmlNode *attrs = pcmk__xe_first_child(state,
PCMK__XE_TRANSIENT_ATTRIBUTES,
NULL, NULL);
add_node_attrs(attrs, node, TRUE, scheduler);
if (pcmk__is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in standby mode", pcmk__node_name(node));
pcmk__set_node_flags(node, pcmk__node_standby);
}
if (pcmk__is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL,
pcmk__rsc_node_current))) {
crm_info("%s is in maintenance mode", pcmk__node_name(node));
node->details->maintenance = TRUE;
}
discovery = pcmk__node_attr(node,
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED,
NULL, pcmk__rsc_node_current);
if ((discovery != NULL) && !pcmk__is_true(discovery)) {
pcmk__config_warn("Ignoring "
PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED
" attribute for %s because disabling resource"
" discovery is not allowed for cluster nodes",
pcmk__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not
* the resource history inside it. Multiple passes through the status are needed
* to fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = pcmk__xe_get(state, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without "
PCMK_XA_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = pcmk__xe_get(state, PCMK_XA_UNAME);
if (uname == NULL) {
/* If a joining peer makes the cluster acquire the quorum from Corosync
* but has not joined the controller CPG membership yet, it's possible
* that the created PCMK__XE_NODE_STATE entry doesn't have a
* PCMK_XA_UNAME yet. Recognize the node as pending and wait for it to
* join CPG.
*/
crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" "
"without " PCMK_XA_UNAME,
id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
crm_notice("Ignoring recorded state for removed node with name %s and "
PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id);
return;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
int remote_fenced = 0;
/* We can't determine the online status of Pacemaker Remote nodes until
* after all resource history has been unpacked. In this first pass, we
* do need to mark whether the node has been fenced, as this plays a
* role during unpacking cluster node resource state.
*/
pcmk__scan_min_int(pcmk__xe_get(state, PCMK__XA_NODE_FENCED),
&remote_fenced, 0);
if (remote_fenced) {
pcmk__set_node_flags(this_node, pcmk__node_remote_fenced);
} else {
pcmk__clear_node_flags(this_node, pcmk__node_remote_fenced);
}
return;
}
unpack_transient_attributes(state, this_node, scheduler);
/* Provisionally mark this cluster node as clean. We have at least seen it
* in the current cluster's lifetime.
*/
this_node->details->unclean = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_seen);
crm_trace("Determining online status of cluster node %s (id %s)",
pcmk__node_name(this_node), id);
determine_online_status(state, this_node, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_quorate)
&& this_node->details->online
&& (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(scheduler, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(const xmlNode *status, bool fence,
pcmk_scheduler_t *scheduler)
{
int rc = pcmk_rc_ok;
// Loop through all PCMK__XE_NODE_STATE entries in CIB status
for (const xmlNode *state = pcmk__xe_first_child(status,
PCMK__XE_NODE_STATE, NULL,
NULL);
state != NULL; state = pcmk__xe_next(state, PCMK__XE_NODE_STATE)) {
const char *id = pcmk__xe_id(state);
const char *uname = pcmk__xe_get(state, PCMK_XA_UNAME);
pcmk_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
PCMK__XE_NODE_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) {
crm_trace("Not unpacking resource history for node %s because "
"already unpacked", id);
continue;
}
if (fence) {
// We're processing all remaining nodes
} else if (pcmk__is_guest_or_bundle_node(this_node)) {
/* We can unpack a guest node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection and containing resource are both up.
*/
const pcmk_resource_t *remote = this_node->priv->remote;
const pcmk_resource_t *launcher = remote->priv->launcher;
if ((remote->priv->orig_role != pcmk_role_started)
|| (launcher->priv->orig_role != pcmk_role_started)) {
crm_trace("Not unpacking resource history for guest node %s "
"because launcher and connection are not known to "
"be up", id);
continue;
}
} else if (pcmk__is_remote_node(this_node)) {
/* We can unpack a remote node's history only after we've unpacked
* other resource history to the point that we know that the node's
* connection is up, with the exception of when shutdown locks are
* in use.
*/
pcmk_resource_t *rsc = this_node->priv->remote;
if ((rsc == NULL)
|| (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)
&& (rsc->priv->orig_role != pcmk_role_started))) {
crm_trace("Not unpacking resource history for remote node %s "
"because connection is not known to be up", id);
continue;
}
/* If fencing and shutdown locks are disabled and we're not processing
* unseen nodes, then we don't want to unpack offline nodes until online
* nodes have been unpacked. This allows us to number active clone
* instances first.
*/
} else if (!pcmk_any_flags_set(scheduler->flags,
pcmk__sched_fencing_enabled
|pcmk__sched_shutdown_lock)
&& !this_node->details->online) {
crm_trace("Not unpacking resource history for offline "
"cluster node %s", id);
continue;
}
if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
unpack_handle_remote_attrs(this_node, state, scheduler);
}
crm_trace("Unpacking resource history for %snode %s",
(fence? "unseen " : ""), id);
pcmk__set_node_flags(this_node, pcmk__node_unpacked);
unpack_node_lrm(this_node, state, scheduler);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (scheduler->priv->ticket_constraints == NULL) {
scheduler->priv->ticket_constraints =
pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL;
state = pcmk__xe_next(state, NULL)) {
if (pcmk__xe_is(state, PCMK_XE_TICKETS)) {
pcmk__xe_foreach_child(state, PCMK__XE_TICKET_STATE,
unpack_ticket_state, scheduler);
} else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) {
unpack_node_state(state, scheduler);
}
}
while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(scheduler->flags,
pcmk__sched_fencing_enabled),
scheduler);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (scheduler->priv->stop_needed != NULL) {
for (GList *item = scheduler->priv->stop_needed;
item != NULL; item = item->next) {
pcmk_resource_t *container = item->data;
pcmk_node_t *node = pcmk__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(scheduler->priv->stop_needed);
scheduler->priv->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *this_node = gIter->data;
if (!pcmk__is_pacemaker_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->priv->remote != NULL)) {
pe__set_next_role(this_node->priv->remote, pcmk_role_stopped,
"remote shutdown");
}
if (!pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) {
determine_remote_online_status(scheduler, this_node);
}
}
return TRUE;
}
/*!
* \internal
* \brief Unpack node's time when it became a member at the cluster layer
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = pcmk__xe_get(node_state, PCMK__XA_IN_CCM);
bool is_member = false;
if (member_time == NULL) {
return -1LL;
}
if (pcmk__parse_bool(member_time, &is_member) != pcmk_rc_ok) {
long long when_member = 0LL;
if ((pcmk__scan_ll(member_time, &when_member,
0LL) != pcmk_rc_ok) || (when_member < 0LL)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM
" in " PCMK__XE_NODE_STATE " entry", member_time);
return -1LL;
}
return when_member;
}
/* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was
* recorded as a boolean for a DC < 2.1.7, or the node is pending shutdown
* and has left the CPG, in which case it was set to 1 to avoid fencing for
* PCMK_OPT_NODE_PENDING_TIMEOUT.
*
* We return the effective time for in_ccm=1 because what's important to
* avoid fencing is that effective time minus this value is less than the
* pending node timeout.
*/
return is_member? (long long) pcmk__scheduler_epoch_time(scheduler) : 0LL;
}
/*!
* \internal
* \brief Unpack node's time when it became online in process group
*
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = pcmk__xe_get(node_state, PCMK_XA_CRMD);
// @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline"
if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE,
pcmk__str_casei|pcmk__str_null_matches)) {
return 0LL;
} else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) {
return 1LL;
} else {
long long when_online = 0LL;
if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok)
|| (when_online < 0)) {
crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in "
PCMK__XE_NODE_STATE " entry, assuming offline", peer_time);
return 0LL;
}
return when_online;
}
}
/*!
* \internal
* \brief Unpack node attribute for user-requested fencing
*
* \param[in] node Node to check
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
bool value_b = false;
long long value_ll = 0LL;
int rc = pcmk_rc_ok;
const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE,
NULL, pcmk__rsc_node_current);
// Value may be boolean or an epoch time
if ((value_s != NULL)
&& (pcmk__parse_bool(value_s, &value_b) == pcmk_rc_ok)) {
return value_b;
}
rc = pcmk__scan_ll(value_s, &value_ll, 0LL);
if (rc == pcmk_rc_ok) {
return (value_ll > 0);
}
crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE
"node attribute for %s: %s",
value_s, pcmk__node_name(node), pcmk_rc_str(rc));
return false;
}
static gboolean
determine_online_status_no_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
gboolean online = FALSE;
const char *join = pcmk__xe_get(node_state, PCMK__XA_JOIN);
const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
if (when_member <= 0) {
crm_trace("Node %s is %sdown", pcmk__node_name(this_node),
((when_member < 0)? "presumed " : ""));
} else if (when_online > 0) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node %s is not ready to run resources: %s",
pcmk__node_name(this_node), join);
}
} else if (!pcmk_is_set(this_node->priv->flags,
pcmk__node_expected_up)) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE);
crm_info("Node %s member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->priv->node_pending_ms > 0U)
&& (when_member > 0) && (when_online <= 0)) {
// There is a timeout on pending nodes, and node is pending
time_t timeout = when_member
+ pcmk__timeout_ms2s(scheduler->priv->node_pending_ms);
if (pcmk__scheduler_epoch_time(node->priv->scheduler) >= timeout) {
return true; // Node has timed out
}
// Node is pending, but still has time
pcmk__update_recheck_time(timeout, scheduler, "pending node timeout");
}
return false;
}
static bool
determine_online_status_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
bool termination_requested = unpack_node_terminate(this_node, node_state);
const char *join = pcmk__xe_get(node_state, PCMK__XA_JOIN);
const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
/*
- PCMK__XA_JOIN ::= member|down|pending|banned
- PCMK_XA_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- PCMK__XA_IN_CCM ::= true|false
- PCMK_XA_CRMD ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- PCMK__XA_IN_CCM ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- PCMK_XA_CRMD ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"),
(termination_requested? " (termination requested)" : ""));
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", pcmk__node_name(this_node));
/* Slightly different criteria since we can't shut down a dead peer */
return (when_online > 0);
}
if (when_member < 0) {
pe_fence_node(scheduler, this_node,
"peer has not been seen by the cluster", FALSE);
return false;
}
if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) {
pe_fence_node(scheduler, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (termination_requested) {
if ((when_member <= 0) && (when_online <= 0)
&& pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) {
crm_info("%s was fenced as requested", pcmk__node_name(this_node));
return false;
}
pe_fence_node(scheduler, this_node, "fencing was requested", false);
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN,
pcmk__str_null_matches)) {
if (pending_too_long(scheduler, this_node, when_member, when_online)) {
pe_fence_node(scheduler, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if ((when_member > 0) || (when_online > 0)) {
crm_info("- %s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pcmk__node_name(this_node));
}
} else if (when_member <= 0) {
// Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes
pe_fence_node(scheduler, this_node,
"peer is no longer part of the cluster", TRUE);
} else if (when_online <= 0) {
pe_fence_node(scheduler, this_node,
"peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) {
crm_info("%s is active", pcmk__node_name(this_node));
} else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING,
CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("%s is not ready to run resources",
pcmk__node_name(this_node));
pcmk__set_node_flags(this_node, pcmk__node_standby);
this_node->details->pending = TRUE;
} else {
pe_fence_node(scheduler, this_node, "peer was in an unknown state",
FALSE);
}
return (when_member > 0);
}
static void
determine_remote_online_status(pcmk_scheduler_t *scheduler,
pcmk_node_t *this_node)
{
pcmk_resource_t *rsc = this_node->priv->remote;
pcmk_resource_t *launcher = NULL;
pcmk_node_t *host = NULL;
const char *node_type = "Remote";
if (rsc == NULL) {
/* This is a leftover node state entry for a former Pacemaker Remote
* node whose connection resource was removed. Consider it offline.
*/
crm_trace("Pacemaker Remote node %s is considered OFFLINE because "
"its connection resource has been removed from the CIB",
this_node->priv->id);
this_node->details->online = FALSE;
return;
}
launcher = rsc->priv->launcher;
if (launcher != NULL) {
node_type = "Guest";
if (pcmk__list_of_1(rsc->priv->active_nodes)) {
host = rsc->priv->active_nodes->data;
}
}
/* If the resource is currently started, mark it online. */
if (rsc->priv->orig_role == pcmk_role_started) {
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if ((rsc->priv->orig_role == pcmk_role_started)
&& (rsc->priv->next_role == pcmk_role_stopped)) {
crm_trace("%s node %s shutting down because connection resource is stopping",
node_type, this_node->priv->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if ((launcher != NULL) && pcmk_is_set(launcher->flags, pcmk__rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->priv->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
node_type, this_node->priv->id);
this_node->details->online = FALSE;
} else if ((rsc->priv->orig_role == pcmk_role_stopped)
|| ((launcher != NULL)
&& (launcher->priv->orig_role == pcmk_role_stopped))) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
node_type, this_node->priv->id);
this_node->details->online = FALSE;
pcmk__clear_node_flags(this_node, pcmk__node_remote_reset);
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->priv->id);
this_node->details->online = FALSE;
pcmk__set_node_flags(this_node, pcmk__node_remote_reset);
} else {
crm_trace("%s node %s is %s",
node_type, this_node->priv->id,
this_node->details->online? "ONLINE" : "OFFLINE");
}
}
static void
determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler)
{
gboolean online = FALSE;
const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
if (pe__shutdown_requested(this_node)) {
this_node->details->shutdown = TRUE;
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
pcmk__set_node_flags(this_node, pcmk__node_expected_up);
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
online = determine_online_status_no_fencing(scheduler, node_state,
this_node);
} else {
online = determine_online_status_fencing(scheduler, node_state,
this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->assign->score = -PCMK_SCORE_INFINITY;
}
if (this_node->details->unclean) {
pcmk__sched_warn(scheduler, "%s is unclean",
pcmk__node_name(this_node));
} else if (!this_node->details->online) {
crm_trace("%s is offline", pcmk__node_name(this_node));
} else if (this_node->details->shutdown) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
} else if (this_node->details->pending) {
crm_info("%s is pending", pcmk__node_name(this_node));
} else if (pcmk_is_set(this_node->priv->flags, pcmk__node_standby)) {
crm_info("%s is in standby", pcmk__node_name(this_node));
} else if (this_node->details->maintenance) {
crm_info("%s is in maintenance", pcmk__node_name(this_node));
} else {
crm_info("%s is online", pcmk__node_name(this_node));
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
pcmk__assert(end != NULL);
basename = strndup(last_rsc_id, end - last_rsc_id + 1);
pcmk__assert(basename != NULL);
return basename;
}
/*!
* \internal
* \brief Get the name of the first instance of a cloned resource
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name plus :0
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_zero(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
size_t base_name_len = end - last_rsc_id + 1;
char *zero = NULL;
pcmk__assert(end != NULL);
zero = pcmk__assert_alloc(base_name_len + 3, sizeof(char));
memcpy(zero, last_rsc_id, base_name_len);
zero[base_name_len] = ':';
zero[base_name_len + 1] = '0';
return zero;
}
static pcmk_resource_t *
create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE);
pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none);
pcmk__xe_set(xml_rsc, PCMK_XA_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Removed resource");
if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pcmk_node_t *node;
crm_debug("Detected removed remote node %s", rsc_id);
node = pcmk_find_node(scheduler, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, 0,
scheduler);
}
link_rsc2remotenode(scheduler, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to removed "
"connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (pcmk__xe_get(rsc_entry, PCMK__META_CONTAINER)) {
// This removed resource needs to be mapped to a launcher
crm_trace("Launched resource %s was removed from the configuration",
rsc_id);
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_launched);
}
pcmk__set_rsc_flags(rsc, pcmk__rsc_removed);
scheduler->priv->resources = g_list_append(scheduler->priv->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create "removed" instance for anonymous clone resource history
*
* \param[in,out] parent Clone resource that instance will be added to
* \param[in] rsc_id Instance's resource ID
* \param[in] node Where instance is active (for logging only)
* \param[in,out] scheduler Scheduler data
*
* \return Newly created "removed" instance of \p parent
*/
static pcmk_resource_t *
create_anonymous_removed_instance(pcmk_resource_t *parent, const char *rsc_id,
const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *top = pe__create_clone_child(parent, scheduler);
pcmk_resource_t *instance = NULL;
// find_rsc() because we might be a cloned group
instance = top->priv->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pcmk__rsc_debug(parent, "Created \"removed\" instance %s for %s: %s on %s",
top->id, parent->id, rsc_id, pcmk__node_name(node));
return instance;
}
/*!
* \internal
* \brief Check a node for an instance of an anonymous clone
*
* Return a child instance of the specified anonymous clone, in order of
* preference: (1) the instance running on the specified node, if any;
* (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX
* instances); (3) a newly created "removed" instance (that is,
* \c PCMK_META_CLONE_MAX instances are already active).
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node on which to check for instance
* \param[in,out] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (no instance)
*/
static pcmk_resource_t *
find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
pcmk_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
pcmk__assert(pcmk__is_anonymous_clone(parent));
// Check for active (or partially active, for cloned groups) instance
pcmk__rsc_trace(parent, "Looking for %s on %s in %s",
rsc_id, pcmk__node_name(node), parent->id);
for (rIter = parent->priv->children;
(rIter != NULL) && (rsc == NULL); rIter = rIter->next) {
GList *locations = NULL;
pcmk_resource_t *child = rIter->data;
/* Check whether this instance is already known to be active or pending
* anywhere, at this stage of unpacking. Because this function is called
* for a resource before the resource's individual operation history
* entries are unpacked, locations will generally not contain the
* desired node.
*
* However, there are three exceptions:
* (1) when child is a cloned group and we have already unpacked the
* history of another member of the group on the same node;
* (2) when we've already unpacked the history of another numbered
* instance on the same node (which can happen if
* PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and
* (3) when we re-run calculations on the same scheduler data as part of
* a simulation.
*/
child->priv->fns->location(child, &locations, pcmk__rsc_node_current
|pcmk__rsc_node_pending);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (pcmk__same_node((pcmk_node_t *) locations->data, node)) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry represents a removed instance, rsc will
* be NULL.
*/
rsc = parent->priv->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
if (rsc) {
/* If there are multiple instance history entries for an
* anonymous clone in a single node's history (which can
* happen if PCMK_META_GLOBALLY_UNIQUE is switched from true
* to false), we want to consider the instances beyond the
* first as removed, even if there are inactive instance
* numbers available.
*/
if (rsc->priv->active_nodes != NULL) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple \"removed\" instance histories on "
"%s",
parent->id, pcmk__node_name(node));
skip_inactive = TRUE;
rsc = NULL;
} else {
pcmk__rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pcmk__rsc_blocked)) {
// Remember one inactive instance in case we don't find active
inactive_instance =
parent->priv->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
/* ... but don't use it if it was already associated with a
* pending action on another node
*/
if (inactive_instance != NULL) {
const pcmk_node_t *pending_node = NULL;
pending_node = inactive_instance->priv->pending_node;
if ((pending_node != NULL)
&& !pcmk__same_node(pending_node, node)) {
inactive_instance = NULL;
}
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pcmk__rsc_trace(parent, "Resource %s, empty slot",
inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or
* PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we
* don't want to consume a valid instance number for unclean nodes. Such
* instances may appear to be active according to the history, but should be
* considered inactive, so we can start an instance elsewhere. Treat such
* instances as removed.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pcmk__is_guest_or_bundle_node(node)
&& !pe__is_universal_clone(parent, scheduler)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_removed_instance(parent, rsc_id, node,
scheduler);
pcmk__rsc_trace(parent, "Resource %s, removed", rsc->id);
}
return rsc;
}
static pcmk_resource_t *
unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
const char *rsc_id)
{
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(scheduler->priv->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0,
* we create a single :0 "removed" instance to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pcmk_resource_t *clone0 = pe_find_resource(scheduler->priv->resources,
clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (removed)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->priv->variant > pcmk__rsc_variant_primitive) {
crm_trace("Resource history for %s is considered removed "
"because it is no longer primitive", rsc_id);
return NULL;
} else {
parent = uber_parent(rsc);
}
if (pcmk__is_anonymous_clone(parent)) {
if (pcmk__is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->priv->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(scheduler, node, parent, base);
free(base);
pcmk__assert(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none)
&& !pcmk__str_eq(rsc_id, rsc->priv->history_id, pcmk__str_none)) {
const bool removed = pcmk_is_set(rsc->flags, pcmk__rsc_removed);
pcmk__str_update(&(rsc->priv->history_id), rsc_id);
pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pcmk__node_name(node), rsc->id,
(removed? " (removed)" : ""));
}
return rsc;
}
static pcmk_resource_t *
process_removed_resource(const xmlNode *rsc_entry, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = pcmk__xe_get(rsc_entry, PCMK_XA_ID);
crm_debug("Detected removed resource %s on %s",
rsc_id, pcmk__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) {
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added \"removed\" resource %s", rsc->id);
resource_location(rsc, NULL, -PCMK_SCORE_INFINITY,
"__removed_do_not_run__", scheduler);
}
return rsc;
}
static void
process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node,
enum pcmk__on_fail on_fail)
{
pcmk_node_t *tmpnode = NULL;
char *reason = NULL;
enum pcmk__on_fail save_on_fail = pcmk__on_fail_ignore;
pcmk_scheduler_t *scheduler = NULL;
bool known_active = false;
pcmk__assert(rsc != NULL);
scheduler = rsc->priv->scheduler;
known_active = (rsc->priv->orig_role > pcmk_role_stopped);
pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, pcmk_role_text(rsc->priv->orig_role),
pcmk__node_name(node), pcmk__on_fail_text(on_fail));
/* process current state */
if (rsc->priv->orig_role != pcmk_role_unknown) {
pcmk_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->priv->probed_nodes,
node->priv->id) == NULL) {
pcmk_node_t *n = pe__copy_node(node);
pcmk__rsc_trace(rsc, "%s (%s in history) known on %s",
rsc->id,
pcmk__s(rsc->priv->history_id, "the same"),
pcmk__node_name(n));
g_hash_table_insert(iter->priv->probed_nodes,
(gpointer) n->priv->id, n);
}
if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) {
break;
}
iter = iter->priv->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if (known_active && !node->details->online && !node->details->maintenance
&& pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
gboolean should_fence = FALSE;
/* If this is a guest node, fence it (regardless of whether fencing is
* enabled, because guest node fencing is done by recovery of the
* container resource rather than by the fencer). Mark the resource
* we're processing as failed. When the guest comes back up, its
* operation history in the CIB will be cleared, freeing the affected
* resource to run again once we are sure we know its state.
*/
if (pcmk__is_guest_or_bundle_node(node)) {
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
should_fence = TRUE;
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
if (pcmk__is_remote_node(node)
&& (node->priv->remote != NULL)
&& !pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_failed)) {
/* Setting unseen means that fencing of the remote node will
* occur only if the connection resource is not going to start
* somewhere. This allows connection resources on a failed
* cluster node to move to another node without requiring the
* remote nodes to be fenced as well.
*/
pcmk__clear_node_flags(node, pcmk__node_seen);
reason = crm_strdup_printf("%s is active there (fencing will be"
" revoked if remote connection can "
"be re-established elsewhere)",
rsc->id);
}
should_fence = TRUE;
}
if (should_fence) {
if (reason == NULL) {
reason = crm_strdup_printf("%s is thought to be active there", rsc->id);
}
pe_fence_node(scheduler, node, reason, FALSE);
}
free(reason);
}
/* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */
save_on_fail = on_fail;
if (node->details->unclean) {
/* No extra processing needed
* Also allows resources to be started again after a node is shot
*/
on_fail = pcmk__on_fail_ignore;
}
switch (on_fail) {
case pcmk__on_fail_ignore:
/* nothing to do */
break;
case pcmk__on_fail_demote:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
demote_action(rsc, node, FALSE);
break;
case pcmk__on_fail_fence_node:
/* treat it as if it is still running
* but also mark the node as unclean
*/
reason = crm_strdup_printf("%s failed there", rsc->id);
pe_fence_node(scheduler, node, reason, FALSE);
free(reason);
break;
case pcmk__on_fail_standby_node:
pcmk__set_node_flags(node,
pcmk__node_standby|pcmk__node_fail_standby);
break;
case pcmk__on_fail_block:
/* is_managed == FALSE will prevent any
* actions being sent for the resource
*/
pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked);
break;
case pcmk__on_fail_ban:
/* make sure it comes up somewhere else
* or not at all
*/
resource_location(rsc, node, -PCMK_SCORE_INFINITY,
"__action_migration_auto__", scheduler);
break;
case pcmk__on_fail_stop:
pe__set_next_role(rsc, pcmk_role_stopped,
PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP);
break;
case pcmk__on_fail_restart:
if (known_active) {
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_restart_container:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if ((rsc->priv->launcher != NULL) && pcmk__is_bundled(rsc)) {
/* A bundle's remote connection can run on a different node than
* the bundle's container. We don't necessarily know where the
* container is running yet, so remember it and add a stop
* action for it later.
*/
scheduler->priv->stop_needed =
g_list_prepend(scheduler->priv->stop_needed,
rsc->priv->launcher);
} else if (rsc->priv->launcher != NULL) {
stop_action(rsc->priv->launcher, node, FALSE);
} else if (known_active) {
stop_action(rsc, node, FALSE);
}
break;
case pcmk__on_fail_reset_remote:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
tmpnode = NULL;
if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
}
if (pcmk__is_remote_node(tmpnode)
&& !pcmk_is_set(tmpnode->priv->flags,
pcmk__node_remote_fenced)) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(scheduler, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (known_active) {
stop_action(rsc, node, FALSE);
}
/* if reconnect delay is in use, prevent the connection from exiting the
* "STOPPED" role until the failure is cleared by the delay timeout. */
if (rsc->priv->remote_reconnect_ms > 0U) {
pe__set_next_role(rsc, pcmk_role_stopped, "remote reset");
}
break;
}
/* Ensure a remote connection failure forces an unclean Pacemaker Remote
* node to be fenced. By marking the node as seen, the failure will result
* in a fencing operation regardless if we're going to attempt to reconnect
* in this transition.
*/
if (pcmk_all_flags_set(rsc->flags,
pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) {
tmpnode = pcmk_find_node(scheduler, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
pcmk__set_node_flags(tmpnode, pcmk__node_seen);
}
}
if (known_active) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) {
crm_notice("Removed resource %s is active on %s and will be "
"stopped when possible",
rsc->id, pcmk__node_name(node));
} else {
crm_notice("Removed resource %s must be stopped manually on %s "
"because " PCMK_OPT_STOP_REMOVED_RESOURCES
" is set to false", rsc->id, pcmk__node_name(node));
}
}
native_add_running(rsc, node, scheduler,
(save_on_fail != pcmk__on_fail_ignore));
switch (on_fail) {
case pcmk__on_fail_ignore:
break;
case pcmk__on_fail_demote:
case pcmk__on_fail_block:
pcmk__set_rsc_flags(rsc, pcmk__rsc_failed);
break;
default:
pcmk__set_rsc_flags(rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
break;
}
} else if ((rsc->priv->history_id != NULL)
&& (strchr(rsc->priv->history_id, ':') != NULL)) {
/* @COMPAT This is for older (<1.1.8) status sections that included
* instance numbers, otherwise stopped instances are considered removed.
*
* @TODO We should be able to drop this, but some old regression tests
* will need to be updated. Double-check that this is not still needed
* for unique clones (which may have been later converted to anonymous).
*/
pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)",
rsc->priv->history_id, rsc->id);
free(rsc->priv->history_id);
rsc->priv->history_id = NULL;
} else {
GList *possible_matches = pe__resource_actions(rsc, node,
PCMK_ACTION_STOP, FALSE);
GList *gIter = possible_matches;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_action_t *stop = (pcmk_action_t *) gIter->data;
pcmk__set_action_flags(stop, pcmk__action_optional);
}
g_list_free(possible_matches);
}
/* A successful stop after migrate_to on the migration source doesn't make
* the partially migrated resource stopped on the migration target.
*/
if ((rsc->priv->orig_role == pcmk_role_stopped)
&& (rsc->priv->active_nodes != NULL)
&& (rsc->priv->partial_migration_target != NULL)
&& pcmk__same_node(rsc->priv->partial_migration_source, node)) {
rsc->priv->orig_role = pcmk_role_started;
}
}
/* create active recurring operations as optional */
static void
process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc,
int start_index, int stop_index,
GList *sorted_op_list, pcmk_scheduler_t *scheduler)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
pcmk__assert(rsc != NULL);
pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d",
rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = pcmk__xe_id(rsc_op);
counter++;
if (node->details->online == FALSE) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline",
rsc->id, pcmk__node_name(node));
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pcmk__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pcmk__node_name(node), counter);
continue;
}
pcmk__xe_get_guint(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if (interval_ms == 0) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring",
id, pcmk__node_name(node));
continue;
}
status = pcmk__xe_get(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: status",
id, pcmk__node_name(node));
continue;
}
task = pcmk__xe_get(rsc_op, PCMK_XA_OPERATION);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = pcmk__xe_get(rsc_op, PCMK_XA_OPERATION);
status = pcmk__xe_get(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei)
&& pcmk__str_eq(status, "0", pcmk__str_casei)) {
*stop_index = counter;
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
*start_index = counter;
} else if ((implied_monitor_start <= *stop_index)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
const char *rc = pcmk__xe_get(rsc_op, PCMK__XA_RC_CODE);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
time_t sched_time = 0;
guint shutdown_lock_ms = scheduler->priv->shutdown_lock_ms;
pcmk__xe_get_time(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time);
if (lock_time == 0) {
return;
}
sched_time = pcmk__scheduler_epoch_time(scheduler);
if ((shutdown_lock_ms > 0U)
&& (sched_time > (lock_time + pcmk__timeout_ms2s(shutdown_lock_ms)))) {
pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pcmk__node_name(node));
pe__clear_resource_history(rsc, node);
} else {
rsc->priv->lock_node = node;
rsc->priv->lock_time = lock_time;
}
}
/*!
* \internal
* \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked
* \param[in,out] scheduler Scheduler data
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pcmk_resource_t *
unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource,
pcmk_scheduler_t *scheduler)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = pcmk_role_unknown;
const char *rsc_id = pcmk__xe_id(lrm_resource);
pcmk_resource_t *rsc = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
xmlNode *rsc_op = NULL;
xmlNode *last_failure = NULL;
enum pcmk__on_fail on_fail = pcmk__on_fail_ignore;
enum rsc_role_e saved_role = pcmk_role_unknown;
if (rsc_id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE
" entry: No " PCMK_XA_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s",
rsc_id, pcmk__node_name(node));
/* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort
* them
*/
for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(scheduler, node, rsc_id);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_removed_resource(lrm_resource, node, scheduler);
}
}
pcmk__assert(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, scheduler);
}
/* process operations */
saved_role = rsc->priv->orig_role;
rsc->priv->orig_role = pcmk_role_unknown;
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail);
}
/* create active recurring operations as optional */
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
process_recurring(node, rsc, start_index, stop_index, sorted_op_list,
scheduler);
/* no need to free the contents */
g_list_free(sorted_op_list);
process_rsc_state(rsc, node, on_fail);
if (get_target_role(rsc, &req_role)) {
if ((rsc->priv->next_role == pcmk_role_unknown)
|| (req_role < rsc->priv->next_role)) {
pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE);
} else if (req_role > rsc->priv->next_role) {
pcmk__rsc_info(rsc,
"%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, pcmk_role_text(rsc->priv->next_role),
pcmk_role_text(req_role));
}
}
if (saved_role > rsc->priv->orig_role) {
rsc->priv->orig_role = saved_role;
}
return rsc;
}
static void
handle_removed_launched_resources(const xmlNode *lrm_rsc_list,
pcmk_scheduler_t *scheduler)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
pcmk_resource_t *rsc;
pcmk_resource_t *launcher = NULL;
const char *rsc_id;
const char *launcher_id = NULL;
launcher_id = pcmk__xe_get(rsc_entry, PCMK__META_CONTAINER);
rsc_id = pcmk__xe_get(rsc_entry, PCMK_XA_ID);
if ((launcher_id == NULL) || (rsc_id == NULL)) {
continue;
}
launcher = pe_find_resource(scheduler->priv->resources, launcher_id);
if (launcher == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->priv->resources, rsc_id);
if ((rsc == NULL) || (rsc->priv->launcher != NULL)
|| !pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
continue;
}
pcmk__rsc_trace(rsc, "Mapped launcher of removed resource %s to %s",
rsc->id, launcher_id);
rsc->priv->launcher = launcher;
launcher->priv->launched = g_list_append(launcher->priv->launched,
rsc);
}
}
/*!
* \internal
* \brief Unpack one node's lrm status section
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] xml CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml,
pcmk_scheduler_t *scheduler)
{
bool found_removed_launched_resource = false;
// Drill down to PCMK__XE_LRM_RESOURCES section
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL);
if (xml == NULL) {
return;
}
xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL);
if (xml == NULL) {
return;
}
// Unpack each PCMK__XE_LRM_RESOURCE entry
for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml,
PCMK__XE_LRM_RESOURCE,
NULL, NULL);
rsc_entry != NULL;
rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) {
pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) {
found_removed_launched_resource = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* removed launched resources to their launchers.
*/
if (found_removed_launched_resource) {
handle_removed_launched_resources(xml, scheduler);
}
}
static void
set_active(pcmk_resource_t *rsc)
{
const pcmk_resource_t *top = pe__const_top_resource(rsc, false);
if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) {
rsc->priv->orig_role = pcmk_role_unpromoted;
} else {
rsc->priv->orig_role = pcmk_role_started;
}
}
static void
set_node_score(gpointer key, gpointer value, gpointer user_data)
{
pcmk_node_t *node = value;
int *score = user_data;
node->assign->score = *score;
}
#define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE
#define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \
"/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
int target_rc, pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL),
return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']"
SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'",
NULL);
/* Need to check against transition_magic too? */
if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = pcmk__xpath_find_one(scheduler->input->doc, xpath->str, LOG_DEBUG);
g_string_free(xpath, TRUE);
if (xml && target_rc >= 0) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_EXEC_ERROR;
pcmk__xe_get_int(xml, PCMK__XA_RC_CODE, &rc);
pcmk__xe_get_int(xml, PCMK__XA_OP_STATUS, &status);
if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) {
return NULL;
}
}
return xml;
}
static xmlNode *
find_lrm_resource(const char *rsc_id, const char *node_name,
pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']",
NULL);
xml = pcmk__xpath_find_one(scheduler->input->doc, xpath->str, LOG_DEBUG);
g_string_free(xpath, TRUE);
return xml;
}
/*!
* \internal
* \brief Check whether a resource has no completed action history on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node_name Node to check
*
* \return true if \p rsc_id is unknown on \p node_name, otherwise false
*/
static bool
unknown_on_node(pcmk_resource_t *rsc, const char *node_name)
{
bool result = false;
xmlXPathObject *search;
char *xpath = NULL;
xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
SUB_XPATH_LRM_RSC_OP
"[@" PCMK__XA_RC_CODE "!='%d']",
node_name, rsc->id, PCMK_OCF_UNKNOWN);
search = pcmk__xpath_search(rsc->priv->scheduler->input->doc, xpath);
result = (pcmk__xpath_num_results(search) == 0);
xmlXPathFreeObject(search);
free(xpath);
return result;
}
/*!
* \internal
* \brief Check whether a probe/monitor indicating the resource was not running
* on a node happened after some event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that monitor is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a monitor happened after event, false otherwise
*/
static bool
monitor_not_running_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
/* Any probe/monitor operation on the node indicating it was not running
* there
*/
xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name,
NULL, PCMK_OCF_NOT_RUNNING, scheduler);
return (monitor != NULL) && (pe__is_newer_op(monitor, xml_op) > 0);
}
/*!
* \internal
* \brief Check whether any non-monitor operation on a node happened after some
* event
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] xml_op Event that non-monitor is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
non_monitor_after(const char *rsc_id, const char *node_name,
const xmlNode *xml_op, pcmk_scheduler_t *scheduler)
{
xmlNode *lrm_resource = NULL;
lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler);
if (lrm_resource == NULL) {
return false;
}
for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP,
NULL, NULL);
op != NULL; op = pcmk__xe_next(op, PCMK__XE_LRM_RSC_OP)) {
const char * task = NULL;
if (op == xml_op) {
continue;
}
task = pcmk__xe_get(op, PCMK_XA_OPERATION);
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
NULL)
&& pe__is_newer_op(op, xml_op) > 0) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether the resource has newer state on a node after a migration
* attempt
*
* \param[in] rsc_id Resource being checked
* \param[in] node_name Node being checked
* \param[in] migrate_to Any migrate_to event that is being compared to
* \param[in] migrate_from Any migrate_from event that is being compared to
* \param[in,out] scheduler Scheduler data
*
* \return true if such a operation happened after event, false otherwise
*/
static bool
newer_state_after_migrate(const char *rsc_id, const char *node_name,
const xmlNode *migrate_to,
const xmlNode *migrate_from,
pcmk_scheduler_t *scheduler)
{
const xmlNode *xml_op = (migrate_from != NULL)? migrate_from : migrate_to;
const char *source = pcmk__xe_get(xml_op, PCMK__META_MIGRATE_SOURCE);
/* It's preferred to compare to the migrate event on the same node if
* existing, since call ids are more reliable.
*/
if ((xml_op != migrate_to) && (migrate_to != NULL)
&& pcmk__str_eq(node_name, source, pcmk__str_casei)) {
xml_op = migrate_to;
}
/* If there's any newer non-monitor operation on the node, or any newer
* probe/monitor operation on the node indicating it was not running there,
* the migration events potentially no longer matter for the node.
*/
return non_monitor_after(rsc_id, node_name, xml_op, scheduler)
|| monitor_not_running_after(rsc_id, node_name, xml_op, scheduler);
}
/*!
* \internal
* \brief Parse migration source and target node names from history entry
*
* \param[in] entry Resource history entry for a migration action
* \param[in] source_node If not NULL, source must match this node
* \param[in] target_node If not NULL, target must match this node
* \param[out] source_name Where to store migration source node name
* \param[out] target_name Where to store migration target node name
*
* \return Standard Pacemaker return code
*/
static int
get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node,
const pcmk_node_t *target_node,
const char **source_name, const char **target_name)
{
*source_name = pcmk__xe_get(entry, PCMK__META_MIGRATE_SOURCE);
*target_name = pcmk__xe_get(entry, PCMK__META_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
PCMK__META_MIGRATE_SOURCE " and "
PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry));
return pcmk_rc_unpack_error;
}
if ((source_node != NULL)
&& !pcmk__str_eq(*source_name, source_node->priv->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_SOURCE "='%s' does not match %s",
pcmk__xe_id(entry), *source_name,
pcmk__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->priv->name,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_TARGET "='%s' does not match %s",
pcmk__xe_id(entry), *target_name,
pcmk__node_name(target_node));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*
* \internal
* \brief Add a migration source to a resource's list of dangling migrations
*
* If the migrate_to and migrate_from actions in a live migration both
* succeeded, but there is no stop on the source, the migration is considered
* "dangling." Add the source to the resource's dangling migration list, which
* will be used to schedule a stop on the source without affecting the target.
*
* \param[in,out] rsc Resource involved in migration
* \param[in] node Migration source
*/
static void
add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s",
rsc->id, pcmk__node_name(node));
rsc->priv->orig_role = pcmk_role_stopped;
rsc->priv->dangling_migration_sources =
g_list_prepend(rsc->priv->dangling_migration_sources,
(gpointer) node);
}
/*!
* \internal
* \brief Update resource role etc. after a successful migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_success(struct action_history *history)
{
/* A complete migration sequence is:
* 1. migrate_to on source node (which succeeded if we get to this function)
* 2. migrate_from on target node
* 3. stop on source node
*
* If no migrate_from has happened, the migration is considered to be
* "partial". If the migrate_from succeeded but no stop has happened, the
* migration is considered to be "dangling".
*
* If a successful migrate_to and stop have happened on the source node, we
* still need to check for a partial migration, due to scenarios (easier to
* produce with batch-limit=1) like:
*
* - A resource is migrating from node1 to node2, and a migrate_to is
* initiated for it on node1.
*
* - node2 goes into standby mode while the migrate_to is pending, which
* aborts the transition.
*
* - Upon completion of the migrate_to, a new transition schedules a stop
* on both nodes and a start on node1.
*
* - If the new transition is aborted for any reason while the resource is
* stopping on node1, the transition after that stop completes will see
* the migrate_to and stop on the source, but it's still a partial
* migration, and the resource must be stopped on node2 because it is
* potentially active there due to the migrate_to.
*
* We also need to take into account that either node's history may be
* cleared at any point in the migration process.
*/
int from_rc = PCMK_OCF_OK;
int from_status = PCMK_EXEC_PENDING;
pcmk_node_t *target_node = NULL;
xmlNode *migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
bool source_newer_op = false;
bool target_newer_state = false;
bool active_on_target = false;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
// Check for newer state on the source
source_newer_op = non_monitor_after(history->rsc->id, source, history->xml,
scheduler);
// Check for a migrate_from action from this source on the target
migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM,
target, source, -1, scheduler);
if (migrate_from != NULL) {
if (source_newer_op) {
/* There's a newer non-monitor operation on the source and a
* migrate_from on the target, so this migrate_to is irrelevant to
* the resource's state.
*/
return;
}
pcmk__xe_get_int(migrate_from, PCMK__XA_RC_CODE, &from_rc);
pcmk__xe_get_int(migrate_from, PCMK__XA_OP_STATUS, &from_status);
}
/* If the resource has newer state on both the source and target after the
* migration events, this migrate_to is irrelevant to the resource's state.
*/
target_newer_state = newer_state_after_migrate(history->rsc->id, target,
history->xml, migrate_from,
scheduler);
if (source_newer_op && target_newer_state) {
return;
}
/* Check for dangling migration (migrate_from succeeded but stop not done).
* We know there's no stop because we already returned if the target has a
* migrate_from and the source has any newer non-monitor operation.
*/
if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) {
add_dangling_migration(history->rsc, history->node);
return;
}
/* Without newer state, this migrate_to implies the resource is active.
* (Clones are not allowed to migrate, so role can't be promoted.)
*/
history->rsc->priv->orig_role = pcmk_role_started;
target_node = pcmk_find_node(scheduler, target);
active_on_target = !target_newer_state && (target_node != NULL)
&& target_node->details->online;
if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target
if (active_on_target) {
native_add_running(history->rsc, target_node, scheduler, TRUE);
} else {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
return;
}
// The migrate_from is pending, complete but erased, or to be scheduled
/* If there is no history at all for the resource on an online target, then
* it was likely cleaned. Just return, and we'll schedule a probe. Once we
* have the probe result, it will be reflected in target_newer_state.
*/
if ((target_node != NULL) && target_node->details->online
&& unknown_on_node(history->rsc, target)) {
return;
}
if (active_on_target) {
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
native_add_running(history->rsc, target_node, scheduler, FALSE);
if ((source_node != NULL) && source_node->details->online) {
/* This is a partial migration: the migrate_to completed
* successfully on the source, but the migrate_from has not
* completed. Remember the source and target; if the newly
* chosen target remains the same when we schedule actions
* later, we may continue with the migration.
*/
history->rsc->priv->partial_migration_target = target_node;
history->rsc->priv->partial_migration_source = source_node;
}
} else if (!source_newer_op) {
// Mark resource as failed, require recovery, and prevent migration
pcmk__set_rsc_flags(history->rsc,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_to action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_to_failure(struct action_history *history)
{
xmlNode *target_migrate_from = NULL;
const char *source = NULL;
const char *target = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, history->node, NULL, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->priv->orig_role = pcmk_role_started;
// Check for migrate_from on the target
target_migrate_from = find_lrm_op(history->rsc->id,
PCMK_ACTION_MIGRATE_FROM, target, source,
PCMK_OCF_OK, scheduler);
if (/* If the resource state is unknown on the target, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, target)
/* If the resource has newer state on the target after the migration
* events, this migrate_to no longer matters for the target.
*/
&& !newer_state_after_migrate(history->rsc->id, target, history->xml,
target_migrate_from, scheduler)) {
/* The resource has no newer state on the target, so assume it's still
* active there.
* (if it is up).
*/
pcmk_node_t *target_node = pcmk_find_node(scheduler, target);
if (target_node && target_node->details->online) {
native_add_running(history->rsc, target_node, scheduler, FALSE);
}
} else if (!non_monitor_after(history->rsc->id, source, history->xml,
scheduler)) {
/* We know the resource has newer state on the target, but this
* migrate_to still matters for the source as long as there's no newer
* non-monitor operation there.
*/
// Mark node as having dangling migration so we can force a stop later
history->rsc->priv->dangling_migration_sources =
g_list_prepend(history->rsc->priv->dangling_migration_sources,
(gpointer) history->node);
}
}
/*!
* \internal
* \brief Update resource role etc. after a failed migrate_from action
*
* \param[in,out] history Parsed action result history
*/
static void
unpack_migrate_from_failure(struct action_history *history)
{
xmlNode *source_migrate_to = NULL;
const char *source = NULL;
const char *target = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
// Get source and target node names from XML
if (get_migration_node_names(history->xml, NULL, history->node, &source,
&target) != pcmk_rc_ok) {
return;
}
/* If a migration failed, we have to assume the resource is active. Clones
* are not allowed to migrate, so role can't be promoted.
*/
history->rsc->priv->orig_role = pcmk_role_started;
// Check for a migrate_to on the source
source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO,
source, target, PCMK_OCF_OK, scheduler);
if (/* If the resource state is unknown on the source, it will likely be
* probed there.
* Don't just consider it running there. We will get back here anyway in
* case the probe detects it's running there.
*/
!unknown_on_node(history->rsc, source)
/* If the resource has newer state on the source after the migration
* events, this migrate_from no longer matters for the source.
*/
&& !newer_state_after_migrate(history->rsc->id, source,
source_migrate_to, history->xml,
scheduler)) {
/* The resource has no newer state on the source, so assume it's still
* active there (if it is up).
*/
pcmk_node_t *source_node = pcmk_find_node(scheduler, source);
if (source_node && source_node->details->online) {
native_add_running(history->rsc, source_node, scheduler, TRUE);
}
}
}
/*!
* \internal
* \brief Add an action to cluster's list of failed actions
*
* \param[in,out] history Parsed action result history
*/
static void
record_failed_op(struct action_history *history)
{
const pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
if (!(history->node->details->online)) {
return;
}
for (const xmlNode *xIter = scheduler->priv->failed->children;
xIter != NULL; xIter = xIter->next) {
const char *key = pcmk__xe_history_key(xIter);
const char *uname = pcmk__xe_get(xIter, PCMK_XA_UNAME);
if (pcmk__str_eq(history->key, key, pcmk__str_none)
&& pcmk__str_eq(uname, history->node->priv->name,
pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s",
history->key, pcmk__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pcmk__node_name(history->node));
pcmk__xe_set(history->xml, PCMK_XA_UNAME, history->node->priv->name);
pcmk__xe_set(history->xml, PCMK__XA_RSC_ID, history->rsc->id);
pcmk__xml_copy(scheduler->priv->failed, history->xml);
}
static char *
last_change_str(const xmlNode *xml_op)
{
time_t when;
char *result = NULL;
if (pcmk__xe_get_time(xml_op, PCMK_XA_LAST_RC_CHANGE,
&when) == pcmk_rc_ok) {
char *when_s = pcmk__epoch2str(&when, 0);
const char *p = strchr(when_s, ' ');
// Skip day of week to make message shorter
if ((p != NULL) && (*(++p) != '\0')) {
result = pcmk__str_copy(p);
}
free(when_s);
}
if (result == NULL) {
result = pcmk__str_copy("unknown_time");
}
return result;
}
/*!
* \internal
* \brief Ban a resource (or its clone if an anonymous instance) from all nodes
*
* \param[in,out] rsc Resource to ban
*/
static void
ban_from_all_nodes(pcmk_resource_t *rsc)
{
int score = -PCMK_SCORE_INFINITY;
const pcmk_scheduler_t *scheduler = rsc->priv->scheduler;
if (rsc->priv->parent != NULL) {
pcmk_resource_t *parent = uber_parent(rsc);
if (pcmk__is_anonymous_clone(parent)) {
/* For anonymous clones, if an operation with
* PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the
* entire clone must stop.
*/
rsc = parent;
}
}
// Ban the resource from all nodes
crm_notice("%s will not be started under current conditions", rsc->id);
if (rsc->priv->allowed_nodes != NULL) {
g_hash_table_destroy(rsc->priv->allowed_nodes);
}
rsc->priv->allowed_nodes = pe__node_list2table(scheduler->nodes);
g_hash_table_foreach(rsc->priv->allowed_nodes, set_node_score, &score);
}
/*!
* \internal
* \brief Get configured failure handling and role after failure for an action
*
* \param[in,out] history Unpacked action history entry
* \param[out] on_fail Where to set configured failure handling
* \param[out] fail_role Where to set to role after failure
*/
static void
unpack_failure_handling(struct action_history *history,
enum pcmk__on_fail *on_fail,
enum rsc_role_e *fail_role)
{
xmlNode *config = pcmk__find_action_config(history->rsc, history->task,
history->interval_ms, true);
GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node,
history->task,
history->interval_ms, config);
const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL);
*on_fail = pcmk__parse_on_fail(history->rsc, history->task,
history->interval_ms, on_fail_str);
*fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail,
meta);
g_hash_table_destroy(meta);
}
/*!
* \internal
* \brief Update resource role, failure handling, etc., after a failed action
*
* \param[in,out] history Parsed action result history
* \param[in] config_on_fail Action failure handling from configuration
* \param[in] fail_role Resource's role after failure of this action
* \param[out] last_failure This will be set to the history XML
* \param[in,out] on_fail Actual handling of action result
*/
static void
unpack_rsc_op_failure(struct action_history *history,
enum pcmk__on_fail config_on_fail,
enum rsc_role_e fail_role, xmlNode **last_failure,
enum pcmk__on_fail *on_fail)
{
bool is_probe = false;
char *last_change_s = NULL;
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
*last_failure = history->xml;
is_probe = pcmk_xe_is_probe(history->xml);
last_change_s = last_change_str(history->xml);
if (!pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)
&& (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
crm_trace("Unexpected result (%s%s%s) was recorded for "
"%s of %s on %s at %s " QB_XS " exit-status=%d id=%s",
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
} else {
pcmk__sched_warn(scheduler,
"Unexpected result (%s%s%s) was recorded for %s of "
"%s on %s at %s " QB_XS " exit-status=%d id=%s",
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
if (is_probe && (history->exit_status != PCMK_OCF_OK)
&& (history->exit_status != PCMK_OCF_NOT_RUNNING)
&& (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) {
/* A failed (not just unexpected) probe result could mean the user
* didn't know resources will be probed even where they can't run.
*/
crm_notice("If it is not possible for %s to run on %s, see "
"the " PCMK_XA_RESOURCE_DISCOVERY " option for location "
"constraints",
history->rsc->id, pcmk__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (*on_fail < config_on_fail) {
pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s",
pcmk__on_fail_text(*on_fail),
pcmk__on_fail_text(config_on_fail), history->key);
*on_fail = config_on_fail;
}
if (strcmp(history->task, PCMK_ACTION_STOP) == 0) {
resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY,
"__stop_fail__", scheduler);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) {
unpack_migrate_to_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) {
unpack_migrate_from_failure(history);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->priv->orig_role = pcmk_role_promoted;
} else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) {
if (config_on_fail == pcmk__on_fail_block) {
history->rsc->priv->orig_role = pcmk_role_promoted;
pe__set_next_role(history->rsc, pcmk_role_stopped,
"demote with " PCMK_META_ON_FAIL "=block");
} else if (history->exit_status == PCMK_OCF_NOT_RUNNING) {
history->rsc->priv->orig_role = pcmk_role_stopped;
} else {
/* Staying in the promoted role would put the scheduler and
* controller into a loop. Setting the role to unpromoted is not
* dangerous because the resource will be stopped as part of
* recovery, and any promotion will be ordered after that stop.
*/
history->rsc->priv->orig_role = pcmk_role_unpromoted;
}
}
if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) {
/* leave stopped */
pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id);
history->rsc->priv->orig_role = pcmk_role_stopped;
} else if (history->rsc->priv->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id);
set_active(history->rsc);
}
pcmk__rsc_trace(history->rsc,
"Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s",
history->rsc->id,
pcmk_role_text(history->rsc->priv->orig_role),
pcmk__btoa(history->node->details->unclean),
pcmk__on_fail_text(config_on_fail),
pcmk_role_text(fail_role));
if ((fail_role != pcmk_role_started)
&& (history->rsc->priv->next_role < fail_role)) {
pe__set_next_role(history->rsc, fail_role, "failure");
}
if (fail_role == pcmk_role_stopped) {
ban_from_all_nodes(history->rsc);
}
}
/*!
* \internal
* \brief Block a resource with a failed action if it cannot be recovered
*
* If resource action is a failed stop and fencing is not possible, mark the
* resource as unmanaged and blocked, since recovery cannot be done.
*
* \param[in,out] history Parsed action history entry
*/
static void
block_if_unrecoverable(struct action_history *history)
{
char *last_change_s = NULL;
if (strcmp(history->task, PCMK_ACTION_STOP) != 0) {
return; // All actions besides stop are always recoverable
}
if (pe_can_fence(history->node->priv->scheduler, history->node)) {
return; // Failed stops are recoverable via fencing
}
last_change_s = last_change_str(history->xml);
pcmk__sched_err(history->node->priv->scheduler,
"No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
QB_XS " rc=%d id=%s",
history->rsc->id, history->task,
pcmk__node_name(history->node),
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
last_change_s, history->exit_status, history->id);
free(last_change_s);
pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed);
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked);
}
/*!
* \internal
* \brief Update action history's execution status and why
*
* \param[in,out] history Parsed action history entry
* \param[out] why Where to store reason for update
* \param[in] value New value
* \param[in] reason Description of why value was changed
*/
static inline void
remap_because(struct action_history *history, const char **why, int value,
const char *reason)
{
if (history->execution_status != value) {
history->execution_status = value;
*why = reason;
}
}
/*!
* \internal
* \brief Remap informational monitor results and operation status
*
* For the monitor results, certain OCF codes are for providing extended information
* to the user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by Pacemaker.
*
* For operation status, the action result can be used to determine an appropriate
* status for the purposes of responding to the action. The status provided by the
* executor is not directly usable since the executor does not know what was expected.
*
* \param[in,out] history Parsed action history entry
* \param[in,out] on_fail What should be done about the result
* \param[in] expired Whether result is expired
*
* \note If the result is remapped and the node is not shutting down or failed,
* the operation will be recorded in the scheduler data's list of failed
* operations to highlight it for the user.
*
* \note This may update the resource's current and next role.
*/
static void
remap_operation(struct action_history *history,
enum pcmk__on_fail *on_fail, bool expired)
{
/* @TODO It would probably also be a good idea to map an exit status of
* CRM_EX_PROMOTED or CRM_EX_DEGRADED_PROMOTED to CRM_EX_OK for promote
* actions
*/
bool is_probe = false;
int orig_exit_status = history->exit_status;
int orig_exec_status = history->execution_status;
const char *why = NULL;
const char *task = history->task;
// Remap degraded results to their successful counterparts
history->exit_status = pcmk__effective_rc(history->exit_status);
if (history->exit_status != orig_exit_status) {
why = "degraded result";
if (!expired && (!history->node->details->shutdown
|| history->node->details->online)) {
record_failed_op(history);
}
}
if (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& ((history->execution_status != PCMK_EXEC_DONE)
|| (history->exit_status != PCMK_OCF_NOT_RUNNING))) {
history->execution_status = PCMK_EXEC_DONE;
history->exit_status = PCMK_OCF_NOT_RUNNING;
why = "equivalent probe result";
}
/* If the executor reported an execution status of anything but done or
* error, consider that final. But for done or error, we know better whether
* it should be treated as a failure or not, because we know the expected
* result.
*/
switch (history->execution_status) {
case PCMK_EXEC_DONE:
case PCMK_EXEC_ERROR:
break;
// These should be treated as node-fatal
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"node-fatal error");
goto remap_done;
default:
goto remap_done;
}
is_probe = pcmk_xe_is_probe(history->xml);
if (is_probe) {
task = "probe";
}
if (history->expected_exit_status < 0) {
/* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with
* Heartbeat 2.0.7 or earlier as the cluster layer, did not include the
* expected exit status in the transition key, which (along with the
* similar case of a corrupted transition key in the CIB) will be
* reported to this function as -1. Pacemaker 2.0+ does not support
* rolling upgrades from those versions or processing of saved CIB files
* from those versions, so we do not need to care much about this case.
*/
remap_because(history, &why, PCMK_EXEC_ERROR,
"obsolete history format");
pcmk__config_warn("Expected result not found for %s on %s "
"(corrupt or obsolete CIB?)",
history->key, pcmk__node_name(history->node));
} else if (history->exit_status == history->expected_exit_status) {
remap_because(history, &why, PCMK_EXEC_DONE, "expected result");
} else {
remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result");
pcmk__rsc_debug(history->rsc,
"%s on %s: expected %d (%s), got %d (%s%s%s)",
history->key, pcmk__node_name(history->node),
history->expected_exit_status,
crm_exit_str(history->expected_exit_status),
history->exit_status,
crm_exit_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""));
}
switch (history->exit_status) {
case PCMK_OCF_OK:
if (is_probe
&& (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active on %s at %s",
history->rsc->id, pcmk__node_name(history->node),
last_change_s);
free(last_change_s);
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe
|| (history->expected_exit_status == history->exit_status)
|| !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) {
/* For probes, recurring monitors for the Stopped role, and
* unmanaged resources, "not running" is not considered a
* failure.
*/
remap_because(history, &why, PCMK_EXEC_DONE, "exit status");
history->rsc->priv->orig_role = pcmk_role_stopped;
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe
&& (history->exit_status != history->expected_exit_status)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active and promoted on %s at %s",
history->rsc->id,
pcmk__node_name(history->node), last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->priv->orig_role = pcmk_role_promoted;
}
break;
case PCMK_OCF_FAILED_PROMOTED:
if (!expired) {
history->rsc->priv->orig_role = pcmk_role_promoted;
}
remap_because(history, &why, PCMK_EXEC_ERROR, "exit status");
break;
case PCMK_OCF_NOT_CONFIGURED:
remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status");
break;
case PCMK_OCF_UNIMPLEMENT_FEATURE:
{
guint interval_ms = 0;
pcmk__xe_get_guint(history->xml, PCMK_META_INTERVAL,
&interval_ms);
if (interval_ms == 0) {
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"exit status");
} else {
remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED,
"exit status");
}
}
break;
case PCMK_OCF_NOT_INSTALLED:
case PCMK_OCF_INVALID_PARAM:
case PCMK_OCF_INSUFFICIENT_PRIV:
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status");
break;
default:
if (history->execution_status == PCMK_EXEC_DONE) {
char *last_change_s = last_change_str(history->xml);
crm_info("Treating unknown exit status %d from %s of %s "
"on %s at %s as failure",
history->exit_status, task, history->rsc->id,
pcmk__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pcmk__add_param_check(xml_op, rsc, node, pcmk__check_last_failure);
} else {
pcmk__op_digest_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->priv->scheduler);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, pcmk__xe_history_key(xml_op),
node->priv->id);
break;
case pcmk__digest_match:
break;
default:
return TRUE;
}
}
}
return FALSE;
}
// Order action after fencing of remote node, given connection rsc
static void
order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn,
pcmk_scheduler_t *scheduler)
{
pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id);
if (remote_node) {
pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL,
FALSE, scheduler);
order_actions(fence, action, pcmk__ar_first_implies_then);
}
}
static bool
should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task,
guint interval_ms, bool is_last_failure)
{
/* Clearing failures of recurring monitors has special concerns. The
* executor reports only changes in the monitor result, so if the
* monitor is still active and still getting the same failure result,
* that will go undetected after the failure is cleared.
*
* Also, the operation history will have the time when the recurring
* monitor result changed to the given code, not the time when the
* result last happened.
*
* @TODO We probably should clear such failures only when the failure
* timeout has passed since the last occurrence of the failed result.
* However we don't record that information. We could maybe approximate
* that by clearing only if there is a more recent successful monitor or
* stop result, but we don't even have that information at this point
* since we are still unpacking the resource's operation history.
*
* This is especially important for remote connection resources with a
* reconnect interval, so in that case, we skip clearing failures
* if the remote node hasn't been fenced.
*/
if ((rsc->priv->remote_reconnect_ms > 0U)
&& pcmk_is_set(rsc->priv->scheduler->flags,
pcmk__sched_fencing_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pcmk_node_t *remote_node = pcmk_find_node(rsc->priv->scheduler,
rsc->id);
if (remote_node && !pcmk_is_set(remote_node->priv->flags,
pcmk__node_remote_fenced)) {
if (is_last_failure) {
crm_info("Waiting to clear monitor failure for remote node %s"
" until fencing has occurred", rsc->id);
}
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Check operation age and schedule failure clearing when appropriate
*
* This function has two distinct purposes. The first is to check whether an
* operation history entry is expired (i.e. the resource has a failure timeout,
* the entry is older than the timeout, and the resource either has no fail
* count or its fail count is entirely older than the timeout). The second is to
* schedule fail count clearing when appropriate (i.e. the operation is expired
* and either the resource has an expired fail count or the operation is a
* last_failure for a remote connection resource with a reconnect interval,
* or the operation is a last_failure for a start or monitor operation and the
* resource's parameters have changed since the operation).
*
* \param[in,out] history Parsed action result history
*
* \return true if operation history entry is expired, otherwise false
*/
static bool
check_operation_expiry(struct action_history *history)
{
bool expired = false;
bool is_last_failure = (history->id != NULL)
&& g_str_has_suffix(history->id, "_last_failure_0");
time_t last_run = 0;
int unexpired_fail_count = 0;
const char *clear_reason = NULL;
const guint expiration_sec =
pcmk__timeout_ms2s(history->rsc->priv->failure_expiration_ms);
pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler;
if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) {
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not expired: "
"Not Installed does not expire",
history->id, pcmk__node_name(history->node));
return false; // "Not installed" must always be cleared manually
}
if ((expiration_sec > 0)
&& (pcmk__xe_get_time(history->xml, PCMK_XA_LAST_RC_CHANGE,
&last_run) == pcmk_rc_ok)) {
/* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a
* timestamp
*/
time_t now = pcmk__scheduler_epoch_time(scheduler);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + expiration_sec))
&& !should_ignore_failure_timeout(history->rsc, history->task,
history->interval_ms,
is_last_failure)) {
expired = true;
}
// Does the resource as a whole have an unexpired fail count?
unexpired_fail_count = pe_get_failcount(history->node, history->rsc,
&last_failure,
pcmk__fc_effective,
history->xml);
// Update scheduler recheck time according to *last* failure
crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d "
"expiration=%s last-failure@%lld",
history->id, (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count,
pcmk__readable_interval(expiration_sec * 1000),
(long long) last_failure);
last_failure += expiration_sec + 1;
if (unexpired_fail_count && (now < last_failure)) {
pcmk__update_recheck_time(last_failure, scheduler,
"fail count expiration");
}
}
if (expired) {
if (pe_get_failcount(history->node, history->rsc, NULL,
pcmk__fc_default, history->xml)) {
// There is a fail count ignoring timeout
if (unexpired_fail_count == 0) {
// There is no fail count considering timeout
clear_reason = "it expired";
} else {
/* This operation is old, but there is an unexpired fail count.
* In a properly functioning cluster, this should only be
* possible if this operation is not a failure (otherwise the
* fail count should be expired too), so this is really just a
* failsafe.
*/
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Unexpired fail count",
history->id, pcmk__node_name(history->node));
expired = false;
}
} else if (is_last_failure
&& (history->rsc->priv->remote_reconnect_ms > 0U)) {
/* Clear any expired last failure when reconnect interval is set,
* even if there is no fail count.
*/
clear_reason = "reconnect interval is set";
}
}
if (!expired && is_last_failure
&& should_clear_for_param_change(history->xml, history->task,
history->rsc, history->node)) {
clear_reason = "resource parameters have changed";
}
if (clear_reason != NULL) {
pcmk_action_t *clear_op = NULL;
// Schedule clearing of the fail count
clear_op = pe__clear_failcount(history->rsc, history->node,
clear_reason, scheduler);
if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)
&& (history->rsc->priv->remote_reconnect_ms > 0)) {
/* If we're clearing a remote connection due to a reconnect
* interval, we want to wait until any scheduled fencing
* completes.
*
* We could limit this to remote_node->details->unclean, but at
* this point, that's always true (it won't be reliable until
* after unpack_node_history() is done).
*/
crm_info("Clearing %s failure will wait until any scheduled "
"fencing of %s completes",
history->task, history->rsc->id);
order_after_remote_fencing(clear_op, history->rsc, scheduler);
}
}
if (expired && (history->interval_ms == 0)
&& pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
switch (history->exit_status) {
case PCMK_OCF_OK:
case PCMK_OCF_NOT_RUNNING:
case PCMK_OCF_RUNNING_PROMOTED:
case PCMK_OCF_DEGRADED:
case PCMK_OCF_DEGRADED_PROMOTED:
// Don't expire probes that return these values
pcmk__rsc_trace(history->rsc,
"Resource history entry %s on %s is not "
"expired: Probe result",
history->id, pcmk__node_name(history->node));
expired = false;
break;
}
}
return expired;
}
int
pe__target_rc_from_xml(const xmlNode *xml_op)
{
int target_rc = 0;
const char *key = pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
/*!
* \internal
* \brief Update a resource's state for an action result
*
* \param[in,out] history Parsed action history entry
* \param[in] exit_status Exit status to base new state on
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
update_resource_state(struct action_history *history, int exit_status,
const xmlNode *last_failure,
enum pcmk__on_fail *on_fail)
{
bool clear_past_failure = false;
if ((exit_status == PCMK_OCF_NOT_INSTALLED)
|| (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml))) {
history->rsc->priv->orig_role = pcmk_role_stopped;
} else if (exit_status == PCMK_OCF_NOT_RUNNING) {
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR,
pcmk__str_none)) {
if ((last_failure != NULL)
&& pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure),
pcmk__str_none)) {
clear_past_failure = true;
}
if (history->rsc->priv->orig_role < pcmk_role_started) {
set_active(history->rsc);
}
} else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_stopped;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_promoted;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE,
pcmk__str_none)) {
if (*on_fail == pcmk__on_fail_demote) {
/* Demote clears an error only if
* PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE
*/
clear_past_failure = true;
}
history->rsc->priv->orig_role = pcmk_role_unpromoted;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
history->rsc->priv->orig_role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO,
pcmk__str_none)) {
unpack_migrate_to_success(history);
} else if (history->rsc->priv->orig_role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "%s active on %s",
history->rsc->id, pcmk__node_name(history->node));
set_active(history->rsc);
}
if (!clear_past_failure) {
return;
}
switch (*on_fail) {
case pcmk__on_fail_stop:
case pcmk__on_fail_ban:
case pcmk__on_fail_standby_node:
case pcmk__on_fail_fence_node:
pcmk__rsc_trace(history->rsc,
"%s (%s) is not cleared by a completed %s",
history->rsc->id, pcmk__on_fail_text(*on_fail),
history->task);
break;
case pcmk__on_fail_block:
case pcmk__on_fail_ignore:
case pcmk__on_fail_demote:
case pcmk__on_fail_restart:
case pcmk__on_fail_restart_container:
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures");
break;
case pcmk__on_fail_reset_remote:
if (history->rsc->priv->remote_reconnect_ms == 0U) {
/* With no reconnect interval, the connection is allowed to
* start again after the remote node is fenced and
* completely stopped. (With a reconnect interval, we wait
* for the failure to be cleared entirely before attempting
* to reconnect.)
*/
*on_fail = pcmk__on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures and reset remote");
}
break;
}
}
/*!
* \internal
* \brief Check whether a given history entry matters for resource state
*
* \param[in] history Parsed action history entry
*
* \return true if action can affect resource state, otherwise false
*/
static inline bool
can_affect_state(struct action_history *history)
{
return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
"asyncmon", NULL);
}
/*!
* \internal
* \brief Unpack execution/exit status and exit reason from a history entry
*
* \param[in,out] history Action history entry to unpack
*
* \return Standard Pacemaker return code
*/
static int
unpack_action_result(struct action_history *history)
{
if ((pcmk__xe_get_int(history->xml, PCMK__XA_OP_STATUS,
&(history->execution_status)) != pcmk_rc_ok)
|| (history->execution_status < PCMK_EXEC_PENDING)
|| (history->execution_status > PCMK_EXEC_MAX)
|| (history->execution_status == PCMK_EXEC_CANCELLED)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_OP_STATUS " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(pcmk__xe_get(history->xml, PCMK__XA_OP_STATUS),
""));
return pcmk_rc_unpack_error;
}
if ((pcmk__xe_get_int(history->xml, PCMK__XA_RC_CODE,
&(history->exit_status)) != pcmk_rc_ok)
|| (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " PCMK__XA_RC_CODE " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(pcmk__xe_get(history->xml, PCMK__XA_RC_CODE),
""));
return pcmk_rc_unpack_error;
}
history->exit_reason = pcmk__xe_get(history->xml, PCMK_XA_EXIT_REASON);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Process an action history entry whose result expired
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
*
* \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the
* entry needs no further processing)
*/
static int
process_expired_result(struct action_history *history, int orig_exit_status)
{
if (!pcmk__is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& (orig_exit_status != history->expected_exit_status)) {
if (history->rsc->priv->orig_role <= pcmk_role_stopped) {
history->rsc->priv->orig_role = pcmk_role_unknown;
}
crm_trace("Ignoring resource history entry %s for probe of %s on %s: "
"Masked failure expired",
history->id, history->rsc->id,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->exit_status == history->expected_exit_status) {
return pcmk_rc_undetermined; // Only failures expire
}
if (history->interval_ms == 0) {
crm_notice("Ignoring resource history entry %s for %s of %s on %s: "
"Expired failure",
history->id, history->task, history->rsc->id,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->node->details->online && !history->node->details->unclean) {
/* Reschedule the recurring action. schedule_cancel() won't work at
* this stage, so as a hacky workaround, forcibly change the restart
* digest so pcmk__check_action_config() does what we want later.
*
* @TODO We should skip this if there is a newer successful monitor.
* Also, this causes rescheduling only if the history entry
* has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure
* scheduler regression test doesn't, but that may not be a
* realistic scenario in production).
*/
crm_notice("Rescheduling %s-interval %s of %s on %s "
"after failure expired",
pcmk__readable_interval(history->interval_ms), history->task,
history->rsc->id, pcmk__node_name(history->node));
pcmk__xe_set(history->xml, PCMK__XA_OP_RESTART_DIGEST,
"calculated-failure-timeout");
return pcmk_rc_ok;
}
return pcmk_rc_undetermined;
}
/*!
* \internal
* \brief Process a masked probe failure
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
mask_probe_failure(struct action_history *history, int orig_exit_status,
const xmlNode *last_failure,
enum pcmk__on_fail *on_fail)
{
pcmk_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) {
ban_rsc = uber_parent(history->rsc);
}
crm_notice("Treating probe result '%s' for %s on %s as 'not running'",
crm_exit_str(orig_exit_status), history->rsc->id,
pcmk__node_name(history->node));
update_resource_state(history, history->expected_exit_status, last_failure,
on_fail);
pcmk__xe_set(history->xml, PCMK_XA_UNAME, history->node->priv->name);
record_failed_op(history);
resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY,
"masked-probe-failure", ban_rsc->priv->scheduler);
}
/*!
* \internal Check whether a given failure is for a given pending action
*
* \param[in] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*
* \return true if \p last_failure is failure of pending action in \p history,
* otherwise false
* \note Both \p history and \p last_failure must come from the same
* \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be
* the same.
*/
static bool
failure_is_newer(const struct action_history *history,
const xmlNode *last_failure)
{
guint failure_interval_ms = 0U;
long long failure_change = 0LL;
long long this_change = 0LL;
if (last_failure == NULL) {
return false; // Resource has no last_failure entry
}
if (!pcmk__str_eq(history->task,
pcmk__xe_get(last_failure, PCMK_XA_OPERATION),
pcmk__str_none)) {
return false; // last_failure is for different action
}
if ((pcmk__xe_get_guint(last_failure, PCMK_META_INTERVAL,
&failure_interval_ms) != pcmk_rc_ok)
|| (history->interval_ms != failure_interval_ms)) {
return false; // last_failure is for action with different interval
}
if ((pcmk__scan_ll(pcmk__xe_get(history->xml, PCMK_XA_LAST_RC_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(pcmk__xe_get(last_failure, PCMK_XA_LAST_RC_CHANGE),
&failure_change, 0LL) != pcmk_rc_ok)
|| (failure_change < this_change)) {
return false; // Failure is not known to be newer
}
return true;
}
/*!
* \internal
* \brief Update a resource's role etc. for a pending action
*
* \param[in,out] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*/
static void
process_pending_action(struct action_history *history,
const xmlNode *last_failure)
{
/* For recurring monitors, a failure is recorded only in RSC_last_failure_0,
* and there might be a RSC_monitor_INTERVAL entry with the last successful
* or pending result.
*
* If last_failure contains the failure of the pending recurring monitor
* we're processing here, and is newer, the action is no longer pending.
* (Pending results have call ID -1, which sorts last, so the last failure
* if any should be known.)
*/
if (failure_is_newer(history, last_failure)) {
return;
}
if (strcmp(history->task, PCMK_ACTION_START) == 0) {
pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending);
set_active(history->rsc);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->priv->orig_role = pcmk_role_promoted;
} else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0)
&& history->node->details->unclean) {
/* A migrate_to action is pending on a unclean source, so force a stop
* on the target.
*/
const char *migrate_target = NULL;
pcmk_node_t *target = NULL;
migrate_target = pcmk__xe_get(history->xml, PCMK__META_MIGRATE_TARGET);
target = pcmk_find_node(history->rsc->priv->scheduler,
migrate_target);
if (target != NULL) {
stop_action(history->rsc, target, FALSE);
}
}
if (history->rsc->priv->pending_action != NULL) {
/* There should never be multiple pending actions, but as a failsafe,
* just remember the first one processed for display purposes.
*/
return;
}
if (pcmk_is_probe(history->task, history->interval_ms)) {
/* Pending probes are currently never displayed, even if pending
* operations are requested. If we ever want to change that,
* enable the below and the corresponding part of
* native.c:native_pending_action().
*/
#if 0
history->rsc->private->pending_action = strdup("probe");
history->rsc->private->pending_node = history->node;
#endif
} else {
history->rsc->priv->pending_action = strdup(history->task);
history->rsc->priv->pending_node = history->node;
}
}
static void
unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum pcmk__on_fail *on_fail)
{
int old_rc = 0;
bool expired = false;
pcmk_resource_t *parent = rsc;
enum rsc_role_e fail_role = pcmk_role_unknown;
enum pcmk__on_fail failure_strategy = pcmk__on_fail_restart;
struct action_history history = {
.rsc = rsc,
.node = node,
.xml = xml_op,
.execution_status = PCMK_EXEC_UNKNOWN,
};
CRM_CHECK(rsc && node && xml_op, return);
history.id = pcmk__xe_id(xml_op);
if (history.id == NULL) {
pcmk__config_err("Ignoring resource history entry for %s on %s "
"without ID", rsc->id, pcmk__node_name(node));
return;
}
// Task and interval
history.task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION);
if (history.task == NULL) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"without " PCMK_XA_OPERATION,
history.id, rsc->id, pcmk__node_name(node));
return;
}
pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &(history.interval_ms));
if (!can_affect_state(&history)) {
pcmk__rsc_trace(rsc,
"Ignoring resource history entry %s for %s on %s "
"with irrelevant action '%s'",
history.id, rsc->id, pcmk__node_name(node),
history.task);
return;
}
if (unpack_action_result(&history) != pcmk_rc_ok) {
return; // Error already logged
}
history.expected_exit_status = pe__target_rc_from_xml(xml_op);
history.key = pcmk__xe_history_key(xml_op);
pcmk__xe_get_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pcmk__node_name(node),
pcmk_exec_status_str(history.execution_status),
crm_exit_str(history.exit_status));
if (node->details->unclean) {
pcmk__rsc_trace(rsc,
"%s is running on %s, which is unclean (further action "
"depends on value of stop's on-fail attribute)",
rsc->id, pcmk__node_name(node));
}
expired = check_operation_expiry(&history);
old_rc = history.exit_status;
remap_operation(&history, on_fail, expired);
if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) {
goto done;
}
if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) {
mask_probe_failure(&history, old_rc, *last_failure, on_fail);
goto done;
}
if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
parent = uber_parent(rsc);
}
switch (history.execution_status) {
case PCMK_EXEC_PENDING:
process_pending_action(&history, *last_failure);
goto done;
case PCMK_EXEC_DONE:
update_resource_state(&history, history.exit_status, *last_failure,
on_fail);
goto done;
case PCMK_EXEC_NOT_INSTALLED:
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if (failure_strategy == pcmk__on_fail_ignore) {
crm_warn("Cannot ignore failed %s of %s on %s: "
"Resource agent doesn't exist "
QB_XS " status=%d rc=%d id=%s",
history.task, rsc->id, pcmk__node_name(node),
history.execution_status, history.exit_status,
history.id);
/* Also for printing it as "FAILED" by marking it as
* pcmk__rsc_failed later
*/
*on_fail = pcmk__on_fail_ban;
}
resource_location(parent, node, -PCMK_SCORE_INFINITY,
"hard-error", rsc->priv->scheduler);
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
goto done;
case PCMK_EXEC_NOT_CONNECTED:
if (pcmk__is_pacemaker_remote_node(node)
&& pcmk_is_set(node->priv->remote->flags,
pcmk__rsc_managed)) {
/* We should never get into a situation where a managed remote
* connection resource is considered OK but a resource action
* behind the connection gets a "not connected" status. But as a
* fail-safe in case a bug or unusual circumstances do lead to
* that, ensure the remote connection is considered failed.
*/
pcmk__set_rsc_flags(node->priv->remote,
pcmk__rsc_failed|pcmk__rsc_stop_if_failed);
}
break; // Not done, do error handling
case PCMK_EXEC_ERROR:
case PCMK_EXEC_ERROR_HARD:
case PCMK_EXEC_ERROR_FATAL:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_INVALID:
break; // Not done, do error handling
default: // No other value should be possible at this point
break;
}
unpack_failure_handling(&history, &failure_strategy, &fail_role);
if ((failure_strategy == pcmk__on_fail_ignore)
|| ((failure_strategy == pcmk__on_fail_restart_container)
&& (strcmp(history.task, PCMK_ACTION_STOP) == 0))) {
char *last_change_s = last_change_str(xml_op);
crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded "
QB_XS " %s",
history.task, crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), rsc->id,
pcmk__node_name(node), last_change_s, history.id);
free(last_change_s);
update_resource_state(&history, history.expected_exit_status,
*last_failure, on_fail);
pcmk__xe_set(xml_op, PCMK_XA_UNAME, node->priv->name);
pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure);
record_failed_op(&history);
if ((failure_strategy == pcmk__on_fail_restart_container)
&& (*on_fail <= pcmk__on_fail_restart)) {
*on_fail = failure_strategy;
}
} else {
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
if (history.execution_status == PCMK_EXEC_ERROR_HARD) {
uint8_t log_level = LOG_ERR;
if (history.exit_status == PCMK_OCF_NOT_INSTALLED) {
log_level = LOG_NOTICE;
}
do_crm_log(log_level,
"Preventing %s from restarting on %s because "
"of hard failure (%s%s%s) " QB_XS " %s",
parent->id, pcmk__node_name(node),
crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, node, -PCMK_SCORE_INFINITY,
"hard-error", rsc->priv->scheduler);
} else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) {
pcmk__sched_err(rsc->priv->scheduler,
"Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " QB_XS " %s",
parent->id, crm_exit_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, NULL, -PCMK_SCORE_INFINITY,
"fatal-error", rsc->priv->scheduler);
}
}
done:
pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)",
rsc->id, pcmk__node_name(node), history.id,
pcmk_role_text(rsc->priv->orig_role),
pcmk_role_text(rsc->priv->next_role));
}
/*!
* \internal
* \brief Insert a node attribute with value into a \c GHashTable
*
* \param[in,out] key Key to insert (either freed or owned by
* \p user_data upon return)
* \param[in] value Value to insert (owned by \p user_data upon return)
* \param[in] user_data \c GHashTable to insert into
*/
static gboolean
insert_attr(gpointer key, gpointer value, gpointer user_data)
{
GHashTable *table = user_data;
g_hash_table_insert(table, key, value);
return TRUE;
}
static void
add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite,
pcmk_scheduler_t *scheduler)
{
const char *cluster_name = NULL;
const char *dc_id = pcmk__xe_get(scheduler->input, PCMK_XA_DC_UUID);
const pcmk_rule_input_t rule_input = {
.now = scheduler->priv->now,
};
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_UNAME, node->priv->name);
pcmk__insert_dup(node->priv->attrs, CRM_ATTR_ID, node->priv->id);
if ((scheduler->dc_node == NULL)
&& pcmk__str_eq(node->priv->id, dc_id, pcmk__str_casei)) {
scheduler->dc_node = node;
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_TRUE);
} else if (!pcmk__same_node(node, scheduler->dc_node)) {
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_IS_DC, PCMK_VALUE_FALSE);
}
cluster_name = g_hash_table_lookup(scheduler->priv->options,
PCMK_OPT_CLUSTER_NAME);
if (cluster_name) {
pcmk__insert_dup(node->priv->attrs, CRM_ATTR_CLUSTER_NAME,
cluster_name);
}
if (overwrite) {
/* @TODO Try to reorder some unpacking so that we don't need the
* overwrite argument or to unpack into a temporary table
*/
GHashTable *unpacked = pcmk__strkey_table(free, free);
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, unpacked, NULL, scheduler);
g_hash_table_foreach_steal(unpacked, insert_attr, node->priv->attrs);
g_hash_table_destroy(unpacked);
} else {
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES,
&rule_input, node->priv->attrs, NULL,
scheduler);
}
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_input,
node->priv->utilization, NULL, scheduler);
if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL,
pcmk__rsc_node_current) == NULL) {
const char *site_name = pcmk__node_attr(node, "site-name", NULL,
pcmk__rsc_node_current);
if (site_name) {
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_SITE_NAME, site_name);
} else if (cluster_name) {
/* Default to cluster-name if unset */
pcmk__insert_dup(node->priv->attrs,
CRM_ATTR_SITE_NAME, cluster_name);
}
}
}
static GList *
extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter)
{
int counter = -1;
int stop_index = -1;
int start_index = -1;
xmlNode *rsc_op = NULL;
GList *gIter = NULL;
GList *op_list = NULL;
GList *sorted_op_list = NULL;
/* extract operations */
op_list = NULL;
sorted_op_list = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL,
NULL);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) {
pcmk__xe_set(rsc_op, PCMK_XA_RESOURCE, rsc);
pcmk__xe_set(rsc_op, PCMK_XA_UNAME, node);
op_list = g_list_prepend(op_list, rsc_op);
}
if (op_list == NULL) {
/* if there are no operations, there is nothing to do */
return NULL;
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
/* create active recurring operations as optional */
if (active_filter == FALSE) {
return sorted_op_list;
}
op_list = NULL;
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
counter++;
if (start_index < stop_index) {
crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op));
continue;
}
op_list = g_list_append(op_list, rsc_op);
}
g_list_free(sorted_op_list);
return op_list;
}
GList *
find_operations(const char *rsc, const char *node, gboolean active_filter,
pcmk_scheduler_t *scheduler)
{
GList *output = NULL;
GList *intermediate = NULL;
xmlNode *tmp = NULL;
xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS,
NULL, NULL);
pcmk_node_t *this_node = NULL;
xmlNode *node_state = NULL;
CRM_CHECK(status != NULL, return NULL);
for (node_state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL,
NULL);
node_state != NULL;
node_state = pcmk__xe_next(node_state, PCMK__XE_NODE_STATE)) {
const char *uname = pcmk__xe_get(node_state, PCMK_XA_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pcmk_find_node(scheduler, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pcmk__is_pacemaker_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
} else {
determine_online_status(node_state, this_node, scheduler);
}
if (this_node->details->online
|| pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
- /* offline nodes run no resources...
- * unless stonith is enabled in which case we need to
- * make sure rsc start events happen after the stonith
+
+ /* Offline nodes run no resources if fencing is disabled. If fencing
+ * is enabled, we need to ensure that resource start events happen
+ * after the fencing event.
*/
xmlNode *lrm_rsc = NULL;
tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL,
NULL);
tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL,
NULL);
for (lrm_rsc = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCE,
NULL, NULL);
lrm_rsc != NULL;
lrm_rsc = pcmk__xe_next(lrm_rsc, PCMK__XE_LRM_RESOURCE)) {
const char *rsc_id = pcmk__xe_get(lrm_rsc, PCMK_XA_ID);
if ((rsc != NULL)
&& !pcmk__str_eq(rsc_id, rsc, pcmk__str_none)) {
continue;
}
intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
output = g_list_concat(output, intermediate);
}
}
}
return output;
}
diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c
index ae71f7a57a..0354b53054 100644
--- a/lib/pengine/utils.c
+++ b/lib/pengine/utils.c
@@ -1,922 +1,922 @@
/*
* Copyright 2004-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data);
/*!
* \internal
* \brief Check whether we can fence a particular node
*
* \param[in] scheduler Scheduler data
* \param[in] node Name of node to check
*
* \return true if node can be fenced, false otherwise
*/
bool
pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node)
{
if (pcmk__is_guest_or_bundle_node(node)) {
/* A guest or bundle node is fenced by stopping its launcher, which is
* possible if the launcher's host is either online or fenceable.
*/
pcmk_resource_t *rsc = node->priv->remote->priv->launcher;
for (GList *n = rsc->priv->active_nodes; n != NULL; n = n->next) {
pcmk_node_t *launcher_node = n->data;
if (!launcher_node->details->online
&& !pe_can_fence(scheduler, launcher_node)) {
return false;
}
}
return true;
} else if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) {
return false; /* Turned off */
} else if (!pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) {
return false; /* No devices */
} else if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) {
return true;
} else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) {
return true;
} else if (node == NULL) {
return false;
} else if (node->details->online) {
/* Remote nodes are marked online when we assign their resource to a
* node, not when they are actually started (see remote_connection_assigned)
* so the above test by itself isn't good enough.
*/
if (pcmk__is_pacemaker_remote_node(node)
&& !pcmk_is_set(scheduler->flags, pcmk__sched_fence_remote_no_quorum)) {
/* If we're on a system without quorum, it's entirely possible that
* the remote resource was automatically moved to a node on the
* partition with quorum. We can't tell that from this node - the
* best we can do is check if it's possible for the resource to run
* on another node in the partition with quorum. If so, it has
* likely been moved and we shouldn't fence it.
*
* NOTE: This condition appears to only come up in very limited
* circumstances. It at least requires some very lengthy fencing
* timeouts set, some way for fencing to still take place (a second
* NIC is how I've reproduced it in testing, but fence_scsi or
* sbd could work too), and a resource that runs on the remote node.
*/
pcmk_resource_t *rsc = node->priv->remote;
pcmk_node_t *n = NULL;
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &n)) {
/* A node that's not online according to this non-quorum node
* is a node that's in another partition.
*/
if (!n->details->online) {
return false;
}
}
}
crm_notice("We can fence %s without quorum because they're in our membership",
pcmk__node_name(node));
return true;
}
crm_trace("Cannot fence %s", pcmk__node_name(node));
return false;
}
/*!
* \internal
* \brief Copy a node object
*
* \param[in] this_node Node object to copy
*
* \return Newly allocated shallow copy of this_node
* \note This function asserts on errors and is guaranteed to return non-NULL.
* The caller is responsible for freeing the result using
* pcmk__free_node_copy().
*/
pcmk_node_t *
pe__copy_node(const pcmk_node_t *this_node)
{
pcmk_node_t *new_node = NULL;
pcmk__assert(this_node != NULL);
new_node = pcmk__assert_alloc(1, sizeof(pcmk_node_t));
new_node->assign = pcmk__assert_alloc(1,
sizeof(struct pcmk__node_assignment));
new_node->assign->probe_mode = this_node->assign->probe_mode;
new_node->assign->score = this_node->assign->score;
new_node->assign->count = this_node->assign->count;
new_node->details = this_node->details;
new_node->priv = this_node->priv;
return new_node;
}
/*!
* \internal
* \brief Create a hash table of node copies from a list of nodes
*
* \param[in] list Node list
*
* \return Hash table equivalent of node list
*/
GHashTable *
pe__node_list2table(const GList *list)
{
GHashTable *result = NULL;
result = pcmk__strkey_table(NULL, pcmk__free_node_copy);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *new_node = NULL;
new_node = pe__copy_node((const pcmk_node_t *) gIter->data);
g_hash_table_insert(result, (gpointer) new_node->priv->id, new_node);
}
return result;
}
/*!
* \internal
* \brief Compare two nodes by name, with numeric portions sorted numerically
*
* Sort two node names case-insensitively like strcasecmp(), but with any
* numeric portions of the name sorted numerically. For example, "node10" will
* sort higher than "node9" but lower than "remotenode9".
*
* \param[in] a First node to compare (can be \c NULL)
* \param[in] b Second node to compare (can be \c NULL)
*
* \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not)
* \retval 0 \c a and \c b are equal (or both are \c NULL)
* \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not)
*/
gint
pe__cmp_node_name(gconstpointer a, gconstpointer b)
{
const pcmk_node_t *node1 = (const pcmk_node_t *) a;
const pcmk_node_t *node2 = (const pcmk_node_t *) b;
if ((node1 == NULL) && (node2 == NULL)) {
return 0;
}
if (node1 == NULL) {
return -1;
}
if (node2 == NULL) {
return 1;
}
return pcmk__numeric_strcasecmp(node1->priv->name, node2->priv->name);
}
/*!
* \internal
* \brief Output node weights to stdout
*
* \param[in] rsc Use allowed nodes for this resource
* \param[in] comment Text description to prefix lines with
* \param[in] nodes If rsc is not specified, use these nodes
* \param[in,out] scheduler Scheduler data
*/
static void
pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes, pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->priv->out;
// Sort the nodes so the output is consistent for regression tests
GList *list = g_list_sort(g_hash_table_get_values(nodes),
pe__cmp_node_name);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) gIter->data;
out->message(out, "node-weight", rsc, comment, node->priv->name,
pcmk_readable_score(node->assign->score));
}
g_list_free(list);
}
/*!
* \internal
* \brief Log node weights at trace level
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] rsc If not NULL, include this resource's ID in logs
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be logged
*/
static void
pe__log_node_weights(const char *file, const char *function, int line,
const pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
// Don't waste time if we're not tracing at this point
pcmk__if_tracing({}, return);
g_hash_table_iter_init(&iter, nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (rsc) {
qb_log_from_external_source(function, file,
"%s: %s allocation score on %s: %s",
LOG_TRACE, line, 0,
comment, rsc->id,
pcmk__node_name(node),
pcmk_readable_score(node->assign->score));
} else {
qb_log_from_external_source(function, file, "%s: %s = %s",
LOG_TRACE, line, 0,
comment, pcmk__node_name(node),
pcmk_readable_score(node->assign->score));
}
}
}
/*!
* \internal
* \brief Log or output node weights
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] to_log Log if true, otherwise output
* \param[in] rsc If not NULL, use this resource's ID in logs,
* and show scores recursively for any children
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be shown
* \param[in,out] scheduler Scheduler data
*/
void
pe__show_node_scores_as(const char *file, const char *function, int line,
bool to_log, const pcmk_resource_t *rsc,
const char *comment, GHashTable *nodes,
pcmk_scheduler_t *scheduler)
{
if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) {
// Don't show allocation scores for removed resources
return;
}
if (nodes == NULL) {
// Nothing to show
return;
}
if (to_log) {
pe__log_node_weights(file, function, line, rsc, comment, nodes);
} else {
pe__output_node_weights(rsc, comment, nodes, scheduler);
}
if (rsc == NULL) {
return;
}
// If this resource has children, repeat recursively for each
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) gIter->data;
pe__show_node_scores_as(file, function, line, to_log, child, comment,
child->priv->allowed_nodes, scheduler);
}
}
/*!
* \internal
* \brief Compare two resources by priority
*
* \param[in] a First resource to compare (can be \c NULL)
* \param[in] b Second resource to compare (can be \c NULL)
*
* \retval -1 a's priority > b's priority (or \c b is \c NULL and \c a is not)
* \retval 0 a's priority == b's priority (or both \c a and \c b are \c NULL)
* \retval 1 a's priority < b's priority (or \c a is \c NULL and \c b is not)
*/
gint
pe__cmp_rsc_priority(gconstpointer a, gconstpointer b)
{
const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a;
const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b;
if (a == NULL && b == NULL) {
return 0;
}
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (resource1->priv->priority > resource2->priv->priority) {
return -1;
}
if (resource1->priv->priority < resource2->priv->priority) {
return 1;
}
return 0;
}
static void
resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag)
{
pcmk_node_t *match = NULL;
if ((pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes)
|| (node->assign->probe_mode == pcmk__probe_never))
&& pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) {
/* This string comparision may be fragile, but exclusive resources and
* exclusive nodes should not have the symmetric_default constraint
* applied to them.
*/
return;
} else {
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
resource_node_score(child_rsc, node, score, tag);
}
}
match = g_hash_table_lookup(rsc->priv->allowed_nodes, node->priv->id);
if (match == NULL) {
match = pe__copy_node(node);
g_hash_table_insert(rsc->priv->allowed_nodes,
(gpointer) match->priv->id, match);
}
match->assign->score = pcmk__add_scores(match->assign->score, score);
pcmk__rsc_trace(rsc,
"Enabling %s preference (%s) for %s on %s (now %s)",
tag, pcmk_readable_score(score), rsc->id,
pcmk__node_name(node),
pcmk_readable_score(match->assign->score));
}
void
resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag, pcmk_scheduler_t *scheduler)
{
if (node != NULL) {
resource_node_score(rsc, node, score, tag);
} else if (scheduler != NULL) {
GList *gIter = scheduler->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data;
resource_node_score(rsc, node_iter, score, tag);
}
} else {
GHashTableIter iter;
pcmk_node_t *node_iter = NULL;
g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) {
resource_node_score(rsc, node_iter, score, tag);
}
}
if ((node == NULL) && (score == -PCMK_SCORE_INFINITY)
&& (rsc->priv->assigned_node != NULL)) {
// @TODO Should this be more like pcmk__unassign_resource()?
crm_info("Unassigning %s from %s",
rsc->id, pcmk__node_name(rsc->priv->assigned_node));
pcmk__free_node_copy(rsc->priv->assigned_node);
rsc->priv->assigned_node = NULL;
}
}
gboolean
get_target_role(const pcmk_resource_t *rsc, enum rsc_role_e *role)
{
enum rsc_role_e local_role = pcmk_role_unknown;
const char *value = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_TARGET_ROLE);
CRM_CHECK(role != NULL, return FALSE);
if (pcmk__str_eq(value, PCMK_ROLE_STARTED,
pcmk__str_null_matches|pcmk__str_casei)) {
return FALSE;
}
if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) {
// @COMPAT Deprecated since 2.1.8
pcmk__config_warn("Support for setting " PCMK_META_TARGET_ROLE
" to the explicit value '" PCMK_VALUE_DEFAULT
"' is deprecated and will be removed in a "
"future release (just leave it unset)");
return FALSE;
}
local_role = pcmk_parse_role(value);
if (local_role == pcmk_role_unknown) {
pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s "
"because '%s' is not valid", rsc->id, value);
return FALSE;
} else if (local_role > pcmk_role_started) {
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_promotable)) {
if (local_role > pcmk_role_unpromoted) {
/* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */
return FALSE;
}
} else {
pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s "
"because '%s' only makes sense for promotable "
"clones", rsc->id, value);
return FALSE;
}
}
*role = local_role;
return TRUE;
}
gboolean
order_actions(pcmk_action_t *first, pcmk_action_t *then, uint32_t flags)
{
GList *gIter = NULL;
pcmk__related_action_t *wrapper = NULL;
GList *list = NULL;
if (flags == pcmk__ar_none) {
return FALSE;
}
if ((first == NULL) || (then == NULL)) {
return FALSE;
}
crm_trace("Creating action wrappers for ordering: %s then %s",
first->uuid, then->uuid);
/* Ensure we never create a dependency on ourselves... it's happened */
pcmk__assert(first != then);
/* Filter dups, otherwise update_action_states() has too much work to do */
gIter = first->actions_after;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__related_action_t *after = gIter->data;
if ((after->action == then)
&& pcmk_any_flags_set(after->flags, flags)) {
return FALSE;
}
}
wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t));
wrapper->action = then;
wrapper->flags = flags;
list = first->actions_after;
list = g_list_prepend(list, wrapper);
first->actions_after = list;
wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t));
wrapper->action = first;
wrapper->flags = flags;
list = then->actions_before;
list = g_list_prepend(list, wrapper);
then->actions_before = list;
return TRUE;
}
void
destroy_ticket(gpointer data)
{
pcmk__ticket_t *ticket = data;
if (ticket->state) {
g_hash_table_destroy(ticket->state);
}
free(ticket->id);
free(ticket);
}
pcmk__ticket_t *
ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler)
{
pcmk__ticket_t *ticket = NULL;
if (pcmk__str_empty(ticket_id)) {
return NULL;
}
if (scheduler->priv->ticket_constraints == NULL) {
scheduler->priv->ticket_constraints =
pcmk__strkey_table(free, destroy_ticket);
}
ticket = g_hash_table_lookup(scheduler->priv->ticket_constraints,
ticket_id);
if (ticket == NULL) {
ticket = calloc(1, sizeof(pcmk__ticket_t));
if (ticket == NULL) {
pcmk__sched_err(scheduler, "Cannot allocate ticket '%s'",
ticket_id);
return NULL;
}
crm_trace("Creating ticket entry for %s", ticket_id);
ticket->id = strdup(ticket_id);
ticket->last_granted = -1;
ticket->state = pcmk__strkey_table(free, free);
g_hash_table_insert(scheduler->priv->ticket_constraints,
pcmk__str_copy(ticket->id), ticket);
}
return ticket;
}
const char *
rsc_printable_id(const pcmk_resource_t *rsc)
{
if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) {
return rsc->id;
}
return pcmk__xe_id(rsc->priv->xml);
}
void
pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__clear_rsc_flags(rsc, flags);
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag)
{
for (GList *lpc = scheduler->priv->resources;
lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *r = (pcmk_resource_t *) lpc->data;
pe__clear_resource_flags_recursive(r, flag);
}
}
void
pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__set_rsc_flags(rsc, flags);
for (GList *gIter = rsc->priv->children;
gIter != NULL; gIter = gIter->next) {
pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason,
pcmk_action_t *dependency, pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk__sched_enable_unfencing)) {
/* No resources require it */
return;
} else if ((rsc != NULL)
&& !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) {
- /* Wasn't a stonith device */
+ // Wasn't a fencing device
return;
} else if(node
&& node->details->online
&& node->details->unclean == FALSE
&& node->details->shutdown == FALSE) {
pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE,
reason, FALSE, scheduler);
if(dependency) {
order_actions(unfence, dependency, pcmk__ar_ordered);
}
} else if(rsc) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) {
trigger_unfencing(rsc, node, reason, dependency, scheduler);
}
}
}
}
/*!
* \internal
* \brief Check whether shutdown has been requested for a node
*
* \param[in] node Node to check
*
* \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise
* \note This differs from simply using node->details->shutdown in that it can
* be used before that has been determined (and in fact to determine it),
* and it can also be used to distinguish requested shutdown from implicit
* shutdown of remote nodes by virtue of their connection stopping.
*/
bool
pe__shutdown_requested(const pcmk_node_t *node)
{
const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL,
pcmk__rsc_node_current);
return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches);
}
/*!
* \internal
* \brief Extract nvpair blocks contained by a CIB XML element into a hash table
*
* \param[in] xml_obj XML element containing blocks of nvpair elements
* \param[in] set_name If not NULL, only use blocks of this element
* \param[in] rule_input Values used to evaluate rule criteria
* (node_attrs member must be NULL if \p set_name
* is PCMK_XE_META_ATTRIBUTES)
* \param[out] hash Where to store extracted name/value pairs
* \param[in] always_first If not NULL, process block with this ID first
* \param[in,out] scheduler Scheduler data containing \p xml_obj
*/
void
pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name,
const pcmk_rule_input_t *rule_input,
GHashTable *hash, const char *always_first,
pcmk_scheduler_t *scheduler)
{
crm_time_t *next_change = NULL;
CRM_CHECK((set_name != NULL) && (rule_input != NULL) && (hash != NULL)
&& (scheduler != NULL), return);
// Node attribute expressions are not allowed for meta-attributes
CRM_CHECK((rule_input->node_attrs == NULL)
|| (strcmp(set_name, PCMK_XE_META_ATTRIBUTES) != 0), return);
if (xml_obj == NULL) {
return;
}
next_change = crm_time_new_undefined();
pcmk_unpack_nvpair_blocks(xml_obj, set_name, always_first, rule_input, hash,
next_change);
if (crm_time_is_defined(next_change)) {
time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change);
pcmk__update_recheck_time(recheck, scheduler, "rule evaluation");
}
crm_time_free(next_change);
}
bool
pe__resource_is_disabled(const pcmk_resource_t *rsc)
{
const char *target_role = NULL;
CRM_CHECK(rsc != NULL, return false);
target_role = g_hash_table_lookup(rsc->priv->meta,
PCMK_META_TARGET_ROLE);
if (target_role) {
// If invalid, we've already logged an error when unpacking
enum rsc_role_e target_role_e = pcmk_parse_role(target_role);
if ((target_role_e == pcmk_role_stopped)
|| ((target_role_e == pcmk_role_unpromoted)
&& pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk__rsc_promotable))) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a resource is running only on given node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p rsc is running only on \p node, otherwise false
*/
bool
pe__rsc_running_on_only(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return (rsc != NULL) && pcmk__list_of_1(rsc->priv->active_nodes)
&& pcmk__same_node((const pcmk_node_t *)
rsc->priv->active_nodes->data, node);
}
bool
pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list)
{
if (rsc != NULL) {
for (GList *ele = rsc->priv->active_nodes; ele; ele = ele->next) {
pcmk_node_t *node = (pcmk_node_t *) ele->data;
if (pcmk__str_in_list(node->priv->name, node_list,
pcmk__str_star_matches|pcmk__str_casei)) {
return true;
}
}
}
return false;
}
bool
pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node)
{
return rsc->priv->fns->active(rsc, false)
&& !pe__rsc_running_on_any(rsc, only_node);
}
GList *
pe__filter_rsc_list(GList *rscs, GList *filter)
{
GList *retval = NULL;
for (GList *gIter = rscs; gIter; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
/* I think the second condition is safe here for all callers of this
* function. If not, it needs to move into pe__node_text.
*/
if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) ||
((rsc->priv->parent != NULL)
&& pcmk__str_in_list(rsc_printable_id(rsc->priv->parent),
filter, pcmk__str_star_matches))) {
retval = g_list_prepend(retval, rsc);
}
}
return retval;
}
GList *
pe__build_node_name_list(pcmk_scheduler_t *scheduler, const char *s)
{
GList *nodes = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
/* Nothing was given so return a list of all node names. Or, '*' was
* given. This would normally fall into the pe__unames_with_tag branch
* where it will return an empty list. Catch it here instead.
*/
nodes = g_list_prepend(nodes, strdup("*"));
} else {
pcmk_node_t *node = pcmk_find_node(scheduler, s);
if (node) {
/* The given string was a valid uname for a node. Return a
* singleton list containing just that uname.
*/
nodes = g_list_prepend(nodes, strdup(s));
} else {
/* The given string was not a valid uname. It's either a tag or
* it's a typo or something. In the first case, we'll return a
* list of all the unames of the nodes with the given tag. In the
* second case, we'll return a NULL pointer and nothing will
* get displayed.
*/
nodes = pe__unames_with_tag(scheduler, s);
}
}
return nodes;
}
GList *
pe__build_rsc_list(pcmk_scheduler_t *scheduler, const char *s)
{
GList *resources = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
resources = g_list_prepend(resources, strdup("*"));
} else {
const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename;
pcmk_resource_t *rsc =
pe_find_resource_with_flags(scheduler->priv->resources, s, flags);
if (rsc) {
/* A colon in the name we were given means we're being asked to filter
* on a specific instance of a cloned resource. Put that exact string
* into the filter list. Otherwise, use the printable ID of whatever
* resource was found that matches what was asked for.
*/
if (strstr(s, ":") != NULL) {
resources = g_list_prepend(resources, strdup(rsc->id));
} else {
resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc)));
}
} else {
/* The given string was not a valid resource name. It's a tag or a
* typo or something. See pe__build_node_name_list() for more
* detail.
*/
resources = pe__rscs_with_tag(scheduler, s);
}
}
return resources;
}
xmlNode *
pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name)
{
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
const char *rsc_id = rsc->id;
const pcmk_scheduler_t *scheduler = rsc->priv->scheduler;
if (pcmk__is_clone(parent)) {
rsc_id = pe__clone_child_id(parent);
}
for (xmlNode *xml_op = pcmk__xe_first_child(scheduler->priv->failed,
NULL, NULL, NULL);
xml_op != NULL; xml_op = pcmk__xe_next(xml_op, NULL)) {
const char *value = NULL;
char *op_id = NULL;
/* This resource operation is not a failed probe. */
if (!pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
/* This resource operation was not run on the given node. Note that if name is
* NULL, this will always succeed.
*/
value = pcmk__xe_get(xml_op, PCMK__META_ON_NODE);
if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) {
continue;
}
if (!parse_op_key(pcmk__xe_history_key(xml_op), &op_id, NULL, NULL)) {
continue; // This history entry is missing an operation key
}
/* This resource operation's ID does not match the rsc_id we are looking for. */
if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) {
free(op_id);
continue;
}
free(op_id);
return xml_op;
}
return NULL;
}

File Metadata

Mime Type
text/x-diff
Expires
Tue, Sep 23, 12:58 PM (1 d, 1 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
2392404
Default Alt Text
(384 KB)

Event Timeline