diff --git a/daemons/controld/controld_te_events.c b/daemons/controld/controld_te_events.c
index d0971caf70..335716f39b 100644
--- a/daemons/controld/controld_te_events.c
+++ b/daemons/controld/controld_te_events.c
@@ -1,612 +1,612 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <pacemaker-controld.h>
#include <crm/common/attrd_internal.h>
#include <crm/common/ipc_attrd_internal.h>
/*!
* \internal
* \brief Action numbers of outside events processed in current update diff
*
* This table is to be used as a set. It should be empty when the transitioner
* begins processing a CIB update diff. It ensures that if there are multiple
* events (for example, "_last_0" and "_last_failure_0") for the same action,
* only one of them updates the failcount. Events that originate outside the
* cluster can't be confirmed, since they're not in the transition graph.
*/
static GHashTable *outside_events = NULL;
/*!
* \internal
* \brief Empty the hash table containing action numbers of outside events
*/
void
controld_remove_all_outside_events(void)
{
if (outside_events != NULL) {
g_hash_table_remove_all(outside_events);
}
}
/*!
* \internal
* \brief Destroy the hash table containing action numbers of outside events
*/
void
controld_destroy_outside_events_table(void)
{
if (outside_events != NULL) {
g_hash_table_destroy(outside_events);
outside_events = NULL;
}
}
/*!
* \internal
* \brief Add an outside event's action number to a set
*
* \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the
* event was not already in the set, or \p pcmk_rc_already otherwise.
*/
static int
record_outside_event(gint action_num)
{
if (outside_events == NULL) {
outside_events = g_hash_table_new(NULL, NULL);
}
if (g_hash_table_add(outside_events, GINT_TO_POINTER(action_num))) {
return pcmk_rc_ok;
}
return pcmk_rc_already;
}
gboolean
fail_incompletable_actions(pcmk__graph_t *graph, const char *down_node)
{
const char *target_uuid = NULL;
const char *router = NULL;
const char *router_uuid = NULL;
xmlNode *last_action = NULL;
GList *gIter = NULL;
GList *gIter2 = NULL;
if (graph == NULL || graph->complete) {
return FALSE;
}
gIter = graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data;
if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) {
/* We've already been here */
continue;
}
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data;
if ((action->type == pcmk__pseudo_graph_action)
|| pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) {
continue;
} else if (action->type == pcmk__cluster_graph_action) {
const char *task = crm_element_value(action->xml,
PCMK_XA_OPERATION);
if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) {
continue;
}
}
target_uuid = crm_element_value(action->xml,
PCMK__META_ON_NODE_UUID);
router = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE);
if (router) {
crm_node_t *node = pcmk__get_node(0, router, NULL,
pcmk__node_search_cluster);
if (node) {
router_uuid = node->uuid;
}
}
if (pcmk__str_eq(target_uuid, down_node, pcmk__str_casei) || pcmk__str_eq(router_uuid, down_node, pcmk__str_casei)) {
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
pcmk__set_synapse_flags(synapse, pcmk__synapse_failed);
last_action = action->xml;
stop_te_timer(action);
pcmk__update_graph(graph, action);
if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) {
crm_notice("Action %d (%s) was pending on %s (offline)",
action->id,
crm_element_value(action->xml,
PCMK__XA_OPERATION_KEY),
down_node);
} else {
crm_info("Action %d (%s) is scheduled for %s (offline)",
action->id,
crm_element_value(action->xml, PCMK__XA_OPERATION_KEY),
down_node);
}
}
}
}
if (last_action != NULL) {
crm_info("Node %s shutdown resulted in un-runnable actions", down_node);
abort_transition(INFINITY, pcmk__graph_restart, "Node failure",
last_action);
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Update failure-related node attributes if warranted
*
* \param[in] event XML describing operation that (maybe) failed
* \param[in] event_node_uuid Node that event occurred on
* \param[in] rc Actual operation return code
* \param[in] target_rc Expected operation return code
* \param[in] do_update If TRUE, do update regardless of operation type
* \param[in] ignore_failures If TRUE, update last failure but not fail count
*
* \return TRUE if this was not a direct nack, success or lrm status refresh
*/
static gboolean
update_failcount(const xmlNode *event, const char *event_node_uuid, int rc,
int target_rc, gboolean do_update, gboolean ignore_failures)
{
guint interval_ms = 0;
char *task = NULL;
char *rsc_id = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, PCMK__XA_OPERATION_KEY);
const char *on_uname = crm_peer_uname(event_node_uuid);
const char *origin = crm_element_value(event, PCMK_XA_CRM_DEBUG_ORIGIN);
// Nothing needs to be done for success or status refresh
if (rc == target_rc) {
return FALSE;
} else if (pcmk__str_eq(origin, "build_active_RAs", pcmk__str_casei)) {
crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh",
id, rc, on_uname);
return FALSE;
}
/* Sanity check */
CRM_CHECK(on_uname != NULL, return TRUE);
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval_ms),
crm_err("Couldn't parse: %s", ID(event)); goto bail);
/* Decide whether update is necessary and what value to use */
if ((interval_ms > 0)
|| pcmk__str_eq(task, PCMK_ACTION_PROMOTE, pcmk__str_none)
|| pcmk__str_eq(task, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
do_update = TRUE;
} else if (pcmk__str_eq(task, PCMK_ACTION_START, pcmk__str_none)) {
do_update = TRUE;
value = pcmk__s(controld_globals.transition_graph->failed_start_offset,
CRM_INFINITY_S);
} else if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_none)) {
do_update = TRUE;
value = pcmk__s(controld_globals.transition_graph->failed_stop_offset,
CRM_INFINITY_S);
}
if (do_update) {
pcmk__attrd_query_pair_t *fail_pair = NULL;
pcmk__attrd_query_pair_t *last_pair = NULL;
char *fail_name = NULL;
char *last_name = NULL;
GList *attrs = NULL;
uint32_t opts = pcmk__node_attr_none;
char *now = pcmk__ttoa(time(NULL));
// Fail count will be either incremented or set to infinity
if (!pcmk_str_is_infinity(value)) {
value = PCMK_XA_VALUE "++";
}
if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) {
opts |= pcmk__node_attr_remote;
}
crm_info("Updating %s for %s on %s after failed %s: rc=%d (update=%s, time=%s)",
(ignore_failures? "last failure" : "failcount"),
rsc_id, on_uname, task, rc, value, now);
/* Update the fail count, if we're not ignoring failures */
if (!ignore_failures) {
fail_pair = calloc(1, sizeof(pcmk__attrd_query_pair_t));
CRM_ASSERT(fail_pair != NULL);
fail_name = pcmk__failcount_name(rsc_id, task, interval_ms);
fail_pair->name = fail_name;
fail_pair->value = value;
fail_pair->node = on_uname;
attrs = g_list_prepend(attrs, fail_pair);
}
/* Update the last failure time (even if we're ignoring failures,
* so that failure can still be detected and shown, e.g. by crm_mon)
*/
last_pair = calloc(1, sizeof(pcmk__attrd_query_pair_t));
CRM_ASSERT(last_pair != NULL);
last_name = pcmk__lastfailure_name(rsc_id, task, interval_ms);
last_pair->name = last_name;
last_pair->value = now;
last_pair->node = on_uname;
attrs = g_list_prepend(attrs, last_pair);
update_attrd_list(attrs, opts);
free(fail_name);
free(fail_pair);
free(last_name);
free(last_pair);
g_list_free(attrs);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
pcmk__graph_action_t *
controld_get_action(int id)
{
for (GList *item = controld_globals.transition_graph->synapses;
item != NULL; item = item->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) item->data;
for (GList *item2 = synapse->actions; item2; item2 = item2->next) {
pcmk__graph_action_t *action = (pcmk__graph_action_t *) item2->data;
if (action->id == id) {
return action;
}
}
}
return NULL;
}
pcmk__graph_action_t *
get_cancel_action(const char *id, const char *node)
{
GList *gIter = NULL;
GList *gIter2 = NULL;
gIter = controld_globals.transition_graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data;
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
const char *task = NULL;
const char *target = NULL;
pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data;
task = crm_element_value(action->xml, PCMK_XA_OPERATION);
if (!pcmk__str_eq(PCMK_ACTION_CANCEL, task, pcmk__str_casei)) {
continue;
}
task = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY);
if (!pcmk__str_eq(task, id, pcmk__str_casei)) {
crm_trace("Wrong key %s for %s on %s", task, id, node);
continue;
}
target = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID);
if (node && !pcmk__str_eq(target, node, pcmk__str_casei)) {
crm_trace("Wrong node %s for %s on %s", target, id, node);
continue;
}
crm_trace("Found %s on %s", id, node);
return action;
}
}
return NULL;
}
bool
confirm_cancel_action(const char *id, const char *node_id)
{
const char *op_key = NULL;
const char *node_name = NULL;
pcmk__graph_action_t *cancel = get_cancel_action(id, node_id);
if (cancel == NULL) {
return FALSE;
}
op_key = crm_element_value(cancel->xml, PCMK__XA_OPERATION_KEY);
node_name = crm_element_value(cancel->xml, PCMK__META_ON_NODE);
stop_te_timer(cancel);
te_action_confirmed(cancel, controld_globals.transition_graph);
crm_info("Cancellation of %s on %s confirmed (action %d)",
op_key, node_name, cancel->id);
return TRUE;
}
/* downed nodes are listed like: <downed> <node id="UUID1" /> ... </downed> */
#define XPATH_DOWNED "//" XML_GRAPH_TAG_DOWNED \
"/" XML_CIB_TAG_NODE "[@" PCMK_XA_ID "='%s']"
/*!
* \brief Find a transition event that would have made a specified node down
*
* \param[in] target UUID of node to match
*
* \return Matching event if found, NULL otherwise
*/
pcmk__graph_action_t *
match_down_event(const char *target)
{
pcmk__graph_action_t *match = NULL;
xmlXPathObjectPtr xpath_ret = NULL;
GList *gIter, *gIter2;
char *xpath = crm_strdup_printf(XPATH_DOWNED, target);
for (gIter = controld_globals.transition_graph->synapses;
gIter != NULL && match == NULL;
gIter = gIter->next) {
for (gIter2 = ((pcmk__graph_synapse_t * ) gIter->data)->actions;
gIter2 != NULL && match == NULL;
gIter2 = gIter2->next) {
match = (pcmk__graph_action_t *) gIter2->data;
if (pcmk_is_set(match->flags, pcmk__graph_action_executed)) {
xpath_ret = xpath_search(match->xml, xpath);
if (numXpathResults(xpath_ret) < 1) {
match = NULL;
}
freeXpathObject(xpath_ret);
} else {
// Only actions that were actually started can match
match = NULL;
}
}
}
free(xpath);
if (match != NULL) {
crm_debug("Shutdown action %d (%s) found for node %s", match->id,
crm_element_value(match->xml, PCMK__XA_OPERATION_KEY),
target);
} else {
crm_debug("No reason to expect node %s to be down", target);
}
return match;
}
void
process_graph_event(xmlNode *event, const char *event_node)
{
int rc = -1; // Actual result
int target_rc = -1; // Expected result
int status = -1; // Executor status
int callid = -1; // Executor call ID
int transition_num = -1; // Transition number
int action_num = -1; // Action number within transition
char *update_te_uuid = NULL;
bool ignore_failures = FALSE;
const char *id = NULL;
const char *desc = NULL;
const char *magic = NULL;
const char *uname = NULL;
CRM_ASSERT(event != NULL);
/*
<lrm_rsc_op id="rsc_east-05_last_0" operation_key="rsc_east-05_monitor_0" operation="monitor" crm-debug-origin="do_update_resource" crm_feature_set="3.0.6" transition-key="9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" transition-magic="0:7;9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" call-id="17" rc-code="7" op-status="0" interval="0" last-rc-change="1355361636" exec-time="128" queue-time="0" op-digest="c81f5f40b1c9e859c992e800b1aa6972"/>
*/
magic = crm_element_value(event, PCMK__XA_TRANSITION_KEY);
if (magic == NULL) {
/* non-change */
return;
}
- crm_element_value_int(event, XML_LRM_ATTR_OPSTATUS, &status);
+ crm_element_value_int(event, PCMK__XA_OP_STATUS, &status);
if (status == PCMK_EXEC_PENDING) {
return;
}
id = crm_element_value(event, PCMK__XA_OPERATION_KEY);
crm_element_value_int(event, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(event, XML_LRM_ATTR_CALLID, &callid);
rc = pcmk__effective_rc(rc);
if (decode_transition_key(magic, &update_te_uuid, &transition_num,
&action_num, &target_rc) == FALSE) {
// decode_transition_key() already logged the bad key
crm_err("Can't process action %s result: Incompatible versions? "
CRM_XS " call-id=%d", id, callid);
abort_transition(INFINITY, pcmk__graph_restart, "Bad event", event);
return;
}
if (transition_num == -1) {
// E.g. crm_resource --fail
if (record_outside_event(action_num) != pcmk_rc_ok) {
crm_debug("Outside event with transition key '%s' has already been "
"processed", magic);
goto bail;
}
desc = "initiated outside of the cluster";
abort_transition(INFINITY, pcmk__graph_restart, "Unexpected event",
event);
} else if ((action_num < 0)
|| !pcmk__str_eq(update_te_uuid, controld_globals.te_uuid,
pcmk__str_none)) {
desc = "initiated by a different DC";
abort_transition(INFINITY, pcmk__graph_restart, "Foreign event", event);
} else if ((controld_globals.transition_graph->id != transition_num)
|| controld_globals.transition_graph->complete) {
// Action is not from currently active transition
guint interval_ms = 0;
if (parse_op_key(id, NULL, NULL, &interval_ms)
&& (interval_ms != 0)) {
/* Recurring actions have the transition number they were first
* scheduled in.
*/
if (status == PCMK_EXEC_CANCELLED) {
confirm_cancel_action(id, get_node_id(event));
goto bail;
}
desc = "arrived after initial scheduling";
abort_transition(INFINITY, pcmk__graph_restart,
"Change in recurring result", event);
} else if (controld_globals.transition_graph->id != transition_num) {
desc = "arrived really late";
abort_transition(INFINITY, pcmk__graph_restart, "Old event", event);
} else {
desc = "arrived late";
abort_transition(INFINITY, pcmk__graph_restart, "Inactive graph",
event);
}
} else {
// Event is result of an action from currently active transition
pcmk__graph_action_t *action = controld_get_action(action_num);
if (action == NULL) {
// Should never happen
desc = "unknown";
abort_transition(INFINITY, pcmk__graph_restart, "Unknown event",
event);
} else if (pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) {
/* Nothing further needs to be done if the action has already been
* confirmed. This can happen e.g. when processing both an
* "xxx_last_0" or "xxx_last_failure_0" record as well as the main
* history record, which would otherwise result in incorrectly
* bumping the fail count twice.
*/
crm_log_xml_debug(event, "Event already confirmed:");
goto bail;
} else {
/* An action result needs to be confirmed.
* (This is the only case where desc == NULL.)
*/
if (pcmk__str_eq(crm_meta_value(action->params, PCMK_META_ON_FAIL),
"ignore", pcmk__str_casei)) {
ignore_failures = TRUE;
} else if (rc != target_rc) {
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
}
stop_te_timer(action);
te_action_confirmed(action, controld_globals.transition_graph);
if (pcmk_is_set(action->flags, pcmk__graph_action_failed)) {
abort_transition(action->synapse->priority + 1,
pcmk__graph_restart, "Event failed", event);
}
}
}
if (id == NULL) {
id = "unknown action";
}
uname = crm_element_value(event, PCMK__META_ON_NODE);
if (uname == NULL) {
uname = "unknown node";
}
if (status == PCMK_EXEC_INVALID) {
// We couldn't attempt the action
crm_info("Transition %d action %d (%s on %s): %s",
transition_num, action_num, id, uname,
pcmk_exec_status_str(status));
} else if (desc && update_failcount(event, event_node, rc, target_rc,
(transition_num == -1), FALSE)) {
crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' "
CRM_XS " target-rc=%d rc=%d call-id=%d event='%s'",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(target_rc),
services_ocf_exitcode_str(rc),
target_rc, rc, callid, desc);
} else if (desc) {
crm_info("Transition %d action %d (%s on %s): %s "
CRM_XS " rc=%d target-rc=%d call-id=%d",
transition_num, action_num, id, uname,
desc, rc, target_rc, callid);
} else if (rc == target_rc) {
crm_info("Transition %d action %d (%s on %s) confirmed: %s "
CRM_XS " rc=%d call-id=%d",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(rc), rc, callid);
} else {
update_failcount(event, event_node, rc, target_rc,
(transition_num == -1), ignore_failures);
crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' "
CRM_XS " target-rc=%d rc=%d call-id=%d",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(target_rc),
services_ocf_exitcode_str(rc),
target_rc, rc, callid);
}
bail:
free(update_te_uuid);
}
diff --git a/daemons/fenced/fenced_history.c b/daemons/fenced/fenced_history.c
index fa257b747f..ab0f1a9ad0 100644
--- a/daemons/fenced/fenced_history.c
+++ b/daemons/fenced/fenced_history.c
@@ -1,550 +1,550 @@
/*
* Copyright 2009-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/ipc.h>
#include <crm/common/ipc_internal.h>
#include <crm/cluster/internal.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <pacemaker-fenced.h>
#define MAX_STONITH_HISTORY 500
/*!
* \internal
* \brief Send a broadcast to all nodes to trigger cleanup or
* history synchronisation
*
* \param[in] history Optional history to be attached
* \param[in] callopts We control cleanup via a flag in the callopts
* \param[in] target Cleanup can be limited to certain fence-targets
*/
static void
stonith_send_broadcast_history(xmlNode *history,
int callopts,
const char *target)
{
xmlNode *bcast = create_xml_node(NULL, "stonith_command");
xmlNode *data = create_xml_node(NULL, __func__);
if (target) {
crm_xml_add(data, F_STONITH_TARGET, target);
}
crm_xml_add(bcast, PCMK__XA_T, T_STONITH_NG);
crm_xml_add(bcast, PCMK__XA_SUBT, "broadcast");
crm_xml_add(bcast, F_STONITH_OPERATION, STONITH_OP_FENCE_HISTORY);
crm_xml_add_int(bcast, F_STONITH_CALLOPTS, callopts);
if (history) {
add_node_copy(data, history);
}
add_message_xml(bcast, F_STONITH_CALLDATA, data);
send_cluster_message(NULL, crm_msg_stonith_ng, bcast, FALSE);
free_xml(data);
free_xml(bcast);
}
static gboolean
stonith_remove_history_entry (gpointer key,
gpointer value,
gpointer user_data)
{
remote_fencing_op_t *op = value;
const char *target = (const char *) user_data;
if ((op->state == st_failed) || (op->state == st_done)) {
if ((target) && (strcmp(op->target, target) != 0)) {
return FALSE;
}
return TRUE;
}
return FALSE; /* don't clean pending operations */
}
/*!
* \internal
* \brief Send out a cleanup broadcast or do a local history-cleanup
*
* \param[in] target Cleanup can be limited to certain fence-targets
* \param[in] broadcast Send out a cleanup broadcast
*/
static void
stonith_fence_history_cleanup(const char *target,
gboolean broadcast)
{
if (broadcast) {
stonith_send_broadcast_history(NULL,
st_opt_cleanup | st_opt_discard_reply,
target);
/* we'll do the local clean when we receive back our own broadcast */
} else if (stonith_remote_op_list) {
g_hash_table_foreach_remove(stonith_remote_op_list,
stonith_remove_history_entry,
(gpointer) target);
fenced_send_notification(T_STONITH_NOTIFY_HISTORY, NULL, NULL);
}
}
/* keeping the length of fence-history within bounds
* =================================================
*
* If things are really running wild a lot of fencing-attempts
* might fill up the hash-map, eventually using up a lot
* of memory and creating huge history-sync messages.
* Before the history being synced across nodes at least
* the reboot of a cluster-node helped keeping the
* history within bounds even though not in a reliable
* manner.
*
* stonith_remote_op_list isn't sorted for time-stamps
* thus it would be kind of expensive to delete e.g.
* the oldest entry if it would grow past MAX_STONITH_HISTORY
* entries.
* It is more efficient to purge MAX_STONITH_HISTORY/2
* entries whenever the list grows beyond MAX_STONITH_HISTORY.
* (sort for age + purge the MAX_STONITH_HISTORY/2 oldest)
* That done on a per-node-base might raise the
* probability of large syncs to occur.
* Things like introducing a broadcast to purge
* MAX_STONITH_HISTORY/2 entries or not sync above a certain
* threshold coming to mind ...
* Simplest thing though is to purge the full history
* throughout the cluster once MAX_STONITH_HISTORY is reached.
* On the other hand this leads to purging the history in
* situations where it would be handy to have it probably.
*/
static int
op_time_sort(const void *a_voidp, const void *b_voidp)
{
const remote_fencing_op_t **a = (const remote_fencing_op_t **) a_voidp;
const remote_fencing_op_t **b = (const remote_fencing_op_t **) b_voidp;
gboolean a_pending = ((*a)->state != st_failed) && ((*a)->state != st_done);
gboolean b_pending = ((*b)->state != st_failed) && ((*b)->state != st_done);
if (a_pending && b_pending) {
return 0;
} else if (a_pending) {
return -1;
} else if (b_pending) {
return 1;
} else if ((*b)->completed == (*a)->completed) {
if ((*b)->completed_nsec > (*a)->completed_nsec) {
return 1;
} else if ((*b)->completed_nsec == (*a)->completed_nsec) {
return 0;
}
} else if ((*b)->completed > (*a)->completed) {
return 1;
}
return -1;
}
/*!
* \internal
* \brief Do a local history-trim to MAX_STONITH_HISTORY / 2 entries
* once over MAX_STONITH_HISTORY
*/
void
stonith_fence_history_trim(void)
{
guint num_ops;
if (!stonith_remote_op_list) {
return;
}
num_ops = g_hash_table_size(stonith_remote_op_list);
if (num_ops > MAX_STONITH_HISTORY) {
remote_fencing_op_t *ops[num_ops];
remote_fencing_op_t *op = NULL;
GHashTableIter iter;
int i;
crm_trace("Fencing History growing beyond limit of %d so purge "
"half of failed/successful attempts", MAX_STONITH_HISTORY);
/* write all ops into an array */
i = 0;
g_hash_table_iter_init(&iter, stonith_remote_op_list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&op)) {
ops[i++] = op;
}
/* run quicksort over the array so that we get pending ops
* first and then sorted most recent to oldest
*/
qsort(ops, num_ops, sizeof(remote_fencing_op_t *), op_time_sort);
/* purgest oldest half of the history entries */
for (i = MAX_STONITH_HISTORY / 2; i < num_ops; i++) {
/* keep pending ops even if they shouldn't fill more than
* half of our buffer
*/
if ((ops[i]->state == st_failed) || (ops[i]->state == st_done)) {
g_hash_table_remove(stonith_remote_op_list, ops[i]->id);
}
}
/* we've just purged valid data from the list so there is no need
* to create a notification - if displayed it can stay
*/
}
}
/*!
* \internal
* \brief Convert xml fence-history to a hash-table like stonith_remote_op_list
*
* \param[in] history Fence-history in xml
*
* \return Fence-history as hash-table
*/
static GHashTable *
stonith_xml_history_to_list(const xmlNode *history)
{
xmlNode *xml_op = NULL;
GHashTable *rv = NULL;
init_stonith_remote_op_hash_table(&rv);
CRM_LOG_ASSERT(rv != NULL);
for (xml_op = pcmk__xml_first_child(history); xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
remote_fencing_op_t *op = NULL;
char *id = crm_element_value_copy(xml_op, F_STONITH_REMOTE_OP_ID);
int state;
int exit_status = CRM_EX_OK;
int execution_status = PCMK_EXEC_DONE;
long long completed;
long long completed_nsec = 0L;
if (!id) {
crm_warn("Malformed fencing history received from peer");
continue;
}
crm_trace("Attaching op %s to hashtable", id);
op = calloc(1, sizeof(remote_fencing_op_t));
op->id = id;
op->target = crm_element_value_copy(xml_op, F_STONITH_TARGET);
op->action = crm_element_value_copy(xml_op, F_STONITH_ACTION);
op->originator = crm_element_value_copy(xml_op, F_STONITH_ORIGIN);
op->delegate = crm_element_value_copy(xml_op, F_STONITH_DELEGATE);
op->client_name = crm_element_value_copy(xml_op, F_STONITH_CLIENTNAME);
crm_element_value_ll(xml_op, F_STONITH_DATE, &completed);
op->completed = (time_t) completed;
crm_element_value_ll(xml_op, F_STONITH_DATE_NSEC, &completed_nsec);
op->completed_nsec = completed_nsec;
crm_element_value_int(xml_op, F_STONITH_STATE, &state);
op->state = (enum op_state) state;
/* @COMPAT We can't use stonith__xe_get_result() here because
* fencers <2.1.3 didn't include results, leading it to assume an error
* status. Instead, set an unknown status in that case.
*/
if ((crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &exit_status) < 0)
- || (crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS,
+ || (crm_element_value_int(xml_op, PCMK__XA_OP_STATUS,
&execution_status) < 0)) {
exit_status = CRM_EX_INDETERMINATE;
execution_status = PCMK_EXEC_UNKNOWN;
}
pcmk__set_result(&op->result, exit_status, execution_status,
crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON));
pcmk__set_result_output(&op->result,
crm_element_value_copy(xml_op, F_STONITH_OUTPUT),
NULL);
g_hash_table_replace(rv, id, op);
CRM_LOG_ASSERT(g_hash_table_lookup(rv, id) != NULL);
}
return rv;
}
/*!
* \internal
* \brief Craft xml difference between local fence-history and a history
* coming from remote, and merge the remote history into the local
*
* \param[in,out] remote_history Fence-history as hash-table (may be NULL)
* \param[in] add_id If crafting the answer for an API
* history-request there is no need for the id
* \param[in] target Optionally limit to certain fence-target
*
* \return The fence-history as xml
*/
static xmlNode *
stonith_local_history_diff_and_merge(GHashTable *remote_history,
gboolean add_id, const char *target)
{
xmlNode *history = NULL;
GHashTableIter iter;
remote_fencing_op_t *op = NULL;
gboolean updated = FALSE;
int cnt = 0;
if (stonith_remote_op_list) {
char *id = NULL;
history = create_xml_node(NULL, F_STONITH_HISTORY_LIST);
g_hash_table_iter_init(&iter, stonith_remote_op_list);
while (g_hash_table_iter_next(&iter, (void **)&id, (void **)&op)) {
xmlNode *entry = NULL;
if (remote_history) {
remote_fencing_op_t *remote_op =
g_hash_table_lookup(remote_history, op->id);
if (remote_op) {
if (stonith__op_state_pending(op->state)
&& !stonith__op_state_pending(remote_op->state)) {
crm_debug("Updating outdated pending operation %.8s "
"(state=%s) according to the one (state=%s) from "
"remote peer history",
op->id, stonith_op_state_str(op->state),
stonith_op_state_str(remote_op->state));
g_hash_table_steal(remote_history, op->id);
op->id = remote_op->id;
remote_op->id = id;
g_hash_table_iter_replace(&iter, remote_op);
updated = TRUE;
continue; /* skip outdated entries */
} else if (!stonith__op_state_pending(op->state)
&& stonith__op_state_pending(remote_op->state)) {
crm_debug("Broadcasting operation %.8s (state=%s) to "
"update the outdated pending one "
"(state=%s) in remote peer history",
op->id, stonith_op_state_str(op->state),
stonith_op_state_str(remote_op->state));
g_hash_table_remove(remote_history, op->id);
} else {
g_hash_table_remove(remote_history, op->id);
continue; /* skip entries broadcasted already */
}
}
}
if (!pcmk__str_eq(target, op->target, pcmk__str_null_matches)) {
continue;
}
cnt++;
crm_trace("Attaching op %s", op->id);
entry = create_xml_node(history, STONITH_OP_EXEC);
if (add_id) {
crm_xml_add(entry, F_STONITH_REMOTE_OP_ID, op->id);
}
crm_xml_add(entry, F_STONITH_TARGET, op->target);
crm_xml_add(entry, F_STONITH_ACTION, op->action);
crm_xml_add(entry, F_STONITH_ORIGIN, op->originator);
crm_xml_add(entry, F_STONITH_DELEGATE, op->delegate);
crm_xml_add(entry, F_STONITH_CLIENTNAME, op->client_name);
crm_xml_add_ll(entry, F_STONITH_DATE, op->completed);
crm_xml_add_ll(entry, F_STONITH_DATE_NSEC, op->completed_nsec);
crm_xml_add_int(entry, F_STONITH_STATE, op->state);
stonith__xe_set_result(entry, &op->result);
}
}
if (remote_history) {
init_stonith_remote_op_hash_table(&stonith_remote_op_list);
updated |= g_hash_table_size(remote_history);
g_hash_table_iter_init(&iter, remote_history);
while (g_hash_table_iter_next(&iter, NULL, (void **)&op)) {
if (stonith__op_state_pending(op->state) &&
pcmk__str_eq(op->originator, stonith_our_uname, pcmk__str_casei)) {
crm_warn("Failing pending operation %.8s originated by us but "
"known only from peer history", op->id);
op->state = st_failed;
set_fencing_completed(op);
/* CRM_EX_EXPIRED + PCMK_EXEC_INVALID prevents finalize_op()
* from setting a delegate
*/
pcmk__set_result(&op->result, CRM_EX_EXPIRED, PCMK_EXEC_INVALID,
"Initiated by earlier fencer "
"process and presumed failed");
fenced_broadcast_op_result(op, false);
}
g_hash_table_iter_steal(&iter);
g_hash_table_replace(stonith_remote_op_list, op->id, op);
/* we could trim the history here but if we bail
* out after trim we might miss more recent entries
* of those that might still be in the list
* if we don't bail out trimming once is more
* efficient and memory overhead is minimal as
* we are just moving pointers from one hash to
* another
*/
}
g_hash_table_destroy(remote_history); /* remove what is left */
}
if (updated) {
stonith_fence_history_trim();
fenced_send_notification(T_STONITH_NOTIFY_HISTORY, NULL, NULL);
}
if (cnt == 0) {
free_xml(history);
return NULL;
} else {
return history;
}
}
/*!
* \internal
* \brief Craft xml from the local fence-history
*
* \param[in] add_id If crafting the answer for an API
* history-request there is no need for the id
* \param[in] target Optionally limit to certain fence-target
*
* \return The fence-history as xml
*/
static xmlNode *
stonith_local_history(gboolean add_id, const char *target)
{
return stonith_local_history_diff_and_merge(NULL, add_id, target);
}
/*!
* \internal
* \brief Handle fence-history messages (from API or coming in as broadcasts)
*
* \param[in,out] msg Request XML
* \param[out] output Where to set local history, if requested
* \param[in] remote_peer If broadcast, peer that sent it
* \param[in] options Call options from the request
*/
void
stonith_fence_history(xmlNode *msg, xmlNode **output,
const char *remote_peer, int options)
{
const char *target = NULL;
xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_NEVER);
xmlNode *out_history = NULL;
if (dev) {
target = crm_element_value(dev, F_STONITH_TARGET);
if (target && (options & st_opt_cs_nodeid)) {
int nodeid;
crm_node_t *node;
pcmk__scan_min_int(target, &nodeid, 0);
node = pcmk__search_node_caches(nodeid, NULL,
pcmk__node_search_any
|pcmk__node_search_known);
if (node) {
target = node->uname;
}
}
}
if (options & st_opt_cleanup) {
crm_trace("Cleaning up operations on %s in %p", target,
stonith_remote_op_list);
stonith_fence_history_cleanup(target,
crm_element_value(msg, F_STONITH_CALLID) != NULL);
} else if (options & st_opt_broadcast) {
/* there is no clear sign atm for when a history sync
is done so send a notification for anything
that smells like history-sync
*/
fenced_send_notification(T_STONITH_NOTIFY_HISTORY_SYNCED, NULL, NULL);
if (crm_element_value(msg, F_STONITH_CALLID)) {
/* this is coming from the stonith-API
*
* craft a broadcast with node's history
* so that every node can merge and broadcast
* what it has on top
*/
out_history = stonith_local_history(TRUE, NULL);
crm_trace("Broadcasting history to peers");
stonith_send_broadcast_history(out_history,
st_opt_broadcast | st_opt_discard_reply,
NULL);
} else if (remote_peer &&
!pcmk__str_eq(remote_peer, stonith_our_uname, pcmk__str_casei)) {
xmlNode *history = get_xpath_object("//" F_STONITH_HISTORY_LIST,
msg, LOG_NEVER);
/* either a broadcast created directly upon stonith-API request
* or a diff as response to such a thing
*
* in both cases it may have a history or not
* if we have differential data
* merge in what we've received and stop
* otherwise broadcast what we have on top
* marking as differential and merge in afterwards
*/
if (!history || !pcmk__xe_attr_is_true(history, F_STONITH_DIFFERENTIAL)) {
GHashTable *received_history = NULL;
if (history != NULL) {
received_history = stonith_xml_history_to_list(history);
}
out_history =
stonith_local_history_diff_and_merge(received_history, TRUE, NULL);
if (out_history) {
crm_trace("Broadcasting history-diff to peers");
pcmk__xe_set_bool_attr(out_history, F_STONITH_DIFFERENTIAL, true);
stonith_send_broadcast_history(out_history,
st_opt_broadcast | st_opt_discard_reply,
NULL);
} else {
crm_trace("History-diff is empty - skip broadcast");
}
}
} else {
crm_trace("Skipping history-query-broadcast (%s%s)"
" we sent ourselves",
remote_peer?"remote-peer=":"local-ipc",
remote_peer?remote_peer:"");
}
} else {
/* plain history request */
crm_trace("Looking for operations on %s in %p", target,
stonith_remote_op_list);
*output = stonith_local_history(FALSE, target);
}
free_xml(out_history);
}
diff --git a/daemons/pacemakerd/pcmkd_messages.c b/daemons/pacemakerd/pcmkd_messages.c
index e9a1a23d59..c328e011ea 100644
--- a/daemons/pacemakerd/pcmkd_messages.c
+++ b/daemons/pacemakerd/pcmkd_messages.c
@@ -1,279 +1,279 @@
/*
* Copyright 2010-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include "pacemakerd.h"
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <errno.h>
#include <stdbool.h>
#include <stdint.h>
#include <string.h>
#include <time.h>
#include <sys/types.h>
static GHashTable *pcmkd_handlers = NULL;
static xmlNode *
handle_node_cache_request(pcmk__request_t *request)
{
crm_trace("Ignoring request from client %s to purge node "
"because peer cache is not used",
pcmk__client_name(request->ipc_client));
pcmk__ipc_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags,
"ack", NULL, CRM_EX_OK);
return NULL;
}
static xmlNode *
handle_ping_request(pcmk__request_t *request)
{
xmlNode *msg = request->xml;
const char *value = NULL;
xmlNode *ping = NULL;
xmlNode *reply = NULL;
const char *from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM);
/* Pinged for status */
crm_trace("Pinged from " PCMK__XA_CRM_SYS_FROM "='%s' "
PCMK_XA_ORIGIN "='%s'",
pcmk__s(from, ""),
pcmk__s(crm_element_value(msg, PCMK_XA_ORIGIN), ""));
pcmk__ipc_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags,
"ack", NULL, CRM_EX_INDETERMINATE);
ping = create_xml_node(NULL, XML_CRM_TAG_PING);
value = crm_element_value(msg, PCMK__XA_CRM_SYS_TO);
crm_xml_add(ping, PCMK__XA_CRM_SUBSYSTEM, value);
crm_xml_add(ping, PCMK__XA_PACEMAKERD_STATE, pacemakerd_state);
crm_xml_add_ll(ping, PCMK_XA_CRM_TIMESTAMP,
(long long) subdaemon_check_progress);
crm_xml_add(ping, PCMK_XA_RESULT, "ok");
reply = create_reply(msg, ping);
free_xml(ping);
if (reply == NULL) {
pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR,
"Failed building ping reply for client %s",
pcmk__client_name(request->ipc_client));
} else {
pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL);
}
/* just proceed state on sbd pinging us */
if (from && strstr(from, "sbd")) {
if (pcmk__str_eq(pacemakerd_state, XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE, pcmk__str_none)) {
if (pcmk__get_sbd_sync_resource_startup()) {
crm_notice("Shutdown-complete-state passed to SBD.");
}
shutdown_complete_state_reported_to = request->ipc_client->pid;
} else if (pcmk__str_eq(pacemakerd_state, XML_PING_ATTR_PACEMAKERDSTATE_WAITPING, pcmk__str_none)) {
crm_notice("Received startup-trigger from SBD.");
pacemakerd_state = XML_PING_ATTR_PACEMAKERDSTATE_STARTINGDAEMONS;
mainloop_set_trigger(startup_trigger);
}
}
return reply;
}
static xmlNode *
handle_shutdown_request(pcmk__request_t *request)
{
xmlNode *msg = request->xml;
xmlNode *shutdown = NULL;
xmlNode *reply = NULL;
/* Only allow privileged users (i.e. root or hacluster) to shut down
* Pacemaker from the command line (or direct IPC), so that other users
* are forced to go through the CIB and have ACLs applied.
*/
bool allowed = pcmk_is_set(request->ipc_client->flags, pcmk__client_privileged);
pcmk__ipc_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags,
"ack", NULL, CRM_EX_INDETERMINATE);
shutdown = create_xml_node(NULL, XML_CIB_ATTR_SHUTDOWN);
if (allowed) {
crm_notice("Shutting down in response to IPC request %s from %s",
crm_element_value(msg, PCMK_XA_REFERENCE),
crm_element_value(msg, PCMK_XA_ORIGIN));
- crm_xml_add_int(shutdown, XML_LRM_ATTR_OPSTATUS, CRM_EX_OK);
+ crm_xml_add_int(shutdown, PCMK__XA_OP_STATUS, CRM_EX_OK);
} else {
crm_warn("Ignoring shutdown request from unprivileged client %s",
pcmk__client_name(request->ipc_client));
- crm_xml_add_int(shutdown, XML_LRM_ATTR_OPSTATUS, CRM_EX_INSUFFICIENT_PRIV);
+ crm_xml_add_int(shutdown, PCMK__XA_OP_STATUS, CRM_EX_INSUFFICIENT_PRIV);
}
reply = create_reply(msg, shutdown);
free_xml(shutdown);
if (reply == NULL) {
pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR,
"Failed building shutdown reply for client %s",
pcmk__client_name(request->ipc_client));
} else {
pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL);
}
if (allowed) {
pcmk_shutdown(15);
}
return reply;
}
static xmlNode *
handle_unknown_request(pcmk__request_t *request)
{
pcmk__ipc_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags,
"ack", NULL, CRM_EX_INVALID_PARAM);
pcmk__format_result(&request->result, CRM_EX_PROTOCOL, PCMK_EXEC_INVALID,
"Unknown IPC request type '%s' (bug?)",
pcmk__client_name(request->ipc_client));
return NULL;
}
static void
pcmkd_register_handlers(void)
{
pcmk__server_command_t handlers[] = {
{ CRM_OP_RM_NODE_CACHE, handle_node_cache_request },
{ CRM_OP_PING, handle_ping_request },
{ CRM_OP_QUIT, handle_shutdown_request },
{ NULL, handle_unknown_request },
};
pcmkd_handlers = pcmk__register_handlers(handlers);
}
static int32_t
pcmk_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
/* Error code means? */
static int32_t
pcmk_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p", c);
if (shutdown_complete_state_reported_to == client->pid) {
shutdown_complete_state_reported_client_closed = TRUE;
if (shutdown_trigger) {
mainloop_set_trigger(shutdown_trigger);
}
}
pcmk__free_client(client);
return 0;
}
static void
pcmk_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
pcmk_ipc_closed(c);
}
/* Exit code means? */
static int32_t
pcmk_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
xmlNode *msg = NULL;
pcmk__client_t *c = pcmk__find_client(qbc);
CRM_CHECK(c != NULL, return 0);
if (pcmkd_handlers == NULL) {
pcmkd_register_handlers();
}
msg = pcmk__client_data2xml(c, data, &id, &flags);
if (msg == NULL) {
pcmk__ipc_send_ack(c, id, flags, "ack", NULL, CRM_EX_PROTOCOL);
return 0;
} else {
char *log_msg = NULL;
const char *reason = NULL;
xmlNode *reply = NULL;
pcmk__request_t request = {
.ipc_client = c,
.ipc_id = id,
.ipc_flags = flags,
.peer = NULL,
.xml = msg,
.call_options = 0,
.result = PCMK__UNKNOWN_RESULT,
};
request.op = crm_element_value_copy(request.xml, PCMK__XA_CRM_TASK);
CRM_CHECK(request.op != NULL, return 0);
reply = pcmk__process_request(&request, pcmkd_handlers);
if (reply != NULL) {
pcmk__ipc_send_xml(c, id, reply, crm_ipc_server_event);
free_xml(reply);
}
reason = request.result.exit_reason;
log_msg = crm_strdup_printf("Processed %s request from %s %s: %s%s%s%s",
request.op, pcmk__request_origin_type(&request),
pcmk__request_origin(&request),
pcmk_exec_status_str(request.result.execution_status),
(reason == NULL)? "" : " (",
(reason == NULL)? "" : reason,
(reason == NULL)? "" : ")");
if (!pcmk__result_ok(&request.result)) {
crm_warn("%s", log_msg);
} else {
crm_debug("%s", log_msg);
}
free(log_msg);
pcmk__reset_request(&request);
}
free_xml(msg);
return 0;
}
struct qb_ipcs_service_handlers pacemakerd_ipc_callbacks = {
.connection_accept = pcmk_ipc_accept,
.connection_created = NULL,
.msg_process = pcmk_ipc_dispatch,
.connection_closed = pcmk_ipc_closed,
.connection_destroyed = pcmk_ipc_destroy
};
diff --git a/include/crm_internal.h b/include/crm_internal.h
index 9249095a8f..53b6621468 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -1,193 +1,194 @@
/*
* Copyright 2006-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef CRM_INTERNAL__H
# define CRM_INTERNAL__H
# ifndef PCMK__CONFIG_H
# define PCMK__CONFIG_H
# include <config.h>
# endif
# include <portability.h>
/* Our minimum glib dependency is 2.42. Define that as both the minimum and
* maximum glib APIs that are allowed (i.e. APIs that were already deprecated
* in 2.42, and APIs introduced after 2.42, cannot be used by Pacemaker code).
*/
#define GLIB_VERSION_MIN_REQUIRED GLIB_VERSION_2_42
#define GLIB_VERSION_MAX_ALLOWED GLIB_VERSION_2_42
# include <glib.h>
# include <stdbool.h>
# include <libxml/tree.h>
/* Public API headers can guard including deprecated API headers with this
* symbol, thus preventing internal code (which includes this header) from using
* deprecated APIs, while still allowing external code to use them by default.
*/
#define PCMK_ALLOW_DEPRECATED 0
# include <crm/lrmd.h>
# include <crm/common/logging.h>
# include <crm/common/logging_internal.h>
# include <crm/common/ipc_internal.h>
# include <crm/common/options_internal.h>
# include <crm/common/output_internal.h>
# include <crm/common/xml_internal.h>
# include <crm/common/internal.h>
# include <locale.h>
# include <gettext.h>
#define N_(String) (String)
#ifdef ENABLE_NLS
# define _(String) gettext(String)
#else
# define _(String) (String)
#endif
/*
* XML element names used only by internal code
*/
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_ADDED "diff-added"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_REMOVED "diff-removed"
/* @COMPAT Deprecated since 2.0.0; alias for <clone> with PCMK_META_PROMOTABLE
* set to "true"
*/
#define PCMK__XE_PROMOTABLE_LEGACY "master"
/*
* XML attribute names used only by internal code
*/
#define PCMK__XA_ATTR_DAMPENING "attr_dampening"
#define PCMK__XA_ATTR_FORCE "attrd_is_force_write"
#define PCMK__XA_ATTR_INTERVAL "attr_clear_interval"
#define PCMK__XA_ATTR_IS_PRIVATE "attr_is_private"
#define PCMK__XA_ATTR_IS_REMOTE "attr_is_remote"
#define PCMK__XA_ATTR_NAME "attr_name"
#define PCMK__XA_ATTR_NODE_ID "attr_host_id"
#define PCMK__XA_ATTR_NODE_NAME "attr_host"
#define PCMK__XA_ATTR_OPERATION "attr_clear_operation"
#define PCMK__XA_ATTR_PATTERN "attr_regex"
#define PCMK__XA_ATTR_RESOURCE "attr_resource"
#define PCMK__XA_ATTR_SECTION "attr_section"
#define PCMK__XA_ATTR_SET "attr_set"
#define PCMK__XA_ATTR_SET_TYPE "attr_set_type"
#define PCMK__XA_ATTR_SYNC_POINT "attr_sync_point"
#define PCMK__XA_ATTR_USER "attr_user"
#define PCMK__XA_ATTR_UUID "attr_key"
#define PCMK__XA_ATTR_VALUE "attr_value"
#define PCMK__XA_ATTR_VERSION "attr_version"
#define PCMK__XA_ATTR_WRITER "attr_writer"
#define PCMK__XA_CONFIG_ERRORS "config-errors"
#define PCMK__XA_CONFIG_WARNINGS "config-warnings"
#define PCMK__XA_CONFIRM "confirm"
#define PCMK__XA_CONN_HOST "connection_host"
#define PCMK__XA_CRMD "crmd"
#define PCMK__XA_CRMD_STATE "crmd_state"
#define PCMK__XA_CRM_HOST_TO "crm_host_to"
#define PCMK__XA_CRM_LIMIT_MAX "crm-limit-max"
#define PCMK__XA_CRM_LIMIT_MODE "crm-limit-mode"
#define PCMK__XA_CRM_SUBSYSTEM "crm_subsystem"
#define PCMK__XA_CRM_SYS_FROM "crm_sys_from"
#define PCMK__XA_CRM_SYS_TO "crm_sys_to"
#define PCMK__XA_CRM_TASK "crm_task"
#define PCMK__XA_CRM_TGRAPH_IN "crm-tgraph-in"
#define PCMK__XA_CRM_USER "crm_user"
#define PCMK__XA_DC_LEAVING "dc-leaving"
#define PCMK__XA_DIGEST "digest"
#define PCMK__XA_ELECTION_AGE_SEC "election-age-sec"
#define PCMK__XA_ELECTION_AGE_NANO_SEC "election-age-nano-sec"
#define PCMK__XA_ELECTION_ID "election-id"
#define PCMK__XA_ELECTION_OWNER "election-owner"
#define PCMK__XA_EXPECTED "expected"
#define PCMK__XA_FILE "file"
#define PCMK__XA_GRAPH_ERRORS "graph-errors"
#define PCMK__XA_GRAPH_WARNINGS "graph-warnings"
#define PCMK__XA_IN_CCM "in_ccm"
#define PCMK__XA_JOIN "join"
#define PCMK__XA_JOIN_ID "join_id"
#define PCMK__XA_LONG_ID "long-id"
#define PCMK__XA_MODE "mode"
#define PCMK__XA_NODE_START_STATE "node_start_state"
#define PCMK__XA_OBJECT_TYPE "object_type"
#define PCMK__XA_OPERATION_KEY "operation_key"
+#define PCMK__XA_OP_STATUS "op-status"
#define PCMK__XA_PACEMAKERD_STATE "pacemakerd_state"
#define PCMK__XA_PRIORITY "priority"
#define PCMK__XA_REAP "reap"
/* Actions to be executed on Pacemaker Remote nodes are routed through the
* controller on the cluster node hosting the remote connection. That cluster
* node is considered the router node for the action.
*/
#define PCMK__XA_ROUTER_NODE "router_node"
#define PCMK__XA_RSC_ID "rsc-id"
#define PCMK__XA_SCHEMA "schema"
#define PCMK__XA_SCHEMAS "schemas"
#define PCMK__XA_SRC "src"
#define PCMK__XA_SUBT "subt" // subtype
#define PCMK__XA_T "t" // type
#define PCMK__XA_TASK "task"
#define PCMK__XA_TRANSITION_KEY "transition-key"
#define PCMK__XA_TRANSITION_MAGIC "transition-magic"
#define PCMK__XA_UPTIME "uptime"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_FIRST_INSTANCE "first-instance"
// @COMPAT Deprecated since 2.1.6
#define PCMK__XA_REPLACE "replace"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_RSC_INSTANCE "rsc-instance"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_THEN_INSTANCE "then-instance"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_WITH_RSC_INSTANCE "with-rsc-instance"
/*
* IPC service names that are only used internally
*/
# define PCMK__SERVER_BASED_RO "cib_ro"
# define PCMK__SERVER_BASED_RW "cib_rw"
# define PCMK__SERVER_BASED_SHM "cib_shm"
/*
* IPC commands that can be sent to Pacemaker daemons
*/
#define PCMK__ATTRD_CMD_PEER_REMOVE "peer-remove"
#define PCMK__ATTRD_CMD_UPDATE "update"
#define PCMK__ATTRD_CMD_UPDATE_BOTH "update-both"
#define PCMK__ATTRD_CMD_UPDATE_DELAY "update-delay"
#define PCMK__ATTRD_CMD_QUERY "query"
#define PCMK__ATTRD_CMD_REFRESH "refresh"
#define PCMK__ATTRD_CMD_FLUSH "flush"
#define PCMK__ATTRD_CMD_SYNC "sync"
#define PCMK__ATTRD_CMD_SYNC_RESPONSE "sync-response"
#define PCMK__ATTRD_CMD_CLEAR_FAILURE "clear-failure"
#define PCMK__ATTRD_CMD_CONFIRM "confirm"
#define PCMK__CONTROLD_CMD_NODES "list-nodes"
#endif /* CRM_INTERNAL__H */
diff --git a/lib/common/actions.c b/lib/common/actions.c
index 0f044a5eac..8e786b904f 100644
--- a/lib/common/actions.c
+++ b/lib/common/actions.c
@@ -1,532 +1,532 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <sys/types.h>
#include <ctype.h>
#include <crm/crm.h>
#include <crm/lrmd.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
/*!
* \brief Generate an operation key (RESOURCE_ACTION_INTERVAL)
*
* \param[in] rsc_id ID of resource being operated on
* \param[in] op_type Operation name
* \param[in] interval_ms Operation interval
*
* \return Newly allocated memory containing operation key as string
*
* \note This function asserts on errors, so it will never return NULL.
* The caller is responsible for freeing the result with free().
*/
char *
pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms)
{
CRM_ASSERT(rsc_id != NULL);
CRM_ASSERT(op_type != NULL);
return crm_strdup_printf(PCMK__OP_FMT, rsc_id, op_type, interval_ms);
}
static inline gboolean
convert_interval(const char *s, guint *interval_ms)
{
unsigned long l;
errno = 0;
l = strtoul(s, NULL, 10);
if (errno != 0) {
return FALSE;
}
*interval_ms = (guint) l;
return TRUE;
}
/*!
* \internal
* \brief Check for underbar-separated substring match
*
* \param[in] key Overall string being checked
* \param[in] position Match before underbar at this \p key index
* \param[in] matches Substrings to match (may contain underbars)
*
* \return \p key index of underbar before any matching substring,
* or 0 if none
*/
static size_t
match_before(const char *key, size_t position, const char **matches)
{
for (int i = 0; matches[i] != NULL; ++i) {
const size_t match_len = strlen(matches[i]);
// Must have at least X_MATCH before position
if (position > (match_len + 1)) {
const size_t possible = position - match_len - 1;
if ((key[possible] == '_')
&& (strncmp(key + possible + 1, matches[i], match_len) == 0)) {
return possible;
}
}
}
return 0;
}
gboolean
parse_op_key(const char *key, char **rsc_id, char **op_type, guint *interval_ms)
{
guint local_interval_ms = 0;
const size_t key_len = (key == NULL)? 0 : strlen(key);
// Operation keys must be formatted as RSC_ACTION_INTERVAL
size_t action_underbar = 0; // Index in key of underbar before ACTION
size_t interval_underbar = 0; // Index in key of underbar before INTERVAL
size_t possible = 0;
/* Underbar was a poor choice of separator since both RSC and ACTION can
* contain underbars. Here, list action names and name prefixes that can.
*/
const char *actions_with_underbars[] = {
PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_MIGRATE_TO,
NULL
};
const char *action_prefixes_with_underbars[] = {
"pre_" PCMK_ACTION_NOTIFY,
"post_" PCMK_ACTION_NOTIFY,
"confirmed-pre_" PCMK_ACTION_NOTIFY,
"confirmed-post_" PCMK_ACTION_NOTIFY,
NULL,
};
// Initialize output variables in case of early return
if (rsc_id) {
*rsc_id = NULL;
}
if (op_type) {
*op_type = NULL;
}
if (interval_ms) {
*interval_ms = 0;
}
// RSC_ACTION_INTERVAL implies a minimum of 5 characters
if (key_len < 5) {
return FALSE;
}
// Find, parse, and validate interval
interval_underbar = key_len - 2;
while ((interval_underbar > 2) && (key[interval_underbar] != '_')) {
--interval_underbar;
}
if ((interval_underbar == 2)
|| !convert_interval(key + interval_underbar + 1, &local_interval_ms)) {
return FALSE;
}
// Find the base (OCF) action name, disregarding prefixes
action_underbar = match_before(key, interval_underbar,
actions_with_underbars);
if (action_underbar == 0) {
action_underbar = interval_underbar - 2;
while ((action_underbar > 0) && (key[action_underbar] != '_')) {
--action_underbar;
}
if (action_underbar == 0) {
return FALSE;
}
}
possible = match_before(key, action_underbar,
action_prefixes_with_underbars);
if (possible != 0) {
action_underbar = possible;
}
// Set output variables
if (rsc_id != NULL) {
*rsc_id = strndup(key, action_underbar);
CRM_ASSERT(*rsc_id != NULL);
}
if (op_type != NULL) {
*op_type = strndup(key + action_underbar + 1,
interval_underbar - action_underbar - 1);
CRM_ASSERT(*op_type != NULL);
}
if (interval_ms != NULL) {
*interval_ms = local_interval_ms;
}
return TRUE;
}
char *
pcmk__notify_key(const char *rsc_id, const char *notify_type,
const char *op_type)
{
CRM_CHECK(rsc_id != NULL, return NULL);
CRM_CHECK(op_type != NULL, return NULL);
CRM_CHECK(notify_type != NULL, return NULL);
return crm_strdup_printf("%s_%s_notify_%s_0",
rsc_id, notify_type, op_type);
}
/*!
* \brief Parse a transition magic string into its constituent parts
*
* \param[in] magic Magic string to parse (must be non-NULL)
* \param[out] uuid If non-NULL, where to store copy of parsed UUID
* \param[out] transition_id If non-NULL, where to store parsed transition ID
* \param[out] action_id If non-NULL, where to store parsed action ID
* \param[out] op_status If non-NULL, where to store parsed result status
* \param[out] op_rc If non-NULL, where to store parsed actual rc
* \param[out] target_rc If non-NULL, where to stored parsed target rc
*
* \return TRUE if key was valid, FALSE otherwise
* \note If uuid is supplied and this returns TRUE, the caller is responsible
* for freeing the memory for *uuid using free().
*/
gboolean
decode_transition_magic(const char *magic, char **uuid, int *transition_id, int *action_id,
int *op_status, int *op_rc, int *target_rc)
{
int res = 0;
char *key = NULL;
gboolean result = TRUE;
int local_op_status = -1;
int local_op_rc = -1;
CRM_CHECK(magic != NULL, return FALSE);
#ifdef HAVE_SSCANF_M
res = sscanf(magic, "%d:%d;%ms", &local_op_status, &local_op_rc, &key);
#else
key = calloc(1, strlen(magic) - 3); // magic must have >=4 other characters
CRM_ASSERT(key);
res = sscanf(magic, "%d:%d;%s", &local_op_status, &local_op_rc, key);
#endif
if (res == EOF) {
crm_err("Could not decode transition information '%s': %s",
magic, pcmk_rc_str(errno));
result = FALSE;
} else if (res < 3) {
crm_warn("Transition information '%s' incomplete (%d of 3 expected items)",
magic, res);
result = FALSE;
} else {
if (op_status) {
*op_status = local_op_status;
}
if (op_rc) {
*op_rc = local_op_rc;
}
result = decode_transition_key(key, uuid, transition_id, action_id,
target_rc);
}
free(key);
return result;
}
char *
pcmk__transition_key(int transition_id, int action_id, int target_rc,
const char *node)
{
CRM_CHECK(node != NULL, return NULL);
return crm_strdup_printf("%d:%d:%d:%-*s",
action_id, transition_id, target_rc, 36, node);
}
/*!
* \brief Parse a transition key into its constituent parts
*
* \param[in] key Transition key to parse (must be non-NULL)
* \param[out] uuid If non-NULL, where to store copy of parsed UUID
* \param[out] transition_id If non-NULL, where to store parsed transition ID
* \param[out] action_id If non-NULL, where to store parsed action ID
* \param[out] target_rc If non-NULL, where to stored parsed target rc
*
* \return TRUE if key was valid, FALSE otherwise
* \note If uuid is supplied and this returns TRUE, the caller is responsible
* for freeing the memory for *uuid using free().
*/
gboolean
decode_transition_key(const char *key, char **uuid, int *transition_id, int *action_id,
int *target_rc)
{
int local_transition_id = -1;
int local_action_id = -1;
int local_target_rc = -1;
char local_uuid[37] = { '\0' };
// Initialize any supplied output arguments
if (uuid) {
*uuid = NULL;
}
if (transition_id) {
*transition_id = -1;
}
if (action_id) {
*action_id = -1;
}
if (target_rc) {
*target_rc = -1;
}
CRM_CHECK(key != NULL, return FALSE);
if (sscanf(key, "%d:%d:%d:%36s", &local_action_id, &local_transition_id,
&local_target_rc, local_uuid) != 4) {
crm_err("Invalid transition key '%s'", key);
return FALSE;
}
if (strlen(local_uuid) != 36) {
crm_warn("Invalid UUID '%s' in transition key '%s'", local_uuid, key);
}
if (uuid) {
*uuid = strdup(local_uuid);
CRM_ASSERT(*uuid);
}
if (transition_id) {
*transition_id = local_transition_id;
}
if (action_id) {
*action_id = local_action_id;
}
if (target_rc) {
*target_rc = local_target_rc;
}
return TRUE;
}
// Return true if a is an attribute that should be filtered
static bool
should_filter_for_digest(xmlAttrPtr a, void *user_data)
{
if (strncmp((const char *) a->name, CRM_META "_",
sizeof(CRM_META " ") - 1) == 0) {
return true;
}
return pcmk__str_any_of((const char *) a->name,
PCMK_XA_ID,
PCMK_XA_CRM_FEATURE_SET,
XML_LRM_ATTR_OP_DIGEST,
PCMK__META_ON_NODE,
PCMK__META_ON_NODE_UUID,
"pcmk_external_ip",
NULL);
}
/*!
* \internal
* \brief Remove XML attributes not needed for operation digest
*
* \param[in,out] param_set XML with operation parameters
*/
void
pcmk__filter_op_for_digest(xmlNode *param_set)
{
char *key = NULL;
char *timeout = NULL;
guint interval_ms = 0;
if (param_set == NULL) {
return;
}
/* Timeout is useful for recurring operation digests, so grab it before
* removing meta-attributes
*/
key = crm_meta_name(PCMK_META_INTERVAL);
if (crm_element_value_ms(param_set, key, &interval_ms) != pcmk_ok) {
interval_ms = 0;
}
free(key);
key = NULL;
if (interval_ms != 0) {
key = crm_meta_name(PCMK_META_TIMEOUT);
timeout = crm_element_value_copy(param_set, key);
}
// Remove all CRM_meta_* attributes and certain other attributes
pcmk__xe_remove_matching_attrs(param_set, should_filter_for_digest, NULL);
// Add timeout back for recurring operation digests
if (timeout != NULL) {
crm_xml_add(param_set, key, timeout);
}
free(timeout);
free(key);
}
int
rsc_op_expected_rc(const lrmd_event_data_t *op)
{
int rc = 0;
if (op && op->user_data) {
decode_transition_key(op->user_data, NULL, NULL, NULL, &rc);
}
return rc;
}
gboolean
did_rsc_op_fail(lrmd_event_data_t * op, int target_rc)
{
switch (op->op_status) {
case PCMK_EXEC_CANCELLED:
case PCMK_EXEC_PENDING:
return FALSE;
case PCMK_EXEC_NOT_SUPPORTED:
case PCMK_EXEC_TIMEOUT:
case PCMK_EXEC_ERROR:
case PCMK_EXEC_NOT_CONNECTED:
case PCMK_EXEC_NO_FENCE_DEVICE:
case PCMK_EXEC_NO_SECRETS:
case PCMK_EXEC_INVALID:
return TRUE;
default:
if (target_rc != op->rc) {
return TRUE;
}
}
return FALSE;
}
/*!
* \brief Create a CIB XML element for an operation
*
* \param[in,out] parent If not NULL, make new XML node a child of this
* \param[in] prefix Generate an ID using this prefix
* \param[in] task Operation task to set
* \param[in] interval_spec Operation interval to set
* \param[in] timeout If not NULL, operation timeout to set
*
* \return New XML object on success, NULL otherwise
*/
xmlNode *
crm_create_op_xml(xmlNode *parent, const char *prefix, const char *task,
const char *interval_spec, const char *timeout)
{
xmlNode *xml_op;
CRM_CHECK(prefix && task && interval_spec, return NULL);
xml_op = create_xml_node(parent, PCMK_XE_OP);
crm_xml_set_id(xml_op, "%s-%s-%s", prefix, task, interval_spec);
crm_xml_add(xml_op, PCMK_META_INTERVAL, interval_spec);
crm_xml_add(xml_op, PCMK_XA_NAME, task);
if (timeout) {
crm_xml_add(xml_op, PCMK_META_TIMEOUT, timeout);
}
return xml_op;
}
/*!
* \brief Check whether an operation requires resource agent meta-data
*
* \param[in] rsc_class Resource agent class (or NULL to skip class check)
* \param[in] op Operation action (or NULL to skip op check)
*
* \return true if operation needs meta-data, false otherwise
* \note At least one of rsc_class and op must be specified.
*/
bool
crm_op_needs_metadata(const char *rsc_class, const char *op)
{
/* Agent metadata is used to determine whether an agent reload is possible,
* so if this op is not relevant to that feature, we don't need metadata.
*/
CRM_CHECK((rsc_class != NULL) || (op != NULL), return false);
if ((rsc_class != NULL)
&& !pcmk_is_set(pcmk_get_ra_caps(rsc_class), pcmk_ra_cap_params)) {
// Metadata is needed only for resource classes that use parameters
return false;
}
if (op == NULL) {
return true;
}
// Metadata is needed only for these actions
return pcmk__str_any_of(op, PCMK_ACTION_START, PCMK_ACTION_MONITOR,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_NOTIFY, NULL);
}
/*!
* \internal
* \brief Check whether an action name is for a fencing action
*
* \param[in] action Action name to check
*
* \return true if \p action is "off", "reboot", or "poweroff", otherwise false
*/
bool
pcmk__is_fencing_action(const char *action)
{
return pcmk__str_any_of(action, PCMK_ACTION_OFF, PCMK_ACTION_REBOOT,
"poweroff", NULL);
}
bool
pcmk_is_probe(const char *task, guint interval)
{
if (task == NULL) {
return false;
}
return (interval == 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_none);
}
bool
pcmk_xe_is_probe(const xmlNode *xml_op)
{
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
const char *interval_ms_s = crm_element_value(xml_op, PCMK_META_INTERVAL);
int interval_ms;
pcmk__scan_min_int(interval_ms_s, &interval_ms, 0);
return pcmk_is_probe(task, interval_ms);
}
bool
pcmk_xe_mask_probe_failure(const xmlNode *xml_op)
{
int status = PCMK_EXEC_UNKNOWN;
int rc = PCMK_OCF_OK;
if (!pcmk_xe_is_probe(xml_op)) {
return false;
}
- crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status);
+ crm_element_value_int(xml_op, PCMK__XA_OP_STATUS, &status);
crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc);
return rc == PCMK_OCF_NOT_INSTALLED || rc == PCMK_OCF_INVALID_PARAM ||
status == PCMK_EXEC_NOT_INSTALLED;
}
diff --git a/lib/common/ipc_pacemakerd.c b/lib/common/ipc_pacemakerd.c
index bd738345d1..3f2cc9798c 100644
--- a/lib/common/ipc_pacemakerd.c
+++ b/lib/common/ipc_pacemakerd.c
@@ -1,316 +1,318 @@
/*
* Copyright 2020-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdlib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/ipc.h>
#include <crm/common/ipc_internal.h>
#include <crm/common/ipc_pacemakerd.h>
#include "crmcommon_private.h"
typedef struct pacemakerd_api_private_s {
enum pcmk_pacemakerd_state state;
char *client_uuid;
} pacemakerd_api_private_t;
static const char *pacemakerd_state_str[] = {
XML_PING_ATTR_PACEMAKERDSTATE_INIT,
XML_PING_ATTR_PACEMAKERDSTATE_STARTINGDAEMONS,
XML_PING_ATTR_PACEMAKERDSTATE_WAITPING,
XML_PING_ATTR_PACEMAKERDSTATE_RUNNING,
XML_PING_ATTR_PACEMAKERDSTATE_SHUTTINGDOWN,
XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE,
XML_PING_ATTR_PACEMAKERDSTATE_REMOTE,
};
enum pcmk_pacemakerd_state
pcmk_pacemakerd_api_daemon_state_text2enum(const char *state)
{
int i;
if (state == NULL) {
return pcmk_pacemakerd_state_invalid;
}
for (i=pcmk_pacemakerd_state_init; i <= pcmk_pacemakerd_state_max;
i++) {
if (pcmk__str_eq(state, pacemakerd_state_str[i], pcmk__str_none)) {
return i;
}
}
return pcmk_pacemakerd_state_invalid;
}
const char *
pcmk_pacemakerd_api_daemon_state_enum2text(
enum pcmk_pacemakerd_state state)
{
if ((state >= pcmk_pacemakerd_state_init) &&
(state <= pcmk_pacemakerd_state_max)) {
return pacemakerd_state_str[state];
}
return "invalid";
}
/*!
* \internal
* \brief Return a friendly string representation of a \p pacemakerd state
*
* \param[in] state \p pacemakerd state
*
* \return A user-friendly string representation of \p state, or
* <tt>"Invalid pacemakerd state"</tt>
*/
const char *
pcmk__pcmkd_state_enum2friendly(enum pcmk_pacemakerd_state state)
{
switch (state) {
case pcmk_pacemakerd_state_init:
return "Initializing pacemaker";
case pcmk_pacemakerd_state_starting_daemons:
return "Pacemaker daemons are starting";
case pcmk_pacemakerd_state_wait_for_ping:
return "Waiting for startup trigger from SBD";
case pcmk_pacemakerd_state_running:
return "Pacemaker is running";
case pcmk_pacemakerd_state_shutting_down:
return "Pacemaker daemons are shutting down";
case pcmk_pacemakerd_state_shutdown_complete:
/* Assuming pacemakerd won't process messages while in
* shutdown_complete state unless reporting to SBD
*/
return "Pacemaker daemons are shut down (reporting to SBD)";
case pcmk_pacemakerd_state_remote:
return "pacemaker-remoted is running (on a Pacemaker Remote node)";
default:
return "Invalid pacemakerd state";
}
}
/*!
* \internal
* \brief Get a string representation of a \p pacemakerd API reply type
*
* \param[in] reply \p pacemakerd API reply type
*
* \return String representation of a \p pacemakerd API reply type
*/
const char *
pcmk__pcmkd_api_reply2str(enum pcmk_pacemakerd_api_reply reply)
{
switch (reply) {
case pcmk_pacemakerd_reply_ping:
return "ping";
case pcmk_pacemakerd_reply_shutdown:
return "shutdown";
default:
return "unknown";
}
}
// \return Standard Pacemaker return code
static int
new_data(pcmk_ipc_api_t *api)
{
struct pacemakerd_api_private_s *private = NULL;
api->api_data = calloc(1, sizeof(struct pacemakerd_api_private_s));
if (api->api_data == NULL) {
return errno;
}
private = api->api_data;
private->state = pcmk_pacemakerd_state_invalid;
/* other as with cib, controld, ... we are addressing pacemakerd just
from the local node -> pid is unique and thus sufficient as an ID
*/
private->client_uuid = pcmk__getpid_s();
return pcmk_rc_ok;
}
static void
free_data(void *data)
{
free(((struct pacemakerd_api_private_s *) data)->client_uuid);
free(data);
}
// \return Standard Pacemaker return code
static int
post_connect(pcmk_ipc_api_t *api)
{
struct pacemakerd_api_private_s *private = NULL;
if (api->api_data == NULL) {
return EINVAL;
}
private = api->api_data;
private->state = pcmk_pacemakerd_state_invalid;
return pcmk_rc_ok;
}
static void
post_disconnect(pcmk_ipc_api_t *api)
{
struct pacemakerd_api_private_s *private = NULL;
if (api->api_data == NULL) {
return;
}
private = api->api_data;
private->state = pcmk_pacemakerd_state_invalid;
return;
}
static bool
reply_expected(pcmk_ipc_api_t *api, const xmlNode *request)
{
const char *command = crm_element_value(request, PCMK__XA_CRM_TASK);
if (command == NULL) {
return false;
}
// We only need to handle commands that functions in this file can send
return pcmk__str_any_of(command, CRM_OP_PING, CRM_OP_QUIT, NULL);
}
static bool
dispatch(pcmk_ipc_api_t *api, xmlNode *reply)
{
crm_exit_t status = CRM_EX_OK;
xmlNode *msg_data = NULL;
pcmk_pacemakerd_api_reply_t reply_data = {
pcmk_pacemakerd_reply_unknown
};
const char *value = NULL;
long long value_ll = 0;
if (pcmk__str_eq((const char *) reply->name, "ack", pcmk__str_none)) {
long long int ack_status = 0;
pcmk__scan_ll(crm_element_value(reply, "status"), &ack_status, CRM_EX_OK);
return ack_status == CRM_EX_INDETERMINATE;
}
value = crm_element_value(reply, PCMK__XA_SUBT);
if (!pcmk__str_eq(value, PCMK__VALUE_RESPONSE, pcmk__str_none)) {
crm_info("Unrecognizable message from %s: "
"message type '%s' not '" PCMK__VALUE_RESPONSE "'",
pcmk_ipc_name(api, true), pcmk__s(value, ""));
status = CRM_EX_PROTOCOL;
goto done;
}
if (pcmk__str_empty(crm_element_value(reply, PCMK_XA_REFERENCE))) {
crm_info("Unrecognizable message from %s: no reference",
pcmk_ipc_name(api, true));
status = CRM_EX_PROTOCOL;
goto done;
}
value = crm_element_value(reply, PCMK__XA_CRM_TASK);
// Parse useful info from reply
msg_data = get_message_xml(reply, F_CRM_DATA);
crm_element_value_ll(msg_data, PCMK_XA_CRM_TIMESTAMP, &value_ll);
if (pcmk__str_eq(value, CRM_OP_PING, pcmk__str_none)) {
reply_data.reply_type = pcmk_pacemakerd_reply_ping;
reply_data.data.ping.state =
pcmk_pacemakerd_api_daemon_state_text2enum(
crm_element_value(msg_data, PCMK__XA_PACEMAKERD_STATE));
reply_data.data.ping.status =
pcmk__str_eq(crm_element_value(msg_data, PCMK_XA_RESULT), "ok",
pcmk__str_casei)?pcmk_rc_ok:pcmk_rc_error;
reply_data.data.ping.last_good = (value_ll < 0)? 0 : (time_t) value_ll;
reply_data.data.ping.sys_from =
crm_element_value(msg_data, PCMK__XA_CRM_SUBSYSTEM);
} else if (pcmk__str_eq(value, CRM_OP_QUIT, pcmk__str_none)) {
+ const char *op_status = crm_element_value(msg_data, PCMK__XA_OP_STATUS);
+
reply_data.reply_type = pcmk_pacemakerd_reply_shutdown;
- reply_data.data.shutdown.status = atoi(crm_element_value(msg_data, XML_LRM_ATTR_OPSTATUS));
+ reply_data.data.shutdown.status = atoi(op_status);
} else {
crm_info("Unrecognizable message from %s: unknown command '%s'",
pcmk_ipc_name(api, true), pcmk__s(value, ""));
status = CRM_EX_PROTOCOL;
goto done;
}
done:
pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data);
return false;
}
pcmk__ipc_methods_t *
pcmk__pacemakerd_api_methods(void)
{
pcmk__ipc_methods_t *cmds = calloc(1, sizeof(pcmk__ipc_methods_t));
if (cmds != NULL) {
cmds->new_data = new_data;
cmds->free_data = free_data;
cmds->post_connect = post_connect;
cmds->reply_expected = reply_expected;
cmds->dispatch = dispatch;
cmds->post_disconnect = post_disconnect;
}
return cmds;
}
static int
do_pacemakerd_api_call(pcmk_ipc_api_t *api, const char *ipc_name, const char *task)
{
pacemakerd_api_private_t *private;
xmlNode *cmd;
int rc;
if (api == NULL) {
return EINVAL;
}
private = api->api_data;
CRM_ASSERT(private != NULL);
cmd = create_request(task, NULL, NULL, CRM_SYSTEM_MCP,
pcmk__ipc_sys_name(ipc_name, "client"),
private->client_uuid);
if (cmd) {
rc = pcmk__send_ipc_request(api, cmd);
if (rc != pcmk_rc_ok) {
crm_debug("Couldn't send request to %s: %s rc=%d",
pcmk_ipc_name(api, true), pcmk_rc_str(rc), rc);
}
free_xml(cmd);
} else {
rc = ENOMSG;
}
return rc;
}
int
pcmk_pacemakerd_api_ping(pcmk_ipc_api_t *api, const char *ipc_name)
{
return do_pacemakerd_api_call(api, ipc_name, CRM_OP_PING);
}
int
pcmk_pacemakerd_api_shutdown(pcmk_ipc_api_t *api, const char *ipc_name)
{
return do_pacemakerd_api_call(api, ipc_name, CRM_OP_QUIT);
}
diff --git a/lib/fencing/st_actions.c b/lib/fencing/st_actions.c
index 67285b27f0..2629db319c 100644
--- a/lib/fencing/st_actions.c
+++ b/lib/fencing/st_actions.c
@@ -1,710 +1,710 @@
/*
- * Copyright 2004-2022 the Pacemaker project contributors
+ * Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <libgen.h>
#include <inttypes.h>
#include <sys/types.h>
#include <glib.h>
#include <crm/crm.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
#include <crm/msg_xml.h>
#include <crm/services_internal.h>
#include "fencing_private.h"
struct stonith_action_s {
/*! user defined data */
char *agent;
char *action;
GHashTable *args;
int timeout;
bool async;
void *userdata;
void (*done_cb) (int pid, const pcmk__action_result_t *result,
void *user_data);
void (*fork_cb) (int pid, void *user_data);
svc_action_t *svc_action;
/*! internal timing information */
time_t initial_start_time;
int tries;
int remaining_timeout;
int max_retries;
int pid;
pcmk__action_result_t result;
};
static int internal_stonith_action_execute(stonith_action_t *action);
static void log_action(stonith_action_t *action, pid_t pid);
/*!
* \internal
* \brief Set an action's result based on services library result
*
* \param[in,out] action Fence action to set result for
* \param[in,out] svc_action Service action to get result from
*/
static void
set_result_from_svc_action(stonith_action_t *action, svc_action_t *svc_action)
{
pcmk__set_result(&(action->result), svc_action->rc, svc_action->status,
services__exit_reason(svc_action));
pcmk__set_result_output(&(action->result),
services__grab_stdout(svc_action),
services__grab_stderr(svc_action));
}
static void
log_action(stonith_action_t *action, pid_t pid)
{
/* The services library has already logged the output at info or debug
* level, so just raise to warning for stderr.
*/
if (action->result.action_stderr != NULL) {
/* Logging the whole string confuses syslog when the string is xml */
char *prefix = crm_strdup_printf("%s[%d] stderr:", action->agent, pid);
crm_log_output(LOG_WARNING, prefix, action->result.action_stderr);
free(prefix);
}
}
static void
append_config_arg(gpointer key, gpointer value, gpointer user_data)
{
/* The fencer will filter "action" out when it registers the device,
* but ignore it here in case any external API users don't.
*
* Also filter out parameters handled directly by Pacemaker.
*/
if (!pcmk__str_eq(key, STONITH_ATTR_ACTION_OP, pcmk__str_casei)
&& !pcmk_stonith_param(key)
&& (strstr(key, CRM_META) == NULL)
&& !pcmk__str_eq(key, PCMK_XA_CRM_FEATURE_SET, pcmk__str_none)) {
crm_trace("Passing %s=%s with fence action",
(const char *) key, (const char *) (value? value : ""));
g_hash_table_insert((GHashTable *) user_data,
strdup(key), strdup(value? value : ""));
}
}
/*!
* \internal
* \brief Create a table of arguments for a fencing action
*
* \param[in] agent Fencing agent name
* \param[in] action Name of fencing action
* \param[in] target Name of target node for fencing action
* \param[in] target_nodeid Node ID of target node for fencing action
* \param[in] device_args Fence device parameters
* \param[in] port_map Target node-to-port mapping for fence device
* \param[in] host_arg Argument name for passing target
*
* \return Newly created hash table of arguments for fencing action
*/
static GHashTable *
make_args(const char *agent, const char *action, const char *target,
uint32_t target_nodeid, GHashTable *device_args,
GHashTable *port_map, const char *host_arg)
{
GHashTable *arg_list = NULL;
const char *value = NULL;
CRM_CHECK(action != NULL, return NULL);
arg_list = pcmk__strkey_table(free, free);
// Add action to arguments (using an alias if requested)
if (device_args) {
char buffer[512];
snprintf(buffer, sizeof(buffer), "pcmk_%s_action", action);
value = g_hash_table_lookup(device_args, buffer);
if (value) {
crm_debug("Substituting '%s' for fence action %s targeting %s",
value, action, pcmk__s(target, "no node"));
action = value;
}
}
g_hash_table_insert(arg_list, strdup(STONITH_ATTR_ACTION_OP),
strdup(action));
/* If this is a fencing operation against another node, add more standard
* arguments.
*/
if ((target != NULL) && (device_args != NULL)) {
const char *param = NULL;
/* Always pass the target's name, per
* https://github.com/ClusterLabs/fence-agents/blob/main/doc/FenceAgentAPI.md
*/
g_hash_table_insert(arg_list, strdup("nodename"), strdup(target));
// If the target's node ID was specified, pass it, too
if (target_nodeid != 0) {
char *nodeid = crm_strdup_printf("%" PRIu32, target_nodeid);
// cts-fencing looks for this log message
crm_info("Passing '%s' as nodeid with fence action '%s' targeting %s",
nodeid, action, pcmk__s(target, "no node"));
g_hash_table_insert(arg_list, strdup("nodeid"), nodeid);
}
// Check whether target must be specified in some other way
param = g_hash_table_lookup(device_args, PCMK_STONITH_HOST_ARGUMENT);
if (!pcmk__str_eq(agent, "fence_legacy", pcmk__str_none)
&& !pcmk__str_eq(param, PCMK__VALUE_NONE, pcmk__str_casei)) {
if (param == NULL) {
/* Use the caller's default for pcmk_host_argument, or "port" if
* none was given
*/
param = (host_arg == NULL)? "port" : host_arg;
}
value = g_hash_table_lookup(device_args, param);
if (pcmk__str_eq(value, "dynamic",
pcmk__str_casei|pcmk__str_null_matches)) {
/* If the host argument was "dynamic" or not explicitly specified,
* add it with the target
*/
const char *alias = NULL;
if (port_map) {
alias = g_hash_table_lookup(port_map, target);
}
if (alias == NULL) {
alias = target;
}
crm_debug("Passing %s='%s' with fence action %s targeting %s",
param, alias, action, pcmk__s(target, "no node"));
g_hash_table_insert(arg_list, strdup(param), strdup(alias));
}
}
}
if (device_args) {
g_hash_table_foreach(device_args, append_config_arg, arg_list);
}
return arg_list;
}
/*!
* \internal
* \brief Free all memory used by a stonith action
*
* \param[in,out] action Action to free
*/
void
stonith__destroy_action(stonith_action_t *action)
{
if (action) {
free(action->agent);
if (action->args) {
g_hash_table_destroy(action->args);
}
free(action->action);
if (action->svc_action) {
services_action_free(action->svc_action);
}
pcmk__reset_result(&(action->result));
free(action);
}
}
/*!
* \internal
* \brief Get the result of an executed stonith action
*
* \param[in] action Executed action
*
* \return Pointer to action's result (or NULL if \p action is NULL)
*/
pcmk__action_result_t *
stonith__action_result(stonith_action_t *action)
{
return (action == NULL)? NULL : &(action->result);
}
#define FAILURE_MAX_RETRIES 2
/*!
* \internal
* \brief Create a new fencing action to be executed
*
* \param[in] agent Fence agent to use
* \param[in] action_name Fencing action to be executed
* \param[in] target Name of target of fencing action (if known)
* \param[in] target_nodeid Node ID of target of fencing action (if known)
* \param[in] timeout_sec Timeout to be used when executing action
* \param[in] device_args Parameters to pass to fence agent
* \param[in] port_map Mapping of target names to device ports
* \param[in] host_arg Agent parameter used to pass target name
*
* \return Newly created fencing action (asserts on error, never NULL)
*/
stonith_action_t *
stonith__action_create(const char *agent, const char *action_name,
const char *target, uint32_t target_nodeid,
int timeout_sec, GHashTable *device_args,
GHashTable *port_map, const char *host_arg)
{
stonith_action_t *action = calloc(1, sizeof(stonith_action_t));
CRM_ASSERT(action != NULL);
action->args = make_args(agent, action_name, target, target_nodeid,
device_args, port_map, host_arg);
crm_debug("Preparing '%s' action targeting %s using agent %s",
action_name, pcmk__s(target, "no node"), agent);
action->agent = strdup(agent);
action->action = strdup(action_name);
action->timeout = action->remaining_timeout = timeout_sec;
action->max_retries = FAILURE_MAX_RETRIES;
pcmk__set_result(&(action->result), PCMK_OCF_UNKNOWN, PCMK_EXEC_UNKNOWN,
"Initialization bug in fencing library");
if (device_args) {
char buffer[512];
const char *value = NULL;
snprintf(buffer, sizeof(buffer), "pcmk_%s_retries", action_name);
value = g_hash_table_lookup(device_args, buffer);
if (value) {
action->max_retries = atoi(value);
}
}
return action;
}
static gboolean
update_remaining_timeout(stonith_action_t * action)
{
int diff = time(NULL) - action->initial_start_time;
if (action->tries >= action->max_retries) {
crm_info("Attempted to execute agent %s (%s) the maximum number of times (%d) allowed",
action->agent, action->action, action->max_retries);
action->remaining_timeout = 0;
} else if ((action->result.execution_status != PCMK_EXEC_TIMEOUT)
&& (diff < (action->timeout * 0.7))) {
/* only set remaining timeout period if there is 30%
* or greater of the original timeout period left */
action->remaining_timeout = action->timeout - diff;
} else {
action->remaining_timeout = 0;
}
return action->remaining_timeout ? TRUE : FALSE;
}
/*!
* \internal
* \brief Map a fencing action result to a standard return code
*
* \param[in] result Fencing action result to map
*
* \return Standard Pacemaker return code that best corresponds to \p result
*/
int
stonith__result2rc(const pcmk__action_result_t *result)
{
if (pcmk__result_ok(result)) {
return pcmk_rc_ok;
}
switch (result->execution_status) {
case PCMK_EXEC_PENDING: return EINPROGRESS;
case PCMK_EXEC_CANCELLED: return ECANCELED;
case PCMK_EXEC_TIMEOUT: return ETIME;
case PCMK_EXEC_NOT_INSTALLED: return ENOENT;
case PCMK_EXEC_NOT_SUPPORTED: return EOPNOTSUPP;
case PCMK_EXEC_NOT_CONNECTED: return ENOTCONN;
case PCMK_EXEC_NO_FENCE_DEVICE: return ENODEV;
case PCMK_EXEC_NO_SECRETS: return EACCES;
/* For the fencing API, PCMK_EXEC_INVALID is used with fencer API
* operations that don't involve executing an agent (for example,
* registering devices). This allows us to use the CRM_EX_* codes in the
* exit status for finer-grained responses.
*/
case PCMK_EXEC_INVALID:
switch (result->exit_status) {
case CRM_EX_INVALID_PARAM: return EINVAL;
case CRM_EX_INSUFFICIENT_PRIV: return EACCES;
case CRM_EX_PROTOCOL: return EPROTO;
/* CRM_EX_EXPIRED is used for orphaned fencing operations left
* over from a previous instance of the fencer. For API backward
* compatibility, this is mapped to the previously used code for
* this case, EHOSTUNREACH.
*/
case CRM_EX_EXPIRED: return EHOSTUNREACH;
default: break;
}
break;
default:
break;
}
// Try to provide useful error code based on result's error output
if (result->action_stderr == NULL) {
return ENODATA;
} else if (strcasestr(result->action_stderr, "timed out")
|| strcasestr(result->action_stderr, "timeout")) {
return ETIME;
} else if (strcasestr(result->action_stderr, "unrecognised action")
|| strcasestr(result->action_stderr, "unrecognized action")
|| strcasestr(result->action_stderr, "unsupported action")) {
return EOPNOTSUPP;
}
// Oh well, we tried
return pcmk_rc_error;
}
/*!
* \internal
* \brief Determine execution status equivalent of legacy fencer return code
*
* Fence action notifications, and fence action callbacks from older fencers
* (<=2.1.2) in a rolling upgrade, will have only a legacy return code. Map this
* to an execution status as best as possible (essentially, the inverse of
* stonith__result2rc()).
*
* \param[in] rc Legacy return code from fencer
*
* \return Execution status best corresponding to \p rc
*/
int
stonith__legacy2status(int rc)
{
if (rc >= 0) {
return PCMK_EXEC_DONE;
}
switch (-rc) {
case EACCES: return PCMK_EXEC_NO_SECRETS;
case ECANCELED: return PCMK_EXEC_CANCELLED;
case EHOSTUNREACH: return PCMK_EXEC_INVALID;
case EINPROGRESS: return PCMK_EXEC_PENDING;
case ENODEV: return PCMK_EXEC_NO_FENCE_DEVICE;
case ENOENT: return PCMK_EXEC_NOT_INSTALLED;
case ENOTCONN: return PCMK_EXEC_NOT_CONNECTED;
case EOPNOTSUPP: return PCMK_EXEC_NOT_SUPPORTED;
case EPROTO: return PCMK_EXEC_INVALID;
case EPROTONOSUPPORT: return PCMK_EXEC_NOT_SUPPORTED;
case ETIME: return PCMK_EXEC_TIMEOUT;
case ETIMEDOUT: return PCMK_EXEC_TIMEOUT;
default: return PCMK_EXEC_ERROR;
}
}
/*!
* \internal
* \brief Add a fencing result to an XML element as attributes
*
* \param[in,out] xml XML element to add result to
* \param[in] result Fencing result to add (assume success if NULL)
*/
void
stonith__xe_set_result(xmlNode *xml, const pcmk__action_result_t *result)
{
int exit_status = CRM_EX_OK;
enum pcmk_exec_status execution_status = PCMK_EXEC_DONE;
const char *exit_reason = NULL;
const char *action_stdout = NULL;
int rc = pcmk_ok;
CRM_CHECK(xml != NULL, return);
if (result != NULL) {
exit_status = result->exit_status;
execution_status = result->execution_status;
exit_reason = result->exit_reason;
action_stdout = result->action_stdout;
rc = pcmk_rc2legacy(stonith__result2rc(result));
}
- crm_xml_add_int(xml, XML_LRM_ATTR_OPSTATUS, (int) execution_status);
+ crm_xml_add_int(xml, PCMK__XA_OP_STATUS, (int) execution_status);
crm_xml_add_int(xml, XML_LRM_ATTR_RC, exit_status);
crm_xml_add(xml, XML_LRM_ATTR_EXIT_REASON, exit_reason);
crm_xml_add(xml, F_STONITH_OUTPUT, action_stdout);
/* @COMPAT Peers in rolling upgrades, Pacemaker Remote nodes, and external
* code that use libstonithd <=2.1.2 don't check for the full result, and
* need a legacy return code instead.
*/
crm_xml_add_int(xml, F_STONITH_RC, rc);
}
/*!
* \internal
* \brief Find a fencing result beneath an XML element
*
* \param[in] xml XML element to search
*
* \return \p xml or descendant of it that contains a fencing result, else NULL
*/
xmlNode *
stonith__find_xe_with_result(xmlNode *xml)
{
xmlNode *match = get_xpath_object("//@" XML_LRM_ATTR_RC, xml, LOG_NEVER);
if (match == NULL) {
/* @COMPAT Peers <=2.1.2 in a rolling upgrade provide only a legacy
* return code, not a full result, so check for that.
*/
match = get_xpath_object("//@" F_STONITH_RC, xml, LOG_ERR);
}
return match;
}
/*!
* \internal
* \brief Get a fencing result from an XML element's attributes
*
* \param[in] xml XML element with fencing result
* \param[out] result Where to store fencing result
*/
void
stonith__xe_get_result(const xmlNode *xml, pcmk__action_result_t *result)
{
int exit_status = CRM_EX_OK;
int execution_status = PCMK_EXEC_DONE;
const char *exit_reason = NULL;
char *action_stdout = NULL;
CRM_CHECK((xml != NULL) && (result != NULL), return);
exit_reason = crm_element_value(xml, XML_LRM_ATTR_EXIT_REASON);
action_stdout = crm_element_value_copy(xml, F_STONITH_OUTPUT);
// A result must include an exit status and execution status
if ((crm_element_value_int(xml, XML_LRM_ATTR_RC, &exit_status) < 0)
- || (crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS,
+ || (crm_element_value_int(xml, PCMK__XA_OP_STATUS,
&execution_status) < 0)) {
int rc = pcmk_ok;
exit_status = CRM_EX_ERROR;
/* @COMPAT Peers <=2.1.2 in rolling upgrades provide only a legacy
* return code, not a full result, so check for that.
*/
if (crm_element_value_int(xml, F_STONITH_RC, &rc) == 0) {
if ((rc == pcmk_ok) || (rc == -EINPROGRESS)) {
exit_status = CRM_EX_OK;
}
execution_status = stonith__legacy2status(rc);
exit_reason = pcmk_strerror(rc);
} else {
execution_status = PCMK_EXEC_ERROR;
exit_reason = "Fencer reply contained neither a full result "
"nor a legacy return code (bug?)";
}
}
pcmk__set_result(result, exit_status, execution_status, exit_reason);
pcmk__set_result_output(result, action_stdout, NULL);
}
static void
stonith_action_async_done(svc_action_t *svc_action)
{
stonith_action_t *action = (stonith_action_t *) svc_action->cb_data;
set_result_from_svc_action(action, svc_action);
svc_action->params = NULL;
log_action(action, action->pid);
if (!pcmk__result_ok(&(action->result))
&& update_remaining_timeout(action)) {
int rc = internal_stonith_action_execute(action);
if (rc == pcmk_ok) {
return;
}
}
if (action->done_cb) {
action->done_cb(action->pid, &(action->result), action->userdata);
}
action->svc_action = NULL; // don't remove our caller
stonith__destroy_action(action);
}
static void
stonith_action_async_forked(svc_action_t *svc_action)
{
stonith_action_t *action = (stonith_action_t *) svc_action->cb_data;
action->pid = svc_action->pid;
action->svc_action = svc_action;
if (action->fork_cb) {
(action->fork_cb) (svc_action->pid, action->userdata);
}
pcmk__set_result(&(action->result), PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING,
NULL);
crm_trace("Child process %d performing action '%s' successfully forked",
action->pid, action->action);
}
static int
internal_stonith_action_execute(stonith_action_t * action)
{
int rc = -EPROTO;
int is_retry = 0;
svc_action_t *svc_action = NULL;
static int stonith_sequence = 0;
char *buffer = NULL;
CRM_CHECK(action != NULL, return -EINVAL);
if ((action->action == NULL) || (action->args == NULL)
|| (action->agent == NULL)) {
pcmk__set_result(&(action->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR_FATAL, "Bug in fencing library");
return -EINVAL;
}
if (!action->tries) {
action->initial_start_time = time(NULL);
}
action->tries++;
if (action->tries > 1) {
crm_info("Attempt %d to execute %s (%s). remaining timeout is %d",
action->tries, action->agent, action->action, action->remaining_timeout);
is_retry = 1;
}
buffer = crm_strdup_printf(PCMK__FENCE_BINDIR "/%s",
basename(action->agent));
svc_action = services_action_create_generic(buffer, NULL);
free(buffer);
if (svc_action->rc != PCMK_OCF_UNKNOWN) {
set_result_from_svc_action(action, svc_action);
services_action_free(svc_action);
return -E2BIG;
}
svc_action->timeout = 1000 * action->remaining_timeout;
svc_action->standard = strdup(PCMK_RESOURCE_CLASS_STONITH);
svc_action->id = crm_strdup_printf("%s_%s_%dof%d", basename(action->agent),
action->action, action->tries,
action->max_retries);
svc_action->agent = strdup(action->agent);
svc_action->sequence = stonith_sequence++;
svc_action->params = action->args;
svc_action->cb_data = (void *) action;
svc_action->flags = pcmk__set_flags_as(__func__, __LINE__,
LOG_TRACE, "Action",
svc_action->id, svc_action->flags,
SVC_ACTION_NON_BLOCKED,
"SVC_ACTION_NON_BLOCKED");
/* keep retries from executing out of control and free previous results */
if (is_retry) {
pcmk__reset_result(&(action->result));
sleep(1);
}
if (action->async) {
// We never create a recurring action, so this should always return TRUE
CRM_LOG_ASSERT(services_action_async_fork_notify(svc_action,
&stonith_action_async_done,
&stonith_action_async_forked));
return pcmk_ok;
} else if (services_action_sync(svc_action)) { // sync success
rc = pcmk_ok;
} else { // sync failure
rc = -ECONNABORTED;
}
set_result_from_svc_action(action, svc_action);
svc_action->params = NULL;
services_action_free(svc_action);
return rc;
}
/*!
* \internal
* \brief Kick off execution of an async stonith action
*
* \param[in,out] action Action to be executed
* \param[in,out] userdata Datapointer to be passed to callbacks
* \param[in] done Callback to notify action has failed/succeeded
* \param[in] fork_callback Callback to notify successful fork of child
*
* \return pcmk_ok if ownership of action has been taken, -errno otherwise
*/
int
stonith__execute_async(stonith_action_t * action, void *userdata,
void (*done) (int pid,
const pcmk__action_result_t *result,
void *user_data),
void (*fork_cb) (int pid, void *user_data))
{
if (!action) {
return -EINVAL;
}
action->userdata = userdata;
action->done_cb = done;
action->fork_cb = fork_cb;
action->async = true;
return internal_stonith_action_execute(action);
}
/*!
* \internal
* \brief Execute a stonith action
*
* \param[in,out] action Action to execute
*
* \return pcmk_ok on success, -errno otherwise
*/
int
stonith__execute(stonith_action_t *action)
{
int rc = pcmk_ok;
CRM_CHECK(action != NULL, return -EINVAL);
// Keep trying until success, max retries, or timeout
do {
rc = internal_stonith_action_execute(action);
} while ((rc != pcmk_ok) && update_remaining_timeout(action));
return rc;
}
diff --git a/lib/pacemaker/pcmk_sched_actions.c b/lib/pacemaker/pcmk_sched_actions.c
index f23c315781..f59103cdae 100644
--- a/lib/pacemaker/pcmk_sched_actions.c
+++ b/lib/pacemaker/pcmk_sched_actions.c
@@ -1,1934 +1,1934 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <sys/param.h>
#include <glib.h>
#include <crm/lrmd_internal.h>
#include <crm/common/scheduler_internal.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
/*!
* \internal
* \brief Get the action flags relevant to ordering constraints
*
* \param[in,out] action Action to check
* \param[in] node Node that *other* action in the ordering is on
* (used only for clone resource actions)
*
* \return Action flags that should be used for orderings
*/
static uint32_t
action_flags_for_ordering(pcmk_action_t *action, const pcmk_node_t *node)
{
bool runnable = false;
uint32_t flags;
// For non-resource actions, return the action flags
if (action->rsc == NULL) {
return action->flags;
}
/* For non-clone resources, or a clone action not assigned to a node,
* return the flags as determined by the resource method without a node
* specified.
*/
flags = action->rsc->cmds->action_flags(action, NULL);
if ((node == NULL) || !pe_rsc_is_clone(action->rsc)) {
return flags;
}
/* Otherwise (i.e., for clone resource actions on a specific node), first
* remember whether the non-node-specific action is runnable.
*/
runnable = pcmk_is_set(flags, pcmk_action_runnable);
// Then recheck the resource method with the node
flags = action->rsc->cmds->action_flags(action, node);
/* For clones in ordering constraints, the node-specific "runnable" doesn't
* matter, just the non-node-specific setting (i.e., is the action runnable
* anywhere).
*
* This applies only to runnable, and only for ordering constraints. This
* function shouldn't be used for other types of constraints without
* changes. Not very satisfying, but it's logical and appears to work well.
*/
if (runnable && !pcmk_is_set(flags, pcmk_action_runnable)) {
pcmk__set_raw_action_flags(flags, action->rsc->id,
pcmk_action_runnable);
}
return flags;
}
/*!
* \internal
* \brief Get action UUID that should be used with a resource ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the UUID and resource of the first action in an
* ordering, this returns the UUID of the action that should actually be used
* for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0").
*
* \param[in] first_uuid UUID of first action in ordering
* \param[in] first_rsc Resource of first action in ordering
*
* \return Newly allocated copy of UUID to use with ordering
* \note It is the caller's responsibility to free the return value.
*/
static char *
action_uuid_for_ordering(const char *first_uuid,
const pcmk_resource_t *first_rsc)
{
guint interval_ms = 0;
char *uuid = NULL;
char *rid = NULL;
char *first_task_str = NULL;
enum action_tasks first_task = pcmk_action_unspecified;
enum action_tasks remapped_task = pcmk_action_unspecified;
// Only non-notify actions for collective resources need remapping
if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL)
|| (first_rsc->variant < pcmk_rsc_variant_group)) {
goto done;
}
// Only non-recurring actions need remapping
CRM_ASSERT(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms));
if (interval_ms > 0) {
goto done;
}
first_task = text2task(first_task_str);
switch (first_task) {
case pcmk_action_stop:
case pcmk_action_start:
case pcmk_action_notify:
case pcmk_action_promote:
case pcmk_action_demote:
remapped_task = first_task + 1;
break;
case pcmk_action_stopped:
case pcmk_action_started:
case pcmk_action_notified:
case pcmk_action_promoted:
case pcmk_action_demoted:
remapped_task = first_task;
break;
case pcmk_action_monitor:
case pcmk_action_shutdown:
case pcmk_action_fence:
break;
default:
crm_err("Unknown action '%s' in ordering", first_task_str);
break;
}
if (remapped_task != pcmk_action_unspecified) {
/* If a clone or bundle has notifications enabled, the ordering will be
* relative to when notifications have been sent for the remapped task.
*/
if (pcmk_is_set(first_rsc->flags, pcmk_rsc_notify)
&& (pe_rsc_is_clone(first_rsc) || pe_rsc_is_bundled(first_rsc))) {
uuid = pcmk__notify_key(rid, "confirmed-post",
task2text(remapped_task));
} else {
uuid = pcmk__op_key(rid, task2text(remapped_task), 0);
}
pcmk__rsc_trace(first_rsc,
"Remapped action UUID %s to %s for ordering purposes",
first_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(first_uuid);
CRM_ASSERT(uuid != NULL);
}
free(first_task_str);
free(rid);
return uuid;
}
/*!
* \internal
* \brief Get actual action that should be used with an ordering
*
* When an action is ordered relative to an action for a collective resource
* (clone, group, or bundle), it actually needs to be ordered after all
* instances of the collective have completed the relevant action (for example,
* given "start CLONE then start RSC", RSC must wait until all instances of
* CLONE have started). Given the first action in an ordering, this returns the
* the action that should actually be used for ordering (for example, the
* started action instead of the start action).
*
* \param[in] action First action in an ordering
*
* \return Actual action that should be used for the ordering
*/
static pcmk_action_t *
action_for_ordering(pcmk_action_t *action)
{
pcmk_action_t *result = action;
pcmk_resource_t *rsc = action->rsc;
if ((rsc != NULL) && (rsc->variant >= pcmk_rsc_variant_group)
&& (action->uuid != NULL)) {
char *uuid = action_uuid_for_ordering(action->uuid, rsc);
result = find_first_action(rsc->actions, uuid, NULL, NULL);
if (result == NULL) {
crm_warn("Not remapping %s to %s because %s does not have "
"remapped action", action->uuid, uuid, rsc->id);
result = action;
}
free(uuid);
}
return result;
}
/*!
* \internal
* \brief Wrapper for update_ordered_actions() method for readability
*
* \param[in,out] rsc Resource to call method for
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this
* node (only used when interleaving instances)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates
* (may include pcmk_action_optional to affect only
* mandatory actions, and pe_action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pcmk__action_relation_flags to apply
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
static inline uint32_t
update(pcmk_resource_t *rsc, pcmk_action_t *first, pcmk_action_t *then,
const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type,
pcmk_scheduler_t *scheduler)
{
return rsc->cmds->update_ordered_actions(first, then, node, flags, filter,
type, scheduler);
}
/*!
* \internal
* \brief Update flags for ordering's actions appropriately for ordering's flags
*
* \param[in,out] first First action in an ordering
* \param[in,out] then Then action in an ordering
* \param[in] first_flags Action flags for \p first for ordering purposes
* \param[in] then_flags Action flags for \p then for ordering purposes
* \param[in,out] order Action wrapper for \p first in ordering
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags
*/
static uint32_t
update_action_for_ordering_flags(pcmk_action_t *first, pcmk_action_t *then,
uint32_t first_flags, uint32_t then_flags,
pcmk__related_action_t *order,
pcmk_scheduler_t *scheduler)
{
uint32_t changed = pcmk__updated_none;
/* The node will only be used for clones. If interleaved, node will be NULL,
* otherwise the ordering scope will be limited to the node. Normally, the
* whole 'then' clone should restart if 'first' is restarted, so then->node
* is needed.
*/
pcmk_node_t *node = then->node;
if (pcmk_is_set(order->type, pcmk__ar_first_implies_same_node_then)) {
/* For unfencing, only instances of 'then' on the same node as 'first'
* (the unfencing operation) should restart, so reset node to
* first->node, at which point this case is handled like a normal
* pcmk__ar_first_implies_then.
*/
pcmk__clear_relation_flags(order->type,
pcmk__ar_first_implies_same_node_then);
pcmk__set_relation_flags(order->type, pcmk__ar_first_implies_then);
node = first->node;
pcmk__rsc_trace(then->rsc,
"%s then %s: mapped "
"pcmk__ar_first_implies_same_node_then to "
"pcmk__ar_first_implies_then on %s",
first->uuid, then->uuid, pe__node_name(node));
}
if (pcmk_is_set(order->type, pcmk__ar_first_implies_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pcmk_action_optional,
pcmk_action_optional, pcmk__ar_first_implies_then,
scheduler);
} else if (!pcmk_is_set(first_flags, pcmk_action_optional)
&& pcmk_is_set(then->flags, pcmk_action_optional)) {
pcmk__clear_action_flags(then, pcmk_action_optional);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_first_implies_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_intermediate_stop)
&& (then->rsc != NULL)) {
enum pe_action_flags restart = pcmk_action_optional
|pcmk_action_runnable;
changed |= update(then->rsc, first, then, node, first_flags, restart,
pcmk__ar_intermediate_stop, scheduler);
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_intermediate_stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_then_implies_first)) {
if (first->rsc != NULL) {
changed |= update(first->rsc, first, then, node, first_flags,
pcmk_action_optional, pcmk__ar_then_implies_first,
scheduler);
} else if (!pcmk_is_set(first_flags, pcmk_action_optional)
&& pcmk_is_set(first->flags, pcmk_action_runnable)) {
pcmk__clear_action_flags(first, pcmk_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_then_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_promoted_then_implies_first)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node,
first_flags & pcmk_action_optional,
pcmk_action_optional,
pcmk__ar_promoted_then_implies_first, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after "
"pcmk__ar_promoted_then_implies_first",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_min_runnable)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_runnable, pcmk__ar_min_runnable,
scheduler);
} else if (pcmk_is_set(first_flags, pcmk_action_runnable)) {
// We have another runnable instance of "first"
then->runnable_before++;
/* Mark "then" as runnable if it requires a certain number of
* "before" instances to be runnable, and they now are.
*/
if ((then->runnable_before >= then->required_runnable_before)
&& !pcmk_is_set(then->flags, pcmk_action_runnable)) {
pcmk__set_action_flags(then, pcmk_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_min_runnable",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_nested_remote_probe)
&& (then->rsc != NULL)) {
if (!pcmk_is_set(first_flags, pcmk_action_runnable)
&& (first->rsc != NULL) && (first->rsc->running_on != NULL)) {
pcmk__rsc_trace(then->rsc,
"%s then %s: ignoring because first is stopping",
first->uuid, then->uuid);
order->type = (enum pe_ordering) pcmk__ar_none;
} else {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_runnable,
pcmk__ar_unrunnable_first_blocks, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_nested_remote_probe",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_unrunnable_first_blocks)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_runnable,
pcmk__ar_unrunnable_first_blocks, scheduler);
} else if (!pcmk_is_set(first_flags, pcmk_action_runnable)
&& pcmk_is_set(then->flags, pcmk_action_runnable)) {
pcmk__clear_action_flags(then, pcmk_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_unrunnable_first_blocks",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_unmigratable_then_blocks)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_optional,
pcmk__ar_unmigratable_then_blocks, scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after "
"pcmk__ar_unmigratable_then_blocks",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_first_else_then)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_optional, pcmk__ar_first_else_then,
scheduler);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after pcmk__ar_first_else_then",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_ordered)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_runnable, pcmk__ar_ordered,
scheduler);
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_ordered",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(order->type, pcmk__ar_asymmetric)) {
if (then->rsc != NULL) {
changed |= update(then->rsc, first, then, node, first_flags,
pcmk_action_runnable, pcmk__ar_asymmetric,
scheduler);
}
pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_asymmetric",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
if (pcmk_is_set(first->flags, pcmk_action_runnable)
&& pcmk_is_set(order->type, pcmk__ar_first_implies_then_graphed)
&& !pcmk_is_set(first_flags, pcmk_action_optional)) {
pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required",
then->uuid, first->uuid);
pcmk__set_action_flags(then, pcmk_action_always_in_graph);
// Don't bother marking 'then' as changed just for this
}
if (pcmk_is_set(order->type, pcmk__ar_then_implies_first_graphed)
&& !pcmk_is_set(then_flags, pcmk_action_optional)) {
pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required",
first->uuid, then->uuid);
pcmk__set_action_flags(first, pcmk_action_always_in_graph);
// Don't bother marking 'first' as changed just for this
}
if (pcmk_any_flags_set(order->type, pcmk__ar_first_implies_then
|pcmk__ar_then_implies_first
|pcmk__ar_intermediate_stop)
&& (first->rsc != NULL)
&& !pcmk_is_set(first->rsc->flags, pcmk_rsc_managed)
&& pcmk_is_set(first->rsc->flags, pcmk_rsc_blocked)
&& !pcmk_is_set(first->flags, pcmk_action_runnable)
&& pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) {
if (pcmk_is_set(then->flags, pcmk_action_runnable)) {
pcmk__clear_action_flags(then, pcmk_action_runnable);
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
}
pcmk__rsc_trace(then->rsc,
"%s then %s: %s after checking whether first "
"is blocked, unmanaged, unrunnable stop",
first->uuid, then->uuid,
(changed? "changed" : "unchanged"));
}
return changed;
}
// Convenience macros for logging action properties
#define action_type_str(flags) \
(pcmk_is_set((flags), pcmk_action_pseudo)? "pseudo-action" : "action")
#define action_optional_str(flags) \
(pcmk_is_set((flags), pcmk_action_optional)? "optional" : "required")
#define action_runnable_str(flags) \
(pcmk_is_set((flags), pcmk_action_runnable)? "runnable" : "unrunnable")
#define action_node_str(a) \
(((a)->node == NULL)? "no node" : (a)->node->details->uname)
/*!
* \internal
* \brief Update an action's flags for all orderings where it is "then"
*
* \param[in,out] then Action to update
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__update_action_for_orderings(pcmk_action_t *then,
pcmk_scheduler_t *scheduler)
{
GList *lpc = NULL;
uint32_t changed = pcmk__updated_none;
int last_flags = then->flags;
pcmk__rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s",
action_type_str(then->flags), then->uuid,
action_optional_str(then->flags),
action_runnable_str(then->flags), action_node_str(then));
if (pcmk_is_set(then->flags, pcmk_action_min_runnable)) {
/* Initialize current known "runnable before" actions. As
* update_action_for_ordering_flags() is called for each of then's
* before actions, this number will increment as runnable 'first'
* actions are encountered.
*/
then->runnable_before = 0;
if (then->required_runnable_before == 0) {
/* @COMPAT This ordering constraint uses the deprecated
* "require-all=false" attribute. Treat it like
* PCMK_META_CLONE_MIN=1.
*/
then->required_runnable_before = 1;
}
/* The pcmk__ar_min_runnable clause of
* update_action_for_ordering_flags() (called below)
* will reset runnable if appropriate.
*/
pcmk__clear_action_flags(then, pcmk_action_runnable);
}
for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *other = lpc->data;
pcmk_action_t *first = other->action;
pcmk_node_t *then_node = then->node;
pcmk_node_t *first_node = first->node;
if ((first->rsc != NULL)
&& (first->rsc->variant == pcmk_rsc_variant_group)
&& pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) {
first_node = first->rsc->fns->location(first->rsc, NULL, FALSE);
if (first_node != NULL) {
pcmk__rsc_trace(first->rsc, "Found %s for 'first' %s",
pe__node_name(first_node), first->uuid);
}
}
if ((then->rsc != NULL)
&& (then->rsc->variant == pcmk_rsc_variant_group)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) {
then_node = then->rsc->fns->location(then->rsc, NULL, FALSE);
if (then_node != NULL) {
pcmk__rsc_trace(then->rsc, "Found %s for 'then' %s",
pe__node_name(then_node), then->uuid);
}
}
// Disable constraint if it only applies when on same node, but isn't
if (pcmk_is_set(other->type, pcmk__ar_if_on_same_node)
&& (first_node != NULL) && (then_node != NULL)
&& !pe__same_node(first_node, then_node)) {
pcmk__rsc_trace(then->rsc,
"Disabled ordering %s on %s then %s on %s: "
"not same node",
other->action->uuid, pe__node_name(first_node),
then->uuid, pe__node_name(then_node));
other->type = (enum pe_ordering) pcmk__ar_none;
continue;
}
pcmk__clear_updated_flags(changed, then, pcmk__updated_first);
if ((first->rsc != NULL)
&& pcmk_is_set(other->type, pcmk__ar_then_cancels_first)
&& !pcmk_is_set(then->flags, pcmk_action_optional)) {
/* 'then' is required, so we must abandon 'first'
* (e.g. a required stop cancels any agent reload).
*/
pcmk__set_action_flags(other->action, pcmk_action_optional);
if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) {
pcmk__clear_rsc_flags(first->rsc, pcmk_rsc_reload);
}
}
if ((first->rsc != NULL) && (then->rsc != NULL)
&& (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) {
first = action_for_ordering(first);
}
if (first != other->action) {
pcmk__rsc_trace(then->rsc, "Ordering %s after %s instead of %s",
then->uuid, first->uuid, other->action->uuid);
}
pcmk__rsc_trace(then->rsc,
"%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s",
first->uuid, first->flags, then->uuid, then->flags,
other->type, action_node_str(first));
if (first == other->action) {
/* 'first' was not remapped (e.g. from 'start' to 'running'), which
* could mean it is a non-resource action, a primitive resource
* action, or already expanded.
*/
uint32_t first_flags, then_flags;
first_flags = action_flags_for_ordering(first, then_node);
then_flags = action_flags_for_ordering(then, first_node);
changed |= update_action_for_ordering_flags(first, then,
first_flags, then_flags,
other, scheduler);
/* 'first' was for a complex resource (clone, group, etc),
* create a new dependency if necessary
*/
} else if (order_actions(first, then, other->type)) {
/* This was the first time 'first' and 'then' were associated,
* start again to get the new actions_before list
*/
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
pcmk__rsc_trace(then->rsc,
"Disabled ordering %s then %s in favor of %s "
"then %s",
other->action->uuid, then->uuid, first->uuid,
then->uuid);
other->type = (enum pe_ordering) pcmk__ar_none;
}
if (pcmk_is_set(changed, pcmk__updated_first)) {
crm_trace("Re-processing %s and its 'after' actions "
"because it changed", first->uuid);
for (GList *lpc2 = first->actions_after; lpc2 != NULL;
lpc2 = lpc2->next) {
pcmk__related_action_t *other = lpc2->data;
pcmk__update_action_for_orderings(other->action, scheduler);
}
pcmk__update_action_for_orderings(first, scheduler);
}
}
if (pcmk_is_set(then->flags, pcmk_action_min_runnable)) {
if (last_flags == then->flags) {
pcmk__clear_updated_flags(changed, then, pcmk__updated_then);
} else {
pcmk__set_updated_flags(changed, then, pcmk__updated_then);
}
}
if (pcmk_is_set(changed, pcmk__updated_then)) {
crm_trace("Re-processing %s and its 'after' actions because it changed",
then->uuid);
if (pcmk_is_set(last_flags, pcmk_action_runnable)
&& !pcmk_is_set(then->flags, pcmk_action_runnable)) {
pcmk__block_colocation_dependents(then);
}
pcmk__update_action_for_orderings(then, scheduler);
for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) {
pcmk__related_action_t *other = lpc->data;
pcmk__update_action_for_orderings(other->action, scheduler);
}
}
}
static inline bool
is_primitive_action(const pcmk_action_t *action)
{
return (action != NULL) && (action->rsc != NULL)
&& (action->rsc->variant == pcmk_rsc_variant_primitive);
}
/*!
* \internal
* \brief Clear a single action flag and set reason text
*
* \param[in,out] action Action whose flag should be cleared
* \param[in] flag Action flag that should be cleared
* \param[in] reason Action that is the reason why flag is being cleared
*/
#define clear_action_flag_because(action, flag, reason) do { \
if (pcmk_is_set((action)->flags, (flag))) { \
pcmk__clear_action_flags(action, flag); \
if ((action)->rsc != (reason)->rsc) { \
char *reason_text = pe__action2reason((reason), (flag)); \
pe_action_set_reason((action), reason_text, false); \
free(reason_text); \
} \
} \
} while (0)
/*!
* \internal
* \brief Update actions in an asymmetric ordering
*
* If the "first" action in an asymmetric ordering is unrunnable, make the
* "second" action unrunnable as well, if appropriate.
*
* \param[in] first 'First' action in an asymmetric ordering
* \param[in,out] then 'Then' action in an asymmetric ordering
*/
static void
handle_asymmetric_ordering(const pcmk_action_t *first, pcmk_action_t *then)
{
/* Only resource actions after an unrunnable 'first' action need updates for
* asymmetric ordering.
*/
if ((then->rsc == NULL)
|| pcmk_is_set(first->flags, pcmk_action_runnable)) {
return;
}
// Certain optional 'then' actions are unaffected by unrunnable 'first'
if (pcmk_is_set(then->flags, pcmk_action_optional)) {
enum rsc_role_e then_rsc_role = then->rsc->fns->state(then->rsc, TRUE);
if ((then_rsc_role == pcmk_role_stopped)
&& pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) {
/* If 'then' should stop after 'first' but is already stopped, the
* ordering is irrelevant.
*/
return;
} else if ((then_rsc_role >= pcmk_role_started)
&& pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)
&& pe__rsc_running_on_only(then->rsc, then->node)) {
/* Similarly if 'then' should start after 'first' but is already
* started on a single node.
*/
return;
}
}
// 'First' can't run, so 'then' can't either
clear_action_flag_because(then, pcmk_action_optional, first);
clear_action_flag_because(then, pcmk_action_runnable, first);
}
/*!
* \internal
* \brief Set action bits appropriately when pe_restart_order is used
*
* \param[in,out] first 'First' action in an ordering with pe_restart_order
* \param[in,out] then 'Then' action in an ordering with pe_restart_order
* \param[in] filter What action flags to care about
*
* \note pe_restart_order is set for "stop resource before starting it" and
* "stop later group member before stopping earlier group member"
*/
static void
handle_restart_ordering(pcmk_action_t *first, pcmk_action_t *then,
uint32_t filter)
{
const char *reason = NULL;
CRM_ASSERT(is_primitive_action(first));
CRM_ASSERT(is_primitive_action(then));
// We need to update the action in two cases:
// ... if 'then' is required
if (pcmk_is_set(filter, pcmk_action_optional)
&& !pcmk_is_set(then->flags, pcmk_action_optional)) {
reason = "restart";
}
/* ... if 'then' is unrunnable action on same resource (if a resource
* should restart but can't start, we still want to stop)
*/
if (pcmk_is_set(filter, pcmk_action_runnable)
&& !pcmk_is_set(then->flags, pcmk_action_runnable)
&& pcmk_is_set(then->rsc->flags, pcmk_rsc_managed)
&& (first->rsc == then->rsc)) {
reason = "stop";
}
if (reason == NULL) {
return;
}
pcmk__rsc_trace(first->rsc, "Handling %s -> %s for %s",
first->uuid, then->uuid, reason);
// Make 'first' required if it is runnable
if (pcmk_is_set(first->flags, pcmk_action_runnable)) {
clear_action_flag_because(first, pcmk_action_optional, then);
}
// Make 'first' required if 'then' is required
if (!pcmk_is_set(then->flags, pcmk_action_optional)) {
clear_action_flag_because(first, pcmk_action_optional, then);
}
// Make 'first' unmigratable if 'then' is unmigratable
if (!pcmk_is_set(then->flags, pcmk_action_migratable)) {
clear_action_flag_because(first, pcmk_action_migratable, then);
}
// Make 'then' unrunnable if 'first' is required but unrunnable
if (!pcmk_is_set(first->flags, pcmk_action_optional)
&& !pcmk_is_set(first->flags, pcmk_action_runnable)) {
clear_action_flag_because(then, pcmk_action_runnable, first);
}
}
/*!
* \internal
* \brief Update two actions according to an ordering between them
*
* Given information about an ordering of two actions, update the actions' flags
* (and runnable_before members if appropriate) as appropriate for the ordering.
* Effects may cascade to other orderings involving the actions as well.
*
* \param[in,out] first 'First' action in an ordering
* \param[in,out] then 'Then' action in an ordering
* \param[in] node If not NULL, limit scope of ordering to this node
* (ignored)
* \param[in] flags Action flags for \p first for ordering purposes
* \param[in] filter Action flags to limit scope of certain updates (may
* include pcmk_action_optional to affect only
* mandatory actions, and pcmk_action_runnable to
* affect only runnable actions)
* \param[in] type Group of enum pcmk__action_relation_flags to apply
* \param[in,out] scheduler Scheduler data
*
* \return Group of enum pcmk__updated flags indicating what was updated
*/
uint32_t
pcmk__update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then,
const pcmk_node_t *node, uint32_t flags,
uint32_t filter, uint32_t type,
pcmk_scheduler_t *scheduler)
{
uint32_t changed = pcmk__updated_none;
uint32_t then_flags = 0U;
uint32_t first_flags = 0U;
CRM_ASSERT((first != NULL) && (then != NULL) && (scheduler != NULL));
then_flags = then->flags;
first_flags = first->flags;
if (pcmk_is_set(type, pcmk__ar_asymmetric)) {
handle_asymmetric_ordering(first, then);
}
if (pcmk_is_set(type, pcmk__ar_then_implies_first)
&& !pcmk_is_set(then_flags, pcmk_action_optional)) {
// Then is required, and implies first should be, too
if (pcmk_is_set(filter, pcmk_action_optional)
&& !pcmk_is_set(flags, pcmk_action_optional)
&& pcmk_is_set(first_flags, pcmk_action_optional)) {
clear_action_flag_because(first, pcmk_action_optional, then);
}
if (pcmk_is_set(flags, pcmk_action_migratable)
&& !pcmk_is_set(then->flags, pcmk_action_migratable)) {
clear_action_flag_because(first, pcmk_action_migratable, then);
}
}
if (pcmk_is_set(type, pcmk__ar_promoted_then_implies_first)
&& (then->rsc != NULL) && (then->rsc->role == pcmk_role_promoted)
&& pcmk_is_set(filter, pcmk_action_optional)
&& !pcmk_is_set(then->flags, pcmk_action_optional)) {
clear_action_flag_because(first, pcmk_action_optional, then);
if (pcmk_is_set(first->flags, pcmk_action_migratable)
&& !pcmk_is_set(then->flags, pcmk_action_migratable)) {
clear_action_flag_because(first, pcmk_action_migratable, then);
}
}
if (pcmk_is_set(type, pcmk__ar_unmigratable_then_blocks)
&& pcmk_is_set(filter, pcmk_action_optional)) {
if (!pcmk_all_flags_set(then->flags, pcmk_action_migratable
|pcmk_action_runnable)) {
clear_action_flag_because(first, pcmk_action_runnable, then);
}
if (!pcmk_is_set(then->flags, pcmk_action_optional)) {
clear_action_flag_because(first, pcmk_action_optional, then);
}
}
if (pcmk_is_set(type, pcmk__ar_first_else_then)
&& pcmk_is_set(filter, pcmk_action_optional)
&& !pcmk_is_set(first->flags, pcmk_action_runnable)) {
clear_action_flag_because(then, pcmk_action_migratable, first);
pcmk__clear_action_flags(then, pcmk_action_pseudo);
}
if (pcmk_is_set(type, pcmk__ar_unrunnable_first_blocks)
&& pcmk_is_set(filter, pcmk_action_runnable)
&& pcmk_is_set(then->flags, pcmk_action_runnable)
&& !pcmk_is_set(flags, pcmk_action_runnable)) {
clear_action_flag_because(then, pcmk_action_runnable, first);
clear_action_flag_because(then, pcmk_action_migratable, first);
}
if (pcmk_is_set(type, pcmk__ar_first_implies_then)
&& pcmk_is_set(filter, pcmk_action_optional)
&& pcmk_is_set(then->flags, pcmk_action_optional)
&& !pcmk_is_set(flags, pcmk_action_optional)
&& !pcmk_is_set(first->flags, pcmk_action_migratable)) {
clear_action_flag_because(then, pcmk_action_optional, first);
}
if (pcmk_is_set(type, pcmk__ar_intermediate_stop)) {
handle_restart_ordering(first, then, filter);
}
if (then_flags != then->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_then);
pcmk__rsc_trace(then->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'first' %s (%#.6x)",
then->uuid, pe__node_name(then->node),
then->flags, then_flags, first->uuid, first->flags);
if ((then->rsc != NULL) && (then->rsc->parent != NULL)) {
// Required to handle "X_stop then X_start" for cloned groups
pcmk__update_action_for_orderings(then, scheduler);
}
}
if (first_flags != first->flags) {
pcmk__set_updated_flags(changed, first, pcmk__updated_first);
pcmk__rsc_trace(first->rsc,
"%s on %s: flags are now %#.6x (was %#.6x) "
"because of 'then' %s (%#.6x)",
first->uuid, pe__node_name(first->node),
first->flags, first_flags, then->uuid, then->flags);
}
return changed;
}
/*!
* \internal
* \brief Trace-log an action (optionally with its dependent actions)
*
* \param[in] pre_text If not NULL, prefix the log with this plus ": "
* \param[in] action Action to log
* \param[in] details If true, recursively log dependent actions
*/
void
pcmk__log_action(const char *pre_text, const pcmk_action_t *action,
bool details)
{
const char *node_uname = NULL;
const char *node_uuid = NULL;
const char *desc = NULL;
CRM_CHECK(action != NULL, return);
if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) {
if (action->node != NULL) {
node_uname = action->node->details->uname;
node_uuid = action->node->details->id;
} else {
node_uname = "<none>";
}
}
switch (text2task(action->task)) {
case pcmk_action_fence:
case pcmk_action_shutdown:
if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
desc = "Pseudo ";
} else if (pcmk_is_set(action->flags, pcmk_action_optional)) {
desc = "Optional ";
} else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) {
desc = "!!Non-Startable!! ";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
default:
if (pcmk_is_set(action->flags, pcmk_action_optional)) {
desc = "Optional ";
} else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
desc = "Pseudo ";
} else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) {
desc = "!!Non-Startable!! ";
} else {
desc = "(Provisional) ";
}
crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s",
((pre_text == NULL)? "" : pre_text),
((pre_text == NULL)? "" : ": "),
desc, action->id, action->uuid,
(action->rsc? action->rsc->id : "<none>"),
(node_uname? "\ton " : ""), (node_uname? node_uname : ""),
(node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""),
(node_uuid? ")" : ""));
break;
}
if (details) {
const GList *iter = NULL;
const pcmk__related_action_t *other = NULL;
crm_trace("\t\t====== Preceding Actions");
for (iter = action->actions_before; iter != NULL; iter = iter->next) {
other = (const pcmk__related_action_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== Subsequent Actions");
for (iter = action->actions_after; iter != NULL; iter = iter->next) {
other = (const pcmk__related_action_t *) iter->data;
pcmk__log_action("\t\t", other->action, false);
}
crm_trace("\t\t====== End");
} else {
crm_trace("\t\t(before=%d, after=%d)",
g_list_length(action->actions_before),
g_list_length(action->actions_after));
}
}
/*!
* \internal
* \brief Create a new shutdown action for a node
*
* \param[in,out] node Node being shut down
*
* \return Newly created shutdown action for \p node
*/
pcmk_action_t *
pcmk__new_shutdown_action(pcmk_node_t *node)
{
char *shutdown_id = NULL;
pcmk_action_t *shutdown_op = NULL;
CRM_ASSERT(node != NULL);
shutdown_id = crm_strdup_printf("%s-%s", PCMK_ACTION_DO_SHUTDOWN,
node->details->uname);
shutdown_op = custom_action(NULL, shutdown_id, PCMK_ACTION_DO_SHUTDOWN,
node, FALSE, node->details->data_set);
pcmk__order_stops_before_shutdown(node, shutdown_op);
add_hash_param(shutdown_op->meta, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE);
return shutdown_op;
}
/*!
* \internal
* \brief Calculate and add an operation digest to XML
*
* Calculate an operation digest, which enables us to later determine when a
* restart is needed due to the resource's parameters being changed, and add it
* to given XML.
*
* \param[in] op Operation result from executor
* \param[in,out] update XML to add digest to
*/
static void
add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update)
{
char *digest = NULL;
xmlNode *args_xml = NULL;
if (op->params == NULL) {
return;
}
args_xml = create_xml_node(NULL, XML_TAG_PARAMS);
g_hash_table_foreach(op->params, hash2field, args_xml);
pcmk__filter_op_for_digest(args_xml);
digest = calculate_operation_digest(args_xml, NULL);
crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest);
free_xml(args_xml);
free(digest);
}
#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
/*!
* \internal
* \brief Create XML for resource operation history update
*
* \param[in,out] parent Parent XML node to add to
* \param[in,out] op Operation event data
* \param[in] caller_version DC feature set
* \param[in] target_rc Expected result of operation
* \param[in] node Name of node on which operation was performed
* \param[in] origin Arbitrary description of update source
*
* \return Newly created XML node for history update
*/
xmlNode *
pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op,
const char *caller_version, int target_rc,
const char *node, const char *origin)
{
char *key = NULL;
char *magic = NULL;
char *op_id = NULL;
char *op_id_additional = NULL;
char *local_user_data = NULL;
const char *exit_reason = NULL;
xmlNode *xml_op = NULL;
const char *task = NULL;
CRM_CHECK(op != NULL, return NULL);
crm_trace("Creating history XML for %s-interval %s action for %s on %s "
"(DC version: %s, origin: %s)",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
((node == NULL)? "no node" : node), caller_version, origin);
task = op->op_type;
/* Record a successful agent reload as a start, and a failed one as a
* monitor, to make life easier for the scheduler when determining the
* current state.
*
* @COMPAT We should check "reload" here only if the operation was for a
* pre-OCF-1.1 resource agent, but we don't know that here, and we should
* only ever get results for actions scheduled by us, so we can reasonably
* assume any "reload" is actually a pre-1.1 agent reload.
*/
if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT,
NULL)) {
if (op->op_status == PCMK_EXEC_DONE) {
task = PCMK_ACTION_START;
} else {
task = PCMK_ACTION_MONITOR;
}
}
key = pcmk__op_key(op->rsc_id, task, op->interval_ms);
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) {
const char *n_type = crm_meta_value(op->params, "notify_type");
const char *n_task = crm_meta_value(op->params, "notify_operation");
CRM_LOG_ASSERT(n_type != NULL);
CRM_LOG_ASSERT(n_task != NULL);
op_id = pcmk__notify_key(op->rsc_id, n_type, n_task);
if (op->op_status != PCMK_EXEC_PENDING) {
/* Ignore notify errors.
*
* @TODO It might be better to keep the correct result here, and
* ignore it in process_graph_event().
*/
lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
}
/* Migration history is preserved separately, which usually matters for
* multiple nodes and is important for future cluster transitions.
*/
} else if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
op_id = strdup(key);
} else if (did_rsc_op_fail(op, target_rc)) {
op_id = pcmk__op_key(op->rsc_id, "last_failure", 0);
if (op->interval_ms == 0) {
/* Ensure 'last' gets updated, in case PCMK_META_RECORD_PENDING is
* true
*/
op_id_additional = pcmk__op_key(op->rsc_id, "last", 0);
}
exit_reason = op->exit_reason;
} else if (op->interval_ms > 0) {
op_id = strdup(key);
} else {
op_id = pcmk__op_key(op->rsc_id, "last", 0);
}
again:
xml_op = pcmk__xe_match(parent, XML_LRM_TAG_RSC_OP, PCMK_XA_ID, op_id);
if (xml_op == NULL) {
xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP);
}
if (op->user_data == NULL) {
crm_debug("Generating fake transition key for: " PCMK__OP_FMT
" %d from %s", op->rsc_id, op->op_type, op->interval_ms,
op->call_id, origin);
local_user_data = pcmk__transition_key(-1, op->call_id, target_rc,
FAKE_TE_ID);
op->user_data = local_user_data;
}
if (magic == NULL) {
magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc,
(const char *) op->user_data);
}
crm_xml_add(xml_op, PCMK_XA_ID, op_id);
crm_xml_add(xml_op, PCMK__XA_OPERATION_KEY, key);
crm_xml_add(xml_op, PCMK_XA_OPERATION, task);
crm_xml_add(xml_op, PCMK_XA_CRM_DEBUG_ORIGIN, origin);
crm_xml_add(xml_op, PCMK_XA_CRM_FEATURE_SET, caller_version);
crm_xml_add(xml_op, PCMK__XA_TRANSITION_KEY, op->user_data);
crm_xml_add(xml_op, PCMK__XA_TRANSITION_MAGIC, magic);
crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, pcmk__s(exit_reason, ""));
crm_xml_add(xml_op, PCMK__META_ON_NODE, node); // For context during triage
crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id);
crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc);
- crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status);
+ crm_xml_add_int(xml_op, PCMK__XA_OP_STATUS, op->op_status);
crm_xml_add_ms(xml_op, PCMK_META_INTERVAL, op->interval_ms);
if (compare_version("2.1", caller_version) <= 0) {
if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) {
crm_trace("Timing data (" PCMK__OP_FMT
"): last=%u change=%u exec=%u queue=%u",
op->rsc_id, op->op_type, op->interval_ms,
op->t_run, op->t_rcchange, op->exec_time, op->queue_time);
if ((op->interval_ms != 0) && (op->t_rcchange != 0)) {
// Recurring ops may have changed rc after initial run
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_rcchange);
} else {
crm_xml_add_ll(xml_op, XML_RSC_OP_LAST_CHANGE,
(long long) op->t_run);
}
crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time);
crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time);
}
}
if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/*
* Record migrate_source and migrate_target always for migrate ops.
*/
const char *name = XML_LRM_ATTR_MIGRATE_SOURCE;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
name = XML_LRM_ATTR_MIGRATE_TARGET;
crm_xml_add(xml_op, name, crm_meta_value(op->params, name));
}
add_op_digest_to_xml(op, xml_op);
if (op_id_additional) {
free(op_id);
op_id = op_id_additional;
op_id_additional = NULL;
goto again;
}
if (local_user_data) {
free(local_user_data);
op->user_data = NULL;
}
free(magic);
free(op_id);
free(key);
return xml_op;
}
/*!
* \internal
* \brief Check whether an action shutdown-locks a resource to a node
*
* If the PCMK_OPT_SHUTDOWN_LOCK cluster property is set, resources will not be
* recovered on a different node if cleanly stopped, and may start only on that
* same node. This function checks whether that applies to a given action, so
* that the transition graph can be marked appropriately.
*
* \param[in] action Action to check
*
* \return true if \p action locks its resource to the action's node,
* otherwise false
*/
bool
pcmk__action_locks_rsc_to_node(const pcmk_action_t *action)
{
// Only resource actions taking place on resource's lock node are locked
if ((action == NULL) || (action->rsc == NULL)
|| !pe__same_node(action->node, action->rsc->lock_node)) {
return false;
}
/* During shutdown, only stops are locked (otherwise, another action such as
* a demote would cause the controller to clear the lock)
*/
if (action->node->details->shutdown && (action->task != NULL)
&& (strcmp(action->task, PCMK_ACTION_STOP) != 0)) {
return false;
}
return true;
}
/* lowest to highest */
static gint
sort_action_id(gconstpointer a, gconstpointer b)
{
const pcmk__related_action_t *action_wrapper2 = a;
const pcmk__related_action_t *action_wrapper1 = b;
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (action_wrapper1->action->id < action_wrapper2->action->id) {
return 1;
}
if (action_wrapper1->action->id > action_wrapper2->action->id) {
return -1;
}
return 0;
}
/*!
* \internal
* \brief Remove any duplicate action inputs, merging action flags
*
* \param[in,out] action Action whose inputs should be checked
*/
void
pcmk__deduplicate_action_inputs(pcmk_action_t *action)
{
GList *item = NULL;
GList *next = NULL;
pcmk__related_action_t *last_input = NULL;
action->actions_before = g_list_sort(action->actions_before,
sort_action_id);
for (item = action->actions_before; item != NULL; item = next) {
pcmk__related_action_t *input = item->data;
next = item->next;
if ((last_input != NULL)
&& (input->action->id == last_input->action->id)) {
crm_trace("Input %s (%d) duplicate skipped for action %s (%d)",
input->action->uuid, input->action->id,
action->uuid, action->id);
/* For the purposes of scheduling, the ordering flags no longer
* matter, but crm_simulate looks at certain ones when creating a
* dot graph. Combining the flags is sufficient for that purpose.
*/
last_input->type |= input->type;
if (input->state == pe_link_dumped) {
last_input->state = pe_link_dumped;
}
free(item->data);
action->actions_before = g_list_delete_link(action->actions_before,
item);
} else {
last_input = input;
input->state = pe_link_not_dumped;
}
}
}
/*!
* \internal
* \brief Output all scheduled actions
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__output_actions(pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->priv;
// Output node (non-resource) actions
for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) {
char *node_name = NULL;
char *task = NULL;
pcmk_action_t *action = (pcmk_action_t *) iter->data;
if (action->rsc != NULL) {
continue; // Resource actions will be output later
} else if (pcmk_is_set(action->flags, pcmk_action_optional)) {
continue; // This action was not scheduled
}
if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN,
pcmk__str_none)) {
task = strdup("Shutdown");
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta,
"stonith_action");
task = crm_strdup_printf("Fence (%s)", op);
} else {
continue; // Don't display other node action types
}
if (pe__is_guest_node(action->node)) {
const pcmk_resource_t *remote = action->node->details->remote_rsc;
node_name = crm_strdup_printf("%s (resource: %s)",
pe__node_name(action->node),
remote->container->id);
} else if (action->node != NULL) {
node_name = crm_strdup_printf("%s", pe__node_name(action->node));
}
out->message(out, "node-action", task, node_name, action->reason);
free(node_name);
free(task);
}
// Output resource actions
for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
rsc->cmds->output_actions(rsc);
}
}
/*!
* \internal
* \brief Get action name needed to compare digest for configuration changes
*
* \param[in] task Action name from history
* \param[in] interval_ms Action interval (in milliseconds)
*
* \return Action name whose digest should be compared
*/
static const char *
task_for_digest(const char *task, guint interval_ms)
{
/* Certain actions need to be compared against the parameters used to start
* the resource.
*/
if ((interval_ms == 0)
&& pcmk__str_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_FROM,
PCMK_ACTION_PROMOTE, NULL)) {
task = PCMK_ACTION_START;
}
return task;
}
/*!
* \internal
* \brief Check whether only sanitized parameters to an action changed
*
* When collecting CIB files for troubleshooting, crm_report will mask
* sensitive resource parameters. If simulations were run using that, affected
* resources would appear to need a restart, which would complicate
* troubleshooting. To avoid that, we save a "secure digest" of non-sensitive
* parameters. This function used that digest to check whether only masked
* parameters are different.
*
* \param[in] xml_op Resource history entry with secure digest
* \param[in] digest_data Operation digest information being compared
* \param[in] scheduler Scheduler data
*
* \return true if only sanitized parameters changed, otherwise false
*/
static bool
only_sanitized_changed(const xmlNode *xml_op,
const pcmk__op_digest_t *digest_data,
const pcmk_scheduler_t *scheduler)
{
const char *digest_secure = NULL;
if (!pcmk_is_set(scheduler->flags, pcmk_sched_sanitized)) {
// The scheduler is not being run as a simulation
return false;
}
digest_secure = crm_element_value(xml_op, XML_LRM_ATTR_SECURE_DIGEST);
return (digest_data->rc != pcmk__digest_match) && (digest_secure != NULL)
&& (digest_data->digest_secure_calc != NULL)
&& (strcmp(digest_data->digest_secure_calc, digest_secure) == 0);
}
/*!
* \internal
* \brief Force a restart due to a configuration change
*
* \param[in,out] rsc Resource that action is for
* \param[in] task Name of action whose configuration changed
* \param[in] interval_ms Action interval (in milliseconds)
* \param[in,out] node Node where resource should be restarted
*/
static void
force_restart(pcmk_resource_t *rsc, const char *task, guint interval_ms,
pcmk_node_t *node)
{
char *key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk_action_t *required = custom_action(rsc, key, task, NULL, FALSE,
rsc->cluster);
pe_action_set_reason(required, "resource definition change", true);
trigger_unfencing(rsc, node, "Device parameters changed", NULL,
rsc->cluster);
}
/*!
* \internal
* \brief Schedule a reload of a resource on a node
*
* \param[in,out] data Resource to reload
* \param[in] user_data Where resource should be reloaded
*/
static void
schedule_reload(gpointer data, gpointer user_data)
{
pcmk_resource_t *rsc = data;
const pcmk_node_t *node = user_data;
pcmk_action_t *reload = NULL;
// For collective resources, just call recursively for children
if (rsc->variant > pcmk_rsc_variant_primitive) {
g_list_foreach(rsc->children, schedule_reload, user_data);
return;
}
// Skip the reload in certain situations
if ((node == NULL)
|| !pcmk_is_set(rsc->flags, pcmk_rsc_managed)
|| pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pcmk__rsc_trace(rsc, "Skip reload of %s:%s%s %s",
rsc->id,
pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : " unmanaged",
pcmk_is_set(rsc->flags, pcmk_rsc_failed)? " failed" : "",
(node == NULL)? "inactive" : node->details->uname);
return;
}
/* If a resource's configuration changed while a start was pending,
* force a full restart instead of a reload.
*/
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
pcmk__rsc_trace(rsc,
"%s: preventing agent reload because start pending",
rsc->id);
custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE,
rsc->cluster);
return;
}
// Schedule the reload
pcmk__set_rsc_flags(rsc, pcmk_rsc_reload);
reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node,
FALSE, rsc->cluster);
pe_action_set_reason(reload, "resource definition change", FALSE);
// Set orderings so that a required stop or demote cancels the reload
pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL,
pcmk__ar_ordered|pcmk__ar_then_cancels_first,
rsc->cluster);
pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL,
pcmk__ar_ordered|pcmk__ar_then_cancels_first,
rsc->cluster);
}
/*!
* \internal
* \brief Handle any configuration change for an action
*
* Given an action from resource history, if the resource's configuration
* changed since the action was done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, etc.).
*
* \param[in,out] rsc Resource that action is for
* \param[in,out] node Node that action was on
* \param[in] xml_op Action XML from resource history
*
* \return true if action configuration changed, otherwise false
*/
bool
pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node,
const xmlNode *xml_op)
{
guint interval_ms = 0;
const char *task = NULL;
const pcmk__op_digest_t *digest_data = NULL;
CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL),
return false);
task = crm_element_value(xml_op, PCMK_XA_OPERATION);
CRM_CHECK(task != NULL, return false);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
// If this is a recurring action, check whether it has been orphaned
if (interval_ms > 0) {
if (pcmk__find_action_config(rsc, task, interval_ms, false) != NULL) {
pcmk__rsc_trace(rsc,
"%s-interval %s for %s on %s is in configuration",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_cancel_removed_actions)) {
pcmk__schedule_cancel(rsc,
crm_element_value(xml_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "orphan");
return true;
} else {
pcmk__rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
return true;
}
}
crm_trace("Checking %s-interval %s for %s on %s for configuration changes",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node));
task = task_for_digest(task, interval_ms);
digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->cluster);
if (only_sanitized_changed(xml_op, digest_data, rsc->cluster)) {
if (!pcmk__is_daemon && (rsc->cluster->priv != NULL)) {
pcmk__output_t *out = rsc->cluster->priv;
out->info(out,
"Only 'private' parameters to %s-interval %s for %s "
"on %s changed: %s",
pcmk__readable_interval(interval_ms), task, rsc->id,
pe__node_name(node),
crm_element_value(xml_op, PCMK__XA_TRANSITION_MAGIC));
}
return false;
}
switch (digest_data->rc) {
case pcmk__digest_restart:
crm_log_xml_debug(digest_data->params_restart, "params:restart");
force_restart(rsc, task, interval_ms, node);
return true;
case pcmk__digest_unknown:
case pcmk__digest_mismatch:
// Changes that can potentially be handled by an agent reload
if (interval_ms > 0) {
/* Recurring actions aren't reloaded per se, they are just
* re-scheduled so the next run uses the new parameters.
* The old instance will be cancelled automatically.
*/
crm_log_xml_debug(digest_data->params_all, "params:reschedule");
pcmk__reschedule_recurring(rsc, task, interval_ms, node);
} else if (crm_element_value(xml_op,
XML_LRM_ATTR_RESTART_DIGEST) != NULL) {
// Agent supports reload, so use it
trigger_unfencing(rsc, node,
"Device parameters changed (reload)", NULL,
rsc->cluster);
crm_log_xml_debug(digest_data->params_all, "params:reload");
schedule_reload((gpointer) rsc, (gpointer) node);
} else {
pcmk__rsc_trace(rsc,
"Restarting %s "
"because agent doesn't support reload",
rsc->id);
crm_log_xml_debug(digest_data->params_restart,
"params:restart");
force_restart(rsc, task, interval_ms, node);
}
return true;
default:
break;
}
return false;
}
/*!
* \internal
* \brief Create a list of resource's action history entries, sorted by call ID
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[out] start_index Where to store index of start-like action, if any
* \param[out] stop_index Where to store index of stop action, if any
*/
static GList *
rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index)
{
GList *ops = NULL;
for (xmlNode *rsc_op = first_named_child(rsc_entry, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
ops = g_list_prepend(ops, rsc_op);
}
ops = g_list_sort(ops, sort_op_by_callid);
calculate_active_ops(ops, start_index, stop_index);
return ops;
}
/*!
* \internal
* \brief Process a resource's action history from the CIB status
*
* Given a resource's action history, if the resource's configuration
* changed since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in] rsc_entry Resource's <lrm_rsc_op> status XML
* \param[in,out] rsc Resource whose history is being processed
* \param[in,out] node Node whose history is being processed
*/
static void
process_rsc_history(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
pcmk_node_t *node)
{
int offset = -1;
int stop_index = 0;
int start_index = 0;
GList *sorted_op_list = NULL;
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
if (pe_rsc_is_anon_clone(pe__const_top_resource(rsc, false))) {
pcmk__rsc_trace(rsc,
"Skipping configuration check "
"for orphaned clone instance %s",
rsc->id);
} else {
pcmk__rsc_trace(rsc,
"Skipping configuration check and scheduling "
"clean-up for orphaned resource %s", rsc->id);
pcmk__schedule_cleanup(rsc, node, false);
}
return;
}
if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) {
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) {
pcmk__schedule_cleanup(rsc, node, false);
}
pcmk__rsc_trace(rsc,
"Skipping configuration check for %s "
"because no longer active on %s",
rsc->id, pe__node_name(node));
return;
}
pcmk__rsc_trace(rsc, "Checking for configuration changes for %s on %s",
rsc->id, pe__node_name(node));
if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) {
pcmk__schedule_cleanup(rsc, node, false);
}
sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index);
if (start_index < stop_index) {
return; // Resource is stopped
}
for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
xmlNode *rsc_op = (xmlNode *) iter->data;
const char *task = NULL;
guint interval_ms = 0;
if (++offset < start_index) {
// Skip actions that happened before a start
continue;
}
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if ((interval_ms > 0)
&& (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)
|| node->details->maintenance)) {
// Maintenance mode cancels recurring operations
pcmk__schedule_cancel(rsc,
crm_element_value(rsc_op,
XML_LRM_ATTR_CALLID),
task, interval_ms, node, "maintenance mode");
} else if ((interval_ms > 0)
|| pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START,
PCMK_ACTION_PROMOTE,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
/* If a resource operation failed, and the operation's definition
* has changed, clear any fail count so they can be retried fresh.
*/
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't assigned resources to nodes yet, so if the
* REMOTE_CONTAINER_HACK is used, we may calculate the digest
* based on the literal "#uname" value rather than the properly
* substituted value. That would mistakenly make the action
* definition appear to have been changed. Defer the check until
* later in this case.
*/
pe__add_param_check(rsc_op, rsc, node, pcmk__check_active,
rsc->cluster);
} else if (pcmk__check_action_config(rsc, node, rsc_op)
&& (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective,
NULL) != 0)) {
pe__clear_failcount(rsc, node, "action definition changed",
rsc->cluster);
}
}
}
g_list_free(sorted_op_list);
}
/*!
* \internal
* \brief Process a node's action history from the CIB status
*
* Given a node's resource history, if the resource's configuration changed
* since the actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] node Node whose history is being processed
* \param[in] lrm_rscs Node's <lrm_resources> from CIB status XML
*/
static void
process_node_history(pcmk_node_t *node, const xmlNode *lrm_rscs)
{
crm_trace("Processing node history for %s", pe__node_name(node));
for (const xmlNode *rsc_entry = first_named_child(lrm_rscs,
XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
if (rsc_entry->children != NULL) {
GList *result = pcmk__rscs_matching_id(ID(rsc_entry),
node->details->data_set);
for (GList *iter = result; iter != NULL; iter = iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
if (rsc->variant == pcmk_rsc_variant_primitive) {
process_rsc_history(rsc_entry, rsc, node);
}
}
g_list_free(result);
}
}
}
// XPath to find a node's resource history
#define XPATH_NODE_HISTORY "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE "[@" PCMK_XA_UNAME "='%s']" \
"/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES
/*!
* \internal
* \brief Process any resource configuration changes in the CIB status
*
* Go through all nodes' resource history, and if a resource's configuration
* changed since its actions were done, schedule any actions needed (restart,
* reload, unfencing, rescheduling recurring actions, clean-up, etc.).
* (This also cancels recurring actions for maintenance mode, which is not
* entirely related but convenient to do here.)
*
* \param[in,out] scheduler Scheduler data
*/
void
pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler)
{
crm_trace("Check resource and action configuration for changes");
/* Rather than iterate through the status section, iterate through the nodes
* and search for the appropriate status subsection for each. This skips
* orphaned nodes and lets us eliminate some cases before searching the XML.
*/
for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) {
pcmk_node_t *node = (pcmk_node_t *) iter->data;
/* Don't bother checking actions for a node that can't run actions ...
* unless it's in maintenance mode, in which case we still need to
* cancel any existing recurring monitors.
*/
if (node->details->maintenance
|| pcmk__node_available(node, false, false)) {
char *xpath = NULL;
xmlNode *history = NULL;
xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->details->uname);
history = get_xpath_object(xpath, scheduler->input, LOG_NEVER);
free(xpath);
process_node_history(node, history);
}
}
}
diff --git a/lib/pengine/pe_output.c b/lib/pengine/pe_output.c
index 525dd5d75b..92f21b0587 100644
--- a/lib/pengine/pe_output.c
+++ b/lib/pengine/pe_output.c
@@ -1,3182 +1,3182 @@
/*
* Copyright 2019-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdint.h>
#include <crm/common/xml_internal.h>
#include <crm/common/output.h>
#include <crm/common/scheduler_internal.h>
#include <crm/cib/util.h>
#include <crm/msg_xml.h>
#include <crm/pengine/internal.h>
const char *
pe__resource_description(const pcmk_resource_t *rsc, uint32_t show_opts)
{
const char * desc = NULL;
// User-supplied description
if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description)) {
desc = crm_element_value(rsc->xml, PCMK_XA_DESCRIPTION);
}
return desc;
}
/* Never display node attributes whose name starts with one of these prefixes */
#define FILTER_STR { PCMK__FAIL_COUNT_PREFIX, PCMK__LAST_FAILURE_PREFIX, \
"shutdown", PCMK_NODE_ATTR_TERMINATE, "standby", "#", \
NULL }
static int
compare_attribute(gconstpointer a, gconstpointer b)
{
int rc;
rc = strcmp((const char *)a, (const char *)b);
return rc;
}
/*!
* \internal
* \brief Determine whether extended information about an attribute should be added.
*
* \param[in] node Node that ran this resource
* \param[in,out] rsc_list List of resources for this node
* \param[in,out] scheduler Scheduler data
* \param[in] attrname Attribute to find
* \param[out] expected_score Expected value for this attribute
*
* \return true if extended information should be printed, false otherwise
* \note Currently, extended information is only supported for ping/pingd
* resources, for which a message will be printed if connectivity is lost
* or degraded.
*/
static bool
add_extra_info(const pcmk_node_t *node, GList *rsc_list,
pcmk_scheduler_t *scheduler, const char *attrname,
int *expected_score)
{
GList *gIter = NULL;
for (gIter = rsc_list; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
const char *type = g_hash_table_lookup(rsc->meta, PCMK_XA_TYPE);
const char *name = NULL;
GHashTable *params = NULL;
if (rsc->children != NULL) {
if (add_extra_info(node, rsc->children, scheduler, attrname,
expected_score)) {
return true;
}
}
if (!pcmk__strcase_any_of(type, "ping", "pingd", NULL)) {
continue;
}
params = pe_rsc_params(rsc, node, scheduler);
name = g_hash_table_lookup(params, PCMK_XA_NAME);
if (name == NULL) {
name = "pingd";
}
/* To identify the resource with the attribute name. */
if (pcmk__str_eq(name, attrname, pcmk__str_casei)) {
int host_list_num = 0;
const char *hosts = g_hash_table_lookup(params, "host_list");
const char *multiplier = g_hash_table_lookup(params, "multiplier");
int multiplier_i;
if (hosts) {
char **host_list = g_strsplit(hosts, " ", 0);
host_list_num = g_strv_length(host_list);
g_strfreev(host_list);
}
if ((multiplier == NULL)
|| (pcmk__scan_min_int(multiplier, &multiplier_i,
INT_MIN) != pcmk_rc_ok)) {
/* The ocf:pacemaker:ping resource agent defaults multiplier to
* 1. The agent currently does not handle invalid text, but it
* should, and this would be a reasonable choice ...
*/
multiplier_i = 1;
}
*expected_score = host_list_num * multiplier_i;
return true;
}
}
return false;
}
static GList *
filter_attr_list(GList *attr_list, char *name)
{
int i;
const char *filt_str[] = FILTER_STR;
CRM_CHECK(name != NULL, return attr_list);
/* filtering automatic attributes */
for (i = 0; filt_str[i] != NULL; i++) {
if (g_str_has_prefix(name, filt_str[i])) {
return attr_list;
}
}
return g_list_insert_sorted(attr_list, name, compare_attribute);
}
static GList *
get_operation_list(xmlNode *rsc_entry) {
GList *op_list = NULL;
xmlNode *rsc_op = NULL;
for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL;
rsc_op = pcmk__xe_next(rsc_op)) {
const char *task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
const char *interval_ms_s = crm_element_value(rsc_op,
PCMK_META_INTERVAL);
const char *op_rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC);
int op_rc_i;
pcmk__scan_min_int(op_rc, &op_rc_i, 0);
/* Display 0-interval monitors as "probe" */
if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) {
task = "probe";
}
/* Ignore notifies and some probes */
if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)
|| (pcmk__str_eq(task, "probe", pcmk__str_none)
&& (op_rc_i == CRM_EX_NOT_RUNNING))) {
continue;
}
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) {
op_list = g_list_append(op_list, rsc_op);
}
}
op_list = g_list_sort(op_list, sort_op_by_callid);
return op_list;
}
static void
add_dump_node(gpointer key, gpointer value, gpointer user_data)
{
xmlNodePtr node = user_data;
pcmk_create_xml_text_node(node, (const char *) key, (const char *) value);
}
static void
append_dump_text(gpointer key, gpointer value, gpointer user_data)
{
char **dump_text = user_data;
char *new_text = crm_strdup_printf("%s %s=%s",
*dump_text, (char *)key, (char *)value);
free(*dump_text);
*dump_text = new_text;
}
#define XPATH_STACK "//" XML_CIB_TAG_NVPAIR \
"[@" PCMK_XA_NAME "='" \
PCMK_OPT_CLUSTER_INFRASTRUCTURE "']"
static const char *
get_cluster_stack(pcmk_scheduler_t *scheduler)
{
xmlNode *stack = get_xpath_object(XPATH_STACK, scheduler->input, LOG_DEBUG);
return (stack != NULL)? crm_element_value(stack, PCMK_XA_VALUE) : "unknown";
}
static char *
last_changed_string(const char *last_written, const char *user,
const char *client, const char *origin) {
if (last_written != NULL || user != NULL || client != NULL || origin != NULL) {
return crm_strdup_printf("%s%s%s%s%s%s%s",
last_written ? last_written : "",
user ? " by " : "",
user ? user : "",
client ? " via " : "",
client ? client : "",
origin ? " on " : "",
origin ? origin : "");
} else {
return strdup("");
}
}
static char *
op_history_string(xmlNode *xml_op, const char *task, const char *interval_ms_s,
int rc, bool print_timing) {
const char *call = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
char *interval_str = NULL;
char *buf = NULL;
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *pair = pcmk__format_nvpair("interval", interval_ms_s, "ms");
interval_str = crm_strdup_printf(" %s", pair);
free(pair);
}
if (print_timing) {
char *last_change_str = NULL;
char *exec_str = NULL;
char *queue_str = NULL;
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &epoch) == pcmk_ok)
&& (epoch > 0)) {
char *epoch_str = pcmk__epoch2str(&epoch, 0);
last_change_str = crm_strdup_printf(" %s=\"%s\"",
XML_RSC_OP_LAST_CHANGE,
pcmk__s(epoch_str, ""));
free(epoch_str);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
if (value) {
char *pair = pcmk__format_nvpair(XML_RSC_OP_T_EXEC, value, "ms");
exec_str = crm_strdup_printf(" %s", pair);
free(pair);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
if (value) {
char *pair = pcmk__format_nvpair(XML_RSC_OP_T_QUEUE, value, "ms");
queue_str = crm_strdup_printf(" %s", pair);
free(pair);
}
buf = crm_strdup_printf("(%s) %s:%s%s%s%s rc=%d (%s)", call, task,
interval_str ? interval_str : "",
last_change_str ? last_change_str : "",
exec_str ? exec_str : "",
queue_str ? queue_str : "",
rc, services_ocf_exitcode_str(rc));
if (last_change_str) {
free(last_change_str);
}
if (exec_str) {
free(exec_str);
}
if (queue_str) {
free(queue_str);
}
} else {
buf = crm_strdup_printf("(%s) %s%s%s", call, task,
interval_str ? ":" : "",
interval_str ? interval_str : "");
}
if (interval_str) {
free(interval_str);
}
return buf;
}
static char *
resource_history_string(pcmk_resource_t *rsc, const char *rsc_id, bool all,
int failcount, time_t last_failure) {
char *buf = NULL;
if (rsc == NULL) {
buf = crm_strdup_printf("%s: orphan", rsc_id);
} else if (all || failcount || last_failure > 0) {
char *failcount_s = NULL;
char *lastfail_s = NULL;
if (failcount > 0) {
failcount_s = crm_strdup_printf(" %s=%d", PCMK__FAIL_COUNT_PREFIX,
failcount);
} else {
failcount_s = strdup("");
}
if (last_failure > 0) {
buf = pcmk__epoch2str(&last_failure, 0);
lastfail_s = crm_strdup_printf(" %s='%s'",
PCMK__LAST_FAILURE_PREFIX, buf);
free(buf);
}
buf = crm_strdup_printf("%s: " PCMK_META_MIGRATION_THRESHOLD "=%d%s%s",
rsc_id, rsc->migration_threshold, failcount_s,
lastfail_s? lastfail_s : "");
free(failcount_s);
free(lastfail_s);
} else {
buf = crm_strdup_printf("%s:", rsc_id);
}
return buf;
}
/*!
* \internal
* \brief Get a node's feature set for status display purposes
*
* \param[in] node Node to check
*
* \return String representation of feature set if the node is fully up (using
* "<3.15.1" for older nodes that don't set the #feature-set attribute),
* otherwise NULL
*/
static const char *
get_node_feature_set(const pcmk_node_t *node)
{
if (node->details->online && node->details->expected_up
&& !pe__is_guest_or_remote_node(node)) {
const char *feature_set = g_hash_table_lookup(node->details->attrs,
CRM_ATTR_FEATURE_SET);
/* The feature set attribute is present since 3.15.1. If it is missing,
* then the node must be running an earlier version.
*/
return pcmk__s(feature_set, "<3.15.1");
}
return NULL;
}
static bool
is_mixed_version(pcmk_scheduler_t *scheduler)
{
const char *feature_set = NULL;
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
const char *node_feature_set = get_node_feature_set(node);
if (node_feature_set != NULL) {
if (feature_set == NULL) {
feature_set = node_feature_set;
} else if (strcmp(feature_set, node_feature_set) != 0) {
return true;
}
}
}
return false;
}
static char *
formatted_xml_buf(const pcmk_resource_t *rsc, bool raw)
{
if (raw) {
return dump_xml_formatted(rsc->orig_xml ? rsc->orig_xml : rsc->xml);
} else {
return dump_xml_formatted(rsc->xml);
}
}
#define XPATH_DC_VERSION "//" XML_CIB_TAG_NVPAIR \
"[@" PCMK_XA_NAME "='" PCMK_OPT_DC_VERSION "']"
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
if (pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, PCMK_XA_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
PCMK_XA_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
PCMK_XA_CIB_LAST_WRITTEN);
const char *user = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_ORIGIN);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times",
scheduler->localhost, last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->ninstances, scheduler->disabled_resources,
scheduler->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *",
"enum pcmk_pacemakerd_state", "uint32_t", "uint32_t")
static int
cluster_summary_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
int rc = pcmk_rc_no_output;
const char *stack_s = get_cluster_stack(scheduler);
if (pcmk_is_set(section_opts, pcmk_section_stack)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-stack", stack_s, pcmkd_state);
}
/* Always print DC if none, even if not requested */
if ((scheduler->dc_node == NULL)
|| pcmk_is_set(section_opts, pcmk_section_dc)) {
xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION,
scheduler->input, LOG_DEBUG);
const char *dc_version_s = dc_version?
crm_element_value(dc_version, PCMK_XA_VALUE)
: NULL;
const char *quorum = crm_element_value(scheduler->input,
PCMK_XA_HAVE_QUORUM);
char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL;
bool mixed_version = is_mixed_version(scheduler);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-dc", scheduler->dc_node, quorum,
dc_version_s, dc_name, mixed_version);
free(dc_name);
}
if (pcmk_is_set(section_opts, pcmk_section_times)) {
const char *last_written = crm_element_value(scheduler->input,
PCMK_XA_CIB_LAST_WRITTEN);
const char *user = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_USER);
const char *client = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_CLIENT);
const char *origin = crm_element_value(scheduler->input,
PCMK_XA_UPDATE_ORIGIN);
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-times",
scheduler->localhost, last_written, user, client, origin);
}
if (pcmk_is_set(section_opts, pcmk_section_counts)) {
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary");
out->message(out, "cluster-counts", g_list_length(scheduler->nodes),
scheduler->ninstances, scheduler->disabled_resources,
scheduler->blocked_resources);
}
if (pcmk_is_set(section_opts, pcmk_section_options)) {
/* Kind of a hack - close the list we may have opened earlier in this
* function so we can put all the options into their own list. We
* only want to do this on HTML output, though.
*/
PCMK__OUTPUT_LIST_FOOTER(out, rc);
out->begin_list(out, NULL, NULL, "Config Options");
out->message(out, "cluster-options", scheduler);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) {
if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
return rc;
}
char *
pe__node_display_name(pcmk_node_t *node, bool print_detail)
{
char *node_name;
const char *node_host = NULL;
const char *node_id = NULL;
int name_len;
CRM_ASSERT((node != NULL) && (node->details != NULL) && (node->details->uname != NULL));
/* Host is displayed only if this is a guest node and detail is requested */
if (print_detail && pe__is_guest_node(node)) {
const pcmk_resource_t *container = node->details->remote_rsc->container;
const pcmk_node_t *host_node = pe__current_node(container);
if (host_node && host_node->details) {
node_host = host_node->details->uname;
}
if (node_host == NULL) {
node_host = ""; /* so we at least get "uname@" to indicate guest */
}
}
/* Node ID is displayed if different from uname and detail is requested */
if (print_detail && !pcmk__str_eq(node->details->uname, node->details->id, pcmk__str_casei)) {
node_id = node->details->id;
}
/* Determine name length */
name_len = strlen(node->details->uname) + 1;
if (node_host) {
name_len += strlen(node_host) + 1; /* "@node_host" */
}
if (node_id) {
name_len += strlen(node_id) + 3; /* + " (node_id)" */
}
/* Allocate and populate display name */
node_name = malloc(name_len);
CRM_ASSERT(node_name != NULL);
strcpy(node_name, node->details->uname);
if (node_host) {
strcat(node_name, "@");
strcat(node_name, node_host);
}
if (node_id) {
strcat(node_name, " (");
strcat(node_name, node_id);
strcat(node_name, ")");
}
return node_name;
}
int
pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name
, size_t pairs_count, ...)
{
xmlNodePtr xml_node = NULL;
va_list args;
CRM_ASSERT(tag_name != NULL);
xml_node = pcmk__output_xml_peek_parent(out);
CRM_ASSERT(xml_node != NULL);
xml_node = create_xml_node(xml_node, tag_name);
va_start(args, pairs_count);
while(pairs_count--) {
const char *param_name = va_arg(args, const char *);
const char *param_value = va_arg(args, const char *);
if (param_name && param_value) {
crm_xml_add(xml_node, param_name, param_value);
}
};
va_end(args);
if (is_list) {
pcmk__output_xml_push_parent(out, xml_node);
}
return pcmk_rc_ok;
}
static const char *
role_desc(enum rsc_role_e role)
{
if (role == pcmk_role_promoted) {
#ifdef PCMK__COMPAT_2_0
return "as " PCMK__ROLE_PROMOTED_LEGACY " ";
#else
return "in " PCMK__ROLE_PROMOTED " role ";
#endif
}
return "";
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
char *buf = crm_strdup_printf("%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
pcmk__output_create_html_node(out, "li", NULL, NULL, buf);
free(node_name);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts = va_arg(args, uint32_t);
char *node_name = pe__node_display_name(pe_node,
pcmk_is_set(show_opts, pcmk_show_node_id));
out->list_item(out, NULL, "%s\tprevents %s from running %son %s",
location->id, location->rsc->id,
role_desc(location->role_filter), node_name);
free(node_name);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t")
static int
ban_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *);
pcmk__location_t *location = va_arg(args, pcmk__location_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *promoted_only = pcmk__btoa(location->role_filter == pcmk_role_promoted);
char *weight_s = pcmk__itoa(pe_node->weight);
pcmk__output_create_xml_node(out, "ban",
PCMK_XA_ID, location->id,
"resource", location->rsc->id,
"node", pe_node->details->uname,
"weight", weight_s,
"promoted-only", promoted_only,
/* This is a deprecated alias for
* promoted_only. Removing it will break
* backward compatibility of the API schema,
* which will require an API schema major
* version bump.
*/
"master_only", promoted_only,
NULL);
free(weight_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ban-list", "pcmk_scheduler_t *", "const char *", "GList *",
"uint32_t", "bool")
static int
ban_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *prefix = va_arg(args, const char *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *gIter, *gIter2;
int rc = pcmk_rc_no_output;
/* Print each ban */
for (gIter = scheduler->placement_constraints;
gIter != NULL; gIter = gIter->next) {
pcmk__location_t *location = gIter->data;
const pcmk_resource_t *rsc = location->rsc;
if (prefix != NULL && !g_str_has_prefix(location->id, prefix)) {
continue;
}
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(pe__const_top_resource(rsc, false)),
only_rsc, pcmk__str_star_matches)) {
continue;
}
for (gIter2 = location->nodes; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter2->data;
if (node->weight < 0) {
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Negative Location Constraints");
out->message(out, "ban", node, location, show_opts);
}
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_html(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "li", NULL);
char *nnodes_str = crm_strdup_printf("%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
pcmk_create_html_node(nodes_node, "span", NULL, NULL, nnodes_str);
free(nnodes_str);
if (ndisabled && nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
ndisabled);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED");
s = crm_strdup_printf(", %d ", nblocked);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED");
pcmk_create_html_node(resources_node, "span", NULL, NULL,
" from further action due to failure)");
} else if (ndisabled && !nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
ndisabled);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED");
pcmk_create_html_node(resources_node, "span", NULL, NULL, ")");
} else if (!ndisabled && nblocked) {
char *s = crm_strdup_printf("%d resource instance%s configured (%d ",
nresources, pcmk__plural_s(nresources),
nblocked);
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED");
pcmk_create_html_node(resources_node, "span", NULL, NULL,
" from further action due to failure)");
} else {
char *s = crm_strdup_printf("%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
pcmk_create_html_node(resources_node, "span", NULL, NULL, s);
free(s);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_text(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
out->list_item(out, NULL, "%d node%s configured",
nnodes, pcmk__plural_s(nnodes));
if (ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED, %d BLOCKED from "
"further action due to failure)",
nresources, pcmk__plural_s(nresources), ndisabled,
nblocked);
} else if (ndisabled && !nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d DISABLED)",
nresources, pcmk__plural_s(nresources), ndisabled);
} else if (!ndisabled && nblocked) {
out->list_item(out, NULL, "%d resource instance%s configured "
"(%d BLOCKED from further action "
"due to failure)",
nresources, pcmk__plural_s(nresources), nblocked);
} else {
out->list_item(out, NULL, "%d resource instance%s configured",
nresources, pcmk__plural_s(nresources));
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int")
static int
cluster_counts_xml(pcmk__output_t *out, va_list args) {
unsigned int nnodes = va_arg(args, unsigned int);
int nresources = va_arg(args, int);
int ndisabled = va_arg(args, int);
int nblocked = va_arg(args, int);
xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "nodes_configured", NULL);
xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "resources_configured", NULL);
char *s = pcmk__itoa(nnodes);
crm_xml_add(nodes_node, "number", s);
free(s);
s = pcmk__itoa(nresources);
crm_xml_add(resources_node, "number", s);
free(s);
s = pcmk__itoa(ndisabled);
crm_xml_add(resources_node, "disabled", s);
free(s);
s = pcmk__itoa(nblocked);
crm_xml_add(resources_node, "blocked", s);
free(s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, "bold", "Current DC: ");
if (dc) {
char *buf = crm_strdup_printf("%s (version %s) -", dc_name,
dc_version_s ? dc_version_s : "unknown");
pcmk_create_html_node(node, "span", NULL, NULL, buf);
free(buf);
if (mixed_version) {
pcmk_create_html_node(node, "span", NULL, "warning",
" MIXED-VERSION");
}
pcmk_create_html_node(node, "span", NULL, NULL, " partition");
if (crm_is_true(quorum)) {
pcmk_create_html_node(node, "span", NULL, NULL, " with");
} else {
pcmk_create_html_node(node, "span", NULL, "warning", " WITHOUT");
}
pcmk_create_html_node(node, "span", NULL, NULL, " quorum");
} else {
pcmk_create_html_node(node, "span", NULL, "warning", "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
out->list_item(out, "Current DC",
"%s (version %s) - %spartition %s quorum",
dc_name, dc_version_s ? dc_version_s : "unknown",
mixed_version ? "MIXED-VERSION " : "",
crm_is_true(quorum) ? "with" : "WITHOUT");
} else {
out->list_item(out, "Current DC", "NONE");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *",
"char *", "int")
static int
cluster_dc_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *dc = va_arg(args, pcmk_node_t *);
const char *quorum = va_arg(args, const char *);
const char *dc_version_s = va_arg(args, const char *);
char *dc_name G_GNUC_UNUSED = va_arg(args, char *);
bool mixed_version = va_arg(args, int);
if (dc) {
pcmk__output_create_xml_node(out, "current_dc",
"present", PCMK_VALUE_TRUE,
PCMK_XA_VERSION, pcmk__s(dc_version_s, ""),
PCMK_XA_NAME, dc->details->uname,
PCMK_XA_ID, dc->details->id,
"with_quorum", pcmk__btoa(crm_is_true(quorum)),
"mixed_version", pcmk__btoa(mixed_version),
NULL);
} else {
pcmk__output_create_xml_node(out, "current_dc",
"present", PCMK_VALUE_FALSE,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("maint-mode", "unsigned long long int")
static int
cluster_maint_mode_text(pcmk__output_t *out, va_list args) {
unsigned long long flags = va_arg(args, unsigned long long);
if (pcmk_is_set(flags, pcmk_sched_in_maintenance)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will not attempt to start, stop or recover services\n");
return pcmk_rc_ok;
} else if (pcmk_is_set(flags, pcmk_sched_stop_all)) {
pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n");
pcmk__formatted_printf(out, " The cluster will keep all resources stopped\n");
return pcmk_rc_ok;
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_html(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL, "No quorum policy: Suicide");
break;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: ");
pcmk_create_html_node(node, "span", NULL, "bold", "DISABLED");
pcmk_create_html_node(node, "span", NULL, NULL,
" (the cluster will not attempt to start, stop, or recover services)");
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) {
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: ");
pcmk_create_html_node(node, "span", NULL, "bold", "STOPPED");
pcmk_create_html_node(node, "span", NULL, NULL,
" (the cluster will keep all resources stopped)");
} else {
out->list_item(out, NULL, "Resource management: enabled");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_log(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
return out->info(out, "Resource management is DISABLED. The cluster will not attempt to start, stop or recover services.");
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) {
return out->info(out, "Resource management is DISABLED. The cluster has stopped all resources.");
} else {
return pcmk_rc_no_output;
}
}
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_text(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
out->list_item(out, NULL, "STONITH of failed nodes enabled");
} else {
out->list_item(out, NULL, "STONITH of failed nodes disabled");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
out->list_item(out, NULL, "Cluster is symmetric");
} else {
out->list_item(out, NULL, "Cluster is asymmetric");
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
out->list_item(out, NULL, "No quorum policy: Freeze resources");
break;
case pcmk_no_quorum_stop:
out->list_item(out, NULL, "No quorum policy: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
out->list_item(out, NULL, "No quorum policy: Demote promotable "
"resources and stop all other resources");
break;
case pcmk_no_quorum_ignore:
out->list_item(out, NULL, "No quorum policy: Ignore");
break;
case pcmk_no_quorum_fence:
out->list_item(out, NULL, "No quorum policy: Suicide");
break;
}
return pcmk_rc_ok;
}
#define bv(flag) pcmk__btoa(pcmk_is_set(scheduler->flags, (flag)))
PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *")
static int
cluster_options_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
const char *no_quorum_policy = NULL;
char *stonith_timeout_str = pcmk__itoa(scheduler->stonith_timeout);
char *priority_fencing_delay_str = pcmk__itoa(scheduler->priority_fencing_delay * 1000);
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
no_quorum_policy = "freeze";
break;
case pcmk_no_quorum_stop:
no_quorum_policy = "stop";
break;
case pcmk_no_quorum_demote:
no_quorum_policy = "demote";
break;
case pcmk_no_quorum_ignore:
no_quorum_policy = "ignore";
break;
case pcmk_no_quorum_fence:
no_quorum_policy = "suicide";
break;
}
pcmk__output_create_xml_node(out, "cluster_options",
PCMK_OPT_STONITH_ENABLED,
bv(pcmk_sched_fencing_enabled),
PCMK_OPT_SYMMETRIC_CLUSTER,
bv(pcmk_sched_symmetric_cluster),
PCMK_OPT_NO_QUORUM_POLICY, no_quorum_policy,
PCMK_OPT_MAINTENANCE_MODE,
bv(pcmk_sched_in_maintenance),
PCMK_OPT_STOP_ALL_RESOURCES,
bv(pcmk_sched_stop_all),
PCMK_OPT_STONITH_TIMEOUT "-ms",
stonith_timeout_str,
PCMK_OPT_PRIORITY_FENCING_DELAY "-ms",
priority_fencing_delay_str,
NULL);
free(stonith_timeout_str);
free(priority_fencing_delay_str);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_html(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(node, "span", NULL, "bold", "Stack: ");
pcmk_create_html_node(node, "span", NULL, NULL, stack_s);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
pcmk_create_html_node(node, "span", NULL, NULL, " (");
pcmk_create_html_node(node, "span", NULL, NULL,
pcmk__pcmkd_state_enum2friendly(pcmkd_state));
pcmk_create_html_node(node, "span", NULL, NULL, ")");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_text(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
out->list_item(out, "Stack", "%s (%s)",
stack_s, pcmk__pcmkd_state_enum2friendly(pcmkd_state));
} else {
out->list_item(out, "Stack", "%s", stack_s);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state")
static int
cluster_stack_xml(pcmk__output_t *out, va_list args) {
const char *stack_s = va_arg(args, const char *);
enum pcmk_pacemakerd_state pcmkd_state =
(enum pcmk_pacemakerd_state) va_arg(args, int);
const char *state_s = NULL;
if (pcmkd_state != pcmk_pacemakerd_state_invalid) {
state_s = pcmk_pacemakerd_api_daemon_state_enum2text(pcmkd_state);
}
pcmk__output_create_xml_node(out, "stack",
PCMK_XA_TYPE, stack_s,
"pacemakerd-state", state_s,
NULL);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_html(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
xmlNodePtr updated_node = pcmk__output_create_xml_node(out, "li", NULL);
xmlNodePtr changed_node = pcmk__output_create_xml_node(out, "li", NULL);
char *time_s = pcmk__epoch2str(NULL, 0);
pcmk_create_html_node(updated_node, "span", NULL, "bold", "Last updated: ");
pcmk_create_html_node(updated_node, "span", NULL, NULL, time_s);
if (our_nodename != NULL) {
pcmk_create_html_node(updated_node, "span", NULL, NULL, " on ");
pcmk_create_html_node(updated_node, "span", NULL, NULL, our_nodename);
}
free(time_s);
time_s = last_changed_string(last_written, user, client, origin);
pcmk_create_html_node(changed_node, "span", NULL, "bold", "Last change: ");
pcmk_create_html_node(changed_node, "span", NULL, NULL, time_s);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_xml(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
pcmk__output_create_xml_node(out, "last_update",
"time", time_s,
PCMK_XA_ORIGIN, our_nodename,
NULL);
pcmk__output_create_xml_node(out, "last_change",
"time", last_written ? last_written : "",
"user", user ? user : "",
"client", client ? client : "",
PCMK_XA_ORIGIN, pcmk__s(origin, ""),
NULL);
free(time_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *",
"const char *", "const char *", "const char *")
static int
cluster_times_text(pcmk__output_t *out, va_list args) {
const char *our_nodename = va_arg(args, const char *);
const char *last_written = va_arg(args, const char *);
const char *user = va_arg(args, const char *);
const char *client = va_arg(args, const char *);
const char *origin = va_arg(args, const char *);
char *time_s = pcmk__epoch2str(NULL, 0);
out->list_item(out, "Last updated", "%s%s%s",
time_s, (our_nodename != NULL)? " on " : "",
pcmk__s(our_nodename, ""));
free(time_s);
time_s = last_changed_string(last_written, user, client, origin);
out->list_item(out, "Last change", " %s", time_s);
free(time_s);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Display a failed action in less-technical natural language
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_friendly(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
char *rsc_id = NULL;
char *task = NULL;
guint interval_ms = 0;
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)
|| !parse_op_key(op_key, &rsc_id, &task, &interval_ms)) {
rsc_id = strdup("unknown resource");
task = strdup("unknown action");
interval_ms = 0;
}
CRM_ASSERT((rsc_id != NULL) && (task != NULL));
str = g_string_sized_new(256); // Should be sufficient for most messages
pcmk__g_strcat(str, rsc_id, " ", NULL);
if (interval_ms != 0) {
pcmk__g_strcat(str, pcmk__readable_interval(interval_ms), "-interval ",
NULL);
}
pcmk__g_strcat(str, pcmk__readable_action(task, interval_ms), " on ",
node_name, NULL);
if (status == PCMK_EXEC_DONE) {
pcmk__g_strcat(str, " returned '", services_ocf_exitcode_str(rc), "'",
NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, " (", exit_reason, ")", NULL);
}
} else {
pcmk__g_strcat(str, " could not be executed (",
pcmk_exec_status_str(status), NULL);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ": ", exit_reason, NULL);
}
g_string_append_c(str, ')');
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *s = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str, " at ", s, NULL);
free(s);
}
if (!pcmk__str_empty(exec_time)) {
int exec_time_ms = 0;
if ((pcmk__scan_min_int(exec_time, &exec_time_ms, 0) == pcmk_rc_ok)
&& (exec_time_ms > 0)) {
pcmk__g_strcat(str, " after ",
pcmk__readable_interval(exec_time_ms), NULL);
}
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
free(rsc_id);
free(task);
}
/*!
* \internal
* \brief Display a failed action with technical details
*
* \param[in,out] out Output object to use for display
* \param[in] xml_op XML containing failed action
* \param[in] op_key Operation key of failed action
* \param[in] node_name Where failed action occurred
* \param[in] rc OCF exit code of failed action
* \param[in] status Execution status of failed action
* \param[in] exit_reason Exit reason given for failed action
* \param[in] exec_time String containing execution time in milliseconds
*/
static void
failed_action_technical(pcmk__output_t *out, const xmlNode *xml_op,
const char *op_key, const char *node_name, int rc,
int status, const char *exit_reason,
const char *exec_time)
{
const char *call_id = crm_element_value(xml_op, XML_LRM_ATTR_CALLID);
const char *queue_time = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
const char *exit_status = services_ocf_exitcode_str(rc);
const char *lrm_status = pcmk_exec_status_str(status);
time_t last_change_epoch = 0;
GString *str = NULL;
if (pcmk__str_empty(op_key)) {
op_key = "unknown operation";
}
if (pcmk__str_empty(exit_status)) {
exit_status = "unknown exit status";
}
if (pcmk__str_empty(call_id)) {
call_id = "unknown";
}
str = g_string_sized_new(256);
g_string_append_printf(str, "%s on %s '%s' (%d): call=%s, status='%s'",
op_key, node_name, exit_status, rc, call_id,
lrm_status);
if (!pcmk__str_empty(exit_reason)) {
pcmk__g_strcat(str, ", exitreason='", exit_reason, "'", NULL);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change_epoch) == pcmk_ok) {
char *last_change_str = pcmk__epoch2str(&last_change_epoch, 0);
pcmk__g_strcat(str,
", " XML_RSC_OP_LAST_CHANGE "="
"'", last_change_str, "'", NULL);
free(last_change_str);
}
if (!pcmk__str_empty(queue_time)) {
pcmk__g_strcat(str, ", queued=", queue_time, "ms", NULL);
}
if (!pcmk__str_empty(exec_time)) {
pcmk__g_strcat(str, ", exec=", exec_time, "ms", NULL);
}
out->list_item(out, NULL, "%s", str->str);
g_string_free(str, TRUE);
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t")
static int
failed_action_default(pcmk__output_t *out, va_list args)
{
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts = va_arg(args, uint32_t);
const char *op_key = pe__xe_history_key(xml_op);
const char *node_name = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *exit_reason = crm_element_value(xml_op,
XML_LRM_ATTR_EXIT_REASON);
const char *exec_time = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
int rc;
int status;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_RC), &rc, 0);
- pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
- &status, 0);
+ pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
+ 0);
if (pcmk__str_empty(node_name)) {
node_name = "unknown node";
}
if (pcmk_is_set(show_opts, pcmk_show_failed_detail)) {
failed_action_technical(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
} else {
failed_action_friendly(out, xml_op, op_key, node_name, rc, status,
exit_reason, exec_time);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t")
static int
failed_action_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
const char *op_key = pe__xe_history_key(xml_op);
const char *op_key_name = "op_key";
int rc;
int status;
const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON);
time_t epoch = 0;
char *rc_s = NULL;
char *reason_s = crm_xml_escape(exit_reason ? exit_reason : "none");
xmlNodePtr node = NULL;
pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_RC), &rc, 0);
- pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
- &status, 0);
+ pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
+ 0);
rc_s = pcmk__itoa(rc);
if (crm_element_value(xml_op, PCMK__XA_OPERATION_KEY) == NULL) {
op_key_name = PCMK_XA_ID;
}
node = pcmk__output_create_xml_node(out, "failure",
op_key_name, op_key,
"node", uname,
"exitstatus", services_ocf_exitcode_str(rc),
"exitreason", pcmk__s(reason_s, ""),
"exitcode", rc_s,
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"status", pcmk_exec_status_str(status),
NULL);
free(rc_s);
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
guint interval_ms = 0;
char *interval_ms_s = NULL;
char *rc_change = pcmk__epoch2str(&epoch,
crm_time_log_date
|crm_time_log_timeofday
|crm_time_log_with_timezone);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
interval_ms_s = crm_strdup_printf("%u", interval_ms);
pcmk__xe_set_props(node, XML_RSC_OP_LAST_CHANGE, rc_change,
"queued", crm_element_value(xml_op, XML_RSC_OP_T_QUEUE),
"exec", crm_element_value(xml_op, XML_RSC_OP_T_EXEC),
"interval", interval_ms_s,
"task", crm_element_value(xml_op, PCMK_XA_OPERATION),
NULL);
free(interval_ms_s);
free(rc_change);
}
free(reason_s);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("failed-action-list", "pcmk_scheduler_t *", "GList *",
"GList *", "uint32_t", "bool")
static int
failed_action_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *xml_op = NULL;
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(scheduler->failed) == 0) {
return rc;
}
for (xml_op = pcmk__xml_first_child(scheduler->failed); xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
char *rsc = NULL;
if (!pcmk__str_in_list(crm_element_value(xml_op, PCMK_XA_UNAME),
only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
if (pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
if (!parse_op_key(pe__xe_history_key(xml_op), &rsc, NULL, NULL)) {
continue;
}
if (!pcmk__str_in_list(rsc, only_rsc, pcmk__str_star_matches)) {
free(rsc);
continue;
}
free(rsc);
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Failed Resource Actions");
out->message(out, "failed-action", xml_op, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
static void
status_node(pcmk_node_t *node, xmlNodePtr parent, uint32_t show_opts)
{
int health = pe__node_health(node);
// Cluster membership
if (node->details->online) {
pcmk_create_html_node(parent, "span", NULL, "online", " online");
} else {
pcmk_create_html_node(parent, "span", NULL, "offline", " OFFLINE");
}
// Standby mode
if (node->details->standby_onfail && (node->details->running_rsc != NULL)) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby due to on-fail,"
" with active resources)");
} else if (node->details->standby_onfail) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby due to on-fail)");
} else if (node->details->standby && (node->details->running_rsc != NULL)) {
pcmk_create_html_node(parent, "span", NULL, "standby",
" (in standby, with active resources)");
} else if (node->details->standby) {
pcmk_create_html_node(parent, "span", NULL, "standby", " (in standby)");
}
// Maintenance mode
if (node->details->maintenance) {
pcmk_create_html_node(parent, "span", NULL, "maint",
" (in maintenance mode)");
}
// Node health
if (health < 0) {
pcmk_create_html_node(parent, "span", NULL, "health_red",
" (health is RED)");
} else if (health == 0) {
pcmk_create_html_node(parent, "span", NULL, "health_yellow",
" (health is YELLOW)");
}
// Feature set
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
char *buf = crm_strdup_printf(", feature set %s", feature_set);
pcmk_create_html_node(parent, "span", NULL, NULL, buf);
free(buf);
}
}
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool",
"GList *", "GList *")
static int
node_html(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (full) {
xmlNodePtr item_node;
if (pcmk_all_flags_set(show_opts, pcmk_show_brief | pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:");
status_node(node, item_node, show_opts);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
}
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc2 = NULL;
int rc = pcmk_rc_no_output;
out->begin_list(out, NULL, NULL, "%s:", node_name);
item_node = pcmk__output_xml_create_parent(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:");
status_node(node, item_node, show_opts);
for (lpc2 = node->details->running_rsc; lpc2 != NULL; lpc2 = lpc2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc2->data;
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources");
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
pcmk__output_xml_pop_parent(out);
out->end_list(out);
} else {
char *buf = crm_strdup_printf("%s:", node_name);
item_node = pcmk__output_create_xml_node(out, "li", NULL);
pcmk_create_html_node(item_node, "span", NULL, "bold", buf);
status_node(node, item_node, show_opts);
free(buf);
}
} else {
out->begin_list(out, NULL, NULL, "%s:", node_name);
}
free(node_name);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Get a human-friendly textual description of a node's status
*
* \param[in] node Node to check
*
* \return String representation of node's status
*/
static const char *
node_text_status(const pcmk_node_t *node)
{
if (node->details->unclean) {
if (node->details->online) {
return "UNCLEAN (online)";
} else if (node->details->pending) {
return "UNCLEAN (pending)";
} else {
return "UNCLEAN (offline)";
}
} else if (node->details->pending) {
return "pending";
} else if (node->details->standby_onfail && node->details->online) {
return "standby (on-fail)";
} else if (node->details->standby) {
if (node->details->online) {
if (node->details->running_rsc) {
return "standby (with active resources)";
} else {
return "standby";
}
} else {
return "OFFLINE (standby)";
}
} else if (node->details->maintenance) {
if (node->details->online) {
return "maintenance";
} else {
return "OFFLINE (maintenance)";
}
} else if (node->details->online) {
return "online";
}
return "OFFLINE";
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_text(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
GString *str = g_string_sized_new(64);
int health = pe__node_health(node);
// Create a summary line with node type, name, and status
if (pe__is_guest_node(node)) {
g_string_append(str, "GuestNode");
} else if (pe__is_remote_node(node)) {
g_string_append(str, "RemoteNode");
} else {
g_string_append(str, "Node");
}
pcmk__g_strcat(str, " ", node_name, ": ", node_text_status(node), NULL);
if (health < 0) {
g_string_append(str, " (health is RED)");
} else if (health == 0) {
g_string_append(str, " (health is YELLOW)");
}
if (pcmk_is_set(show_opts, pcmk_show_feature_set)) {
const char *feature_set = get_node_feature_set(node);
if (feature_set != NULL) {
pcmk__g_strcat(str, ", feature set ", feature_set, NULL);
}
}
/* If we're grouping by node, print its resources */
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (pcmk_is_set(show_opts, pcmk_show_brief)) {
GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc);
if (rscs != NULL) {
uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
pe__rscs_brief_output(out, rscs, new_show_opts);
out->end_list(out);
out->end_list(out);
g_list_free(rscs);
}
} else {
GList *gIter2 = NULL;
out->begin_list(out, NULL, NULL, "%s", str->str);
out->begin_list(out, NULL, NULL, "Resources");
for (gIter2 = node->details->running_rsc; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter2->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
out->end_list(out);
out->end_list(out);
}
} else {
out->list_item(out, NULL, "%s", str->str);
}
g_string_free(str, TRUE);
free(node_name);
} else {
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
out->begin_list(out, NULL, NULL, "Node: %s", node_name);
free(node_name);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *",
"GList *")
static int
node_xml(pcmk__output_t *out, va_list args) {
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t);
bool full = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
if (full) {
const char *node_type = "unknown";
char *length_s = pcmk__itoa(g_list_length(node->details->running_rsc));
int health = pe__node_health(node);
const char *health_s = NULL;
const char *feature_set;
switch (node->details->type) {
case pcmk_node_variant_cluster:
node_type = "member";
break;
case pcmk_node_variant_remote:
node_type = "remote";
break;
case node_ping:
node_type = "ping";
break;
}
if (health < 0) {
health_s = "red";
} else if (health == 0) {
health_s = "yellow";
} else {
health_s = "green";
}
feature_set = get_node_feature_set(node);
pe__name_and_nvpairs_xml(out, true, "node", 15,
PCMK_XA_NAME, node->details->uname,
PCMK_XA_ID, node->details->id,
"online", pcmk__btoa(node->details->online),
"standby", pcmk__btoa(node->details->standby),
"standby_onfail", pcmk__btoa(node->details->standby_onfail),
"maintenance", pcmk__btoa(node->details->maintenance),
"pending", pcmk__btoa(node->details->pending),
"unclean", pcmk__btoa(node->details->unclean),
"health", health_s,
"feature_set", feature_set,
"shutdown", pcmk__btoa(node->details->shutdown),
"expected_up", pcmk__btoa(node->details->expected_up),
"is_dc", pcmk__btoa(node->details->is_dc),
"resources_running", length_s,
PCMK_XA_TYPE, node_type);
if (pe__is_guest_node(node)) {
xmlNodePtr xml_node = pcmk__output_xml_peek_parent(out);
crm_xml_add(xml_node, "id_as_resource", node->details->remote_rsc->container->id);
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *lpc = NULL;
for (lpc = node->details->running_rsc; lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) lpc->data;
show_opts |= pcmk_show_rsc_only;
out->message(out, crm_map_element_name(rsc->xml), show_opts,
rsc, only_node, only_rsc);
}
}
free(length_s);
out->end_list(out);
} else {
pcmk__output_xml_create_parent(out, "node",
PCMK_XA_NAME, node->details->uname,
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_text(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v;
if (value == NULL) {
v = 0;
} else {
pcmk__scan_min_int(value, &v, INT_MIN);
}
if (v <= 0) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is lost", name, value);
} else if (v < expected_score) {
out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is degraded (Expected=%d)", name, value, expected_score);
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
} else {
out->list_item(out, NULL, "%-32s\t: %-10s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_html(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
if (add_extra) {
int v;
char *s = crm_strdup_printf("%s: %s", name, value);
xmlNodePtr item_node = pcmk__output_create_xml_node(out, "li", NULL);
if (value == NULL) {
v = 0;
} else {
pcmk__scan_min_int(value, &v, INT_MIN);
}
pcmk_create_html_node(item_node, "span", NULL, NULL, s);
free(s);
if (v <= 0) {
pcmk_create_html_node(item_node, "span", NULL, "bold", "(connectivity is lost)");
} else if (v < expected_score) {
char *buf = crm_strdup_printf("(connectivity is degraded -- expected %d", expected_score);
pcmk_create_html_node(item_node, "span", NULL, "bold", buf);
free(buf);
}
} else {
out->list_item(out, NULL, "%s: %s", name, value);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr")
static int
node_and_op(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
gchar *node_str = NULL;
char *last_change_str = NULL;
const char *op_rsc = crm_element_value(xml_op, "resource");
int status;
time_t last_change = 0;
- pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
- &status, PCMK_EXEC_UNKNOWN);
+ pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status,
+ PCMK_EXEC_UNKNOWN);
rsc = pe_find_resource(scheduler->resources, op_rsc);
if (rsc) {
const pcmk_node_t *node = pe__current_node(rsc);
const char *target_role = g_hash_table_lookup(rsc->meta,
PCMK_META_TARGET_ROLE);
uint32_t show_opts = pcmk_show_rsc_only | pcmk_show_pending;
if (node == NULL) {
node = rsc->pending_node;
}
node_str = pcmk__native_output_string(rsc, rsc_printable_id(rsc), node,
show_opts, target_role, false);
} else {
node_str = crm_strdup_printf("Unknown resource %s", op_rsc);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change) == pcmk_ok) {
last_change_str = crm_strdup_printf(", %s='%s', exec=%sms",
XML_RSC_OP_LAST_CHANGE,
pcmk__trim(ctime(&last_change)),
crm_element_value(xml_op, XML_RSC_OP_T_EXEC));
}
out->list_item(out, NULL, "%s: %s (node=%s, call=%s, rc=%s%s): %s",
node_str, pe__xe_history_key(xml_op),
crm_element_value(xml_op, PCMK_XA_UNAME),
crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
crm_element_value(xml_op, XML_LRM_ATTR_RC),
last_change_str ? last_change_str : "",
pcmk_exec_status_str(status));
g_free(node_str);
free(last_change_str);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr")
static int
node_and_op_xml(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
pcmk_resource_t *rsc = NULL;
const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME);
const char *op_rsc = crm_element_value(xml_op, "resource");
int status;
time_t last_change = 0;
xmlNode *node = NULL;
- pcmk__scan_min_int(crm_element_value(xml_op, XML_LRM_ATTR_OPSTATUS),
+ pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS),
&status, PCMK_EXEC_UNKNOWN);
node = pcmk__output_create_xml_node(out, PCMK_XE_OPERATION,
PCMK_XA_OP, pe__xe_history_key(xml_op),
"node", uname,
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"rc", crm_element_value(xml_op, XML_LRM_ATTR_RC),
"status", pcmk_exec_status_str(status),
NULL);
rsc = pe_find_resource(scheduler->resources, op_rsc);
if (rsc) {
const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
const char *provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER);
const char *kind = crm_element_value(rsc->xml, PCMK_XA_TYPE);
bool has_provider = pcmk_is_set(pcmk_get_ra_caps(class),
pcmk_ra_cap_provider);
char *agent_tuple = crm_strdup_printf("%s:%s:%s",
class,
(has_provider? provider : ""),
kind);
pcmk__xe_set_props(node, "rsc", rsc_printable_id(rsc),
"agent", agent_tuple,
NULL);
free(agent_tuple);
}
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&last_change) == pcmk_ok) {
pcmk__xe_set_props(node, XML_RSC_OP_LAST_CHANGE,
pcmk__trim(ctime(&last_change)),
XML_RSC_OP_T_EXEC, crm_element_value(xml_op, XML_RSC_OP_T_EXEC),
NULL);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int")
static int
node_attribute_xml(pcmk__output_t *out, va_list args) {
const char *name = va_arg(args, const char *);
const char *value = va_arg(args, const char *);
bool add_extra = va_arg(args, int);
int expected_score = va_arg(args, int);
xmlNodePtr node = pcmk__output_create_xml_node(out, "attribute",
PCMK_XA_NAME, name,
PCMK_XA_VALUE, value,
NULL);
if (add_extra) {
char *buf = pcmk__itoa(expected_score);
crm_xml_add(node, "expected", buf);
free(buf);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-attribute-list", "pcmk_scheduler_t *", "uint32_t",
"bool", "GList *", "GList *")
static int
node_attribute_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
int rc = pcmk_rc_no_output;
/* Display each node's attributes */
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = gIter->data;
GList *attr_list = NULL;
GHashTableIter iter;
gpointer key;
if (!node || !node->details || !node->details->online) {
continue;
}
g_hash_table_iter_init(&iter, node->details->attrs);
while (g_hash_table_iter_next (&iter, &key, NULL)) {
attr_list = filter_attr_list(attr_list, key);
}
if (attr_list == NULL) {
continue;
}
if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) {
g_list_free(attr_list);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node Attributes");
out->message(out, "node", node, show_opts, false, only_node, only_rsc);
for (GList *aIter = attr_list; aIter != NULL; aIter = aIter->next) {
const char *name = aIter->data;
const char *value = NULL;
int expected_score = 0;
bool add_extra = false;
value = pe_node_attribute_raw(node, name);
add_extra = add_extra_info(node, node->details->running_rsc,
scheduler, name, &expected_score);
/* Print attribute name and value */
out->message(out, "node-attribute", name, value, add_extra,
expected_score);
}
g_list_free(attr_list);
out->end_list(out);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *comment = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s capacity:",
comment, pe__node_name(node));
g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *")
static int
node_capacity_xml(pcmk__output_t *out, va_list args)
{
const pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *comment = va_arg(args, const char *);
xmlNodePtr xml_node = pcmk__output_create_xml_node(out, "capacity",
"node", node->details->uname,
"comment", comment,
NULL);
g_hash_table_foreach(node->details->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-history-list", "pcmk_scheduler_t *", "pcmk_node_t *",
"xmlNodePtr", "GList *", "GList *", "uint32_t", "uint32_t")
static int
node_history_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
xmlNode *node_state = va_arg(args, xmlNode *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
xmlNode *lrm_rsc = NULL;
xmlNode *rsc_entry = NULL;
int rc = pcmk_rc_no_output;
lrm_rsc = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
lrm_rsc = find_xml_node(lrm_rsc, XML_LRM_TAG_RESOURCES, FALSE);
/* Print history of each of the node's resources */
for (rsc_entry = first_named_child(lrm_rsc, XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, rsc_id);
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
/* We can't use is_filtered here to filter group resources. For is_filtered,
* we have to decide whether to check the parent or not. If we check the
* parent, all elements of a group will always be printed because that's how
* is_filtered works for groups. If we do not check the parent, sometimes
* this will filter everything out.
*
* For other resource types, is_filtered is okay.
*/
if (parent->variant == pcmk_rsc_variant_group) {
if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc,
pcmk__str_star_matches)
&& !pcmk__str_in_list(rsc_printable_id(parent), only_rsc,
pcmk__str_star_matches)) {
continue;
}
} else {
if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) {
continue;
}
}
if (!pcmk_is_set(section_opts, pcmk_section_operations)) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
if (failcount <= 0) {
continue;
}
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-history", rsc, rsc_id, false,
failcount, last_failure, false);
} else {
GList *op_list = get_operation_list(rsc_entry);
pcmk_resource_t *rsc = NULL;
if (op_list == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->resources,
crm_element_value(rsc_entry, PCMK_XA_ID));
if (rc == pcmk_rc_no_output) {
rc = pcmk_rc_ok;
out->message(out, "node", node, show_opts, false, only_node,
only_rsc);
}
out->message(out, "resource-operation-list", scheduler, rsc, node,
op_list, show_opts);
}
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_html(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Node List");
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_text(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
/* space-separated lists of node names */
GString *online_nodes = NULL;
GString *online_remote_nodes = NULL;
GString *online_guest_nodes = NULL;
GString *offline_nodes = NULL;
GString *offline_remote_nodes = NULL;
int rc = pcmk_rc_no_output;
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id));
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
free(node_name);
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node List");
// Determine whether to display node individually or in a list
if (node->details->unclean || node->details->pending
|| (node->details->standby_onfail && node->details->online)
|| node->details->standby || node->details->maintenance
|| pcmk_is_set(show_opts, pcmk_show_rscs_by_node)
|| pcmk_is_set(show_opts, pcmk_show_feature_set)
|| (pe__node_health(node) <= 0)) {
// Display node individually
} else if (node->details->online) {
// Display online node in a list
if (pe__is_guest_node(node)) {
pcmk__add_word(&online_guest_nodes, 1024, node_name);
} else if (pe__is_remote_node(node)) {
pcmk__add_word(&online_remote_nodes, 1024, node_name);
} else {
pcmk__add_word(&online_nodes, 1024, node_name);
}
free(node_name);
continue;
} else {
// Display offline node in a list
if (pe__is_remote_node(node)) {
pcmk__add_word(&offline_remote_nodes, 1024, node_name);
} else if (pe__is_guest_node(node)) {
/* ignore offline guest nodes */
} else {
pcmk__add_word(&offline_nodes, 1024, node_name);
}
free(node_name);
continue;
}
/* If we get here, node is in bad state, or we're grouping by node */
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
free(node_name);
}
/* If we're not grouping by node, summarize nodes by status */
if (online_nodes != NULL) {
out->list_item(out, "Online", "[ %s ]",
(const char *) online_nodes->str);
g_string_free(online_nodes, TRUE);
}
if (offline_nodes != NULL) {
out->list_item(out, "OFFLINE", "[ %s ]",
(const char *) offline_nodes->str);
g_string_free(offline_nodes, TRUE);
}
if (online_remote_nodes) {
out->list_item(out, "RemoteOnline", "[ %s ]",
(const char *) online_remote_nodes->str);
g_string_free(online_remote_nodes, TRUE);
}
if (offline_remote_nodes) {
out->list_item(out, "RemoteOFFLINE", "[ %s ]",
(const char *) offline_remote_nodes->str);
g_string_free(offline_remote_nodes, TRUE);
}
if (online_guest_nodes != NULL) {
out->list_item(out, "GuestOnline", "[ %s ]",
(const char *) online_guest_nodes->str);
g_string_free(online_guest_nodes, TRUE);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool")
static int
node_list_xml(pcmk__output_t *out, va_list args) {
GList *nodes = va_arg(args, GList *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer G_GNUC_UNUSED = va_arg(args, int);
out->begin_list(out, NULL, NULL, "nodes");
for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node = (pcmk_node_t *) gIter->data;
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
out->message(out, "node", node, show_opts, true, only_node, only_rsc);
}
out->end_list(out);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-summary", "pcmk_scheduler_t *", "GList *", "GList *",
"uint32_t", "uint32_t", "bool")
static int
node_summary(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
uint32_t section_opts = va_arg(args, uint32_t);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_spacer = va_arg(args, int);
xmlNode *node_state = NULL;
xmlNode *cib_status = pcmk_find_cib_element(scheduler->input,
XML_CIB_TAG_STATUS);
int rc = pcmk_rc_no_output;
if (xmlChildElementCount(cib_status) == 0) {
return rc;
}
for (node_state = first_named_child(cib_status, XML_CIB_TAG_STATE);
node_state != NULL; node_state = crm_next_same_xml(node_state)) {
pcmk_node_t *node = pe_find_node_id(scheduler->nodes, ID(node_state));
if (!node || !node->details || !node->details->online) {
continue;
}
if (!pcmk__str_in_list(node->details->uname, only_node,
pcmk__str_star_matches|pcmk__str_casei)) {
continue;
}
PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc,
pcmk_is_set(section_opts, pcmk_section_operations) ? "Operations" : "Migration Summary");
out->message(out, "node-history-list", scheduler, node, node_state,
only_node, only_rsc, section_opts, show_opts);
}
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
if (rsc) {
out->list_item(out, NULL, "%s: %s allocation score on %s: %s",
prefix, rsc->id, uname, score);
} else {
out->list_item(out, NULL, "%s: %s = %s", prefix, uname, score);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *",
"const char *", "const char *")
static int
node_weight_xml(pcmk__output_t *out, va_list args)
{
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
const char *prefix = va_arg(args, const char *);
const char *uname = va_arg(args, const char *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, "node_weight",
"function", prefix,
"node", uname,
"score", score,
NULL);
if (rsc) {
crm_xml_add(node, PCMK_XA_ID, rsc->id);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t")
static int
op_history_text(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
char *buf = op_history_string(xml_op, task, interval_ms_s, rc,
pcmk_is_set(show_opts, pcmk_show_timing));
out->list_item(out, NULL, "%s", buf);
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t")
static int
op_history_xml(pcmk__output_t *out, va_list args) {
xmlNodePtr xml_op = va_arg(args, xmlNodePtr);
const char *task = va_arg(args, const char *);
const char *interval_ms_s = va_arg(args, const char *);
int rc = va_arg(args, int);
uint32_t show_opts = va_arg(args, uint32_t);
char *rc_s = pcmk__itoa(rc);
xmlNodePtr node = pcmk__output_create_xml_node(out, "operation_history",
"call", crm_element_value(xml_op, XML_LRM_ATTR_CALLID),
"task", task,
"rc", rc_s,
"rc_text", services_ocf_exitcode_str(rc),
NULL);
free(rc_s);
if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) {
char *s = crm_strdup_printf("%sms", interval_ms_s);
crm_xml_add(node, "interval", s);
free(s);
}
if (pcmk_is_set(show_opts, pcmk_show_timing)) {
const char *value = NULL;
time_t epoch = 0;
if ((crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&epoch) == pcmk_ok) && (epoch > 0)) {
char *s = pcmk__epoch2str(&epoch, 0);
crm_xml_add(node, XML_RSC_OP_LAST_CHANGE, s);
free(s);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_EXEC);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, XML_RSC_OP_T_EXEC, s);
free(s);
}
value = crm_element_value(xml_op, XML_RSC_OP_T_QUEUE);
if (value) {
char *s = crm_strdup_printf("%sms", value);
crm_xml_add(node, XML_RSC_OP_T_QUEUE, s);
free(s);
}
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
out->list_item(out, NULL, "%s promotion score on %s: %s",
child_rsc->id,
chosen? chosen->details->uname : "none",
score);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
promotion_score_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *chosen = va_arg(args, pcmk_node_t *);
const char *score = va_arg(args, const char *);
xmlNodePtr node = pcmk__output_create_xml_node(out, "promotion_score",
PCMK_XA_ID, child_rsc->id,
"score", score,
NULL);
if (chosen) {
crm_xml_add(node, "node", chosen->details->uname);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config(pcmk__output_t *out, va_list args) {
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
bool raw = va_arg(args, int);
char *rsc_xml = formatted_xml_buf(rsc, raw);
out->output_xml(out, "xml", rsc_xml);
free(rsc_xml);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool")
static int
resource_config_text(pcmk__output_t *out, va_list args) {
const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *);
bool raw = va_arg(args, int);
char *rsc_xml = formatted_xml_buf(rsc, raw);
pcmk__formatted_printf(out, "Resource XML:\n");
out->output_xml(out, "xml", rsc_xml);
free(rsc_xml);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_text(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
char *buf = resource_history_string(rsc, rsc_id, all, failcount, last_failure);
if (as_header) {
out->begin_list(out, NULL, NULL, "%s", buf);
} else {
out->list_item(out, NULL, "%s", buf);
}
free(buf);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *",
"bool", "int", "time_t", "bool")
static int
resource_history_xml(pcmk__output_t *out, va_list args) {
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
const char *rsc_id = va_arg(args, const char *);
bool all = va_arg(args, int);
int failcount = va_arg(args, int);
time_t last_failure = va_arg(args, time_t);
bool as_header = va_arg(args, int);
xmlNodePtr node = pcmk__output_xml_create_parent(out, "resource_history",
PCMK_XA_ID, rsc_id,
NULL);
if (rsc == NULL) {
pcmk__xe_set_bool_attr(node, "orphan", true);
} else if (all || failcount || last_failure > 0) {
char *migration_s = pcmk__itoa(rsc->migration_threshold);
pcmk__xe_set_props(node, "orphan", PCMK_VALUE_FALSE,
PCMK_META_MIGRATION_THRESHOLD, migration_s,
NULL);
free(migration_s);
if (failcount > 0) {
char *s = pcmk__itoa(failcount);
crm_xml_add(node, PCMK__FAIL_COUNT_PREFIX, s);
free(s);
}
if (last_failure > 0) {
char *s = pcmk__epoch2str(&last_failure, 0);
crm_xml_add(node, PCMK__LAST_FAILURE_PREFIX, s);
free(s);
}
}
if (!as_header) {
pcmk__output_xml_pop_parent(out);
}
return pcmk_rc_ok;
}
static void
print_resource_header(pcmk__output_t *out, uint32_t show_opts)
{
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
/* Active resources have already been printed by node */
out->begin_list(out, NULL, NULL, "Inactive Resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->begin_list(out, NULL, NULL, "Full List of Resources");
} else {
out->begin_list(out, NULL, NULL, "Active Resources");
}
}
PCMK__OUTPUT_ARGS("resource-list", "pcmk_scheduler_t *", "uint32_t", "bool",
"GList *", "GList *", "bool")
static int
resource_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
uint32_t show_opts = va_arg(args, uint32_t);
bool print_summary = va_arg(args, int);
GList *only_node = va_arg(args, GList *);
GList *only_rsc = va_arg(args, GList *);
bool print_spacer = va_arg(args, int);
GList *rsc_iter;
int rc = pcmk_rc_no_output;
bool printed_header = false;
/* If we already showed active resources by node, and
* we're not showing inactive resources, we have nothing to do
*/
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node) &&
!pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
return rc;
}
/* If we haven't already printed resources grouped by node,
* and brief output was requested, print resource summary */
if (pcmk_is_set(show_opts, pcmk_show_brief)
&& !pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
GList *rscs = pe__filter_rsc_list(scheduler->resources, only_rsc);
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
rc = pe__rscs_brief_output(out, rscs, show_opts);
g_list_free(rscs);
}
/* For each resource, display it if appropriate */
for (rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) rsc_iter->data;
int x;
/* Complex resources may have some sub-resources active and some inactive */
gboolean is_active = rsc->fns->active(rsc, TRUE);
gboolean partially_active = rsc->fns->active(rsc, FALSE);
/* Skip inactive orphans (deleted but still in CIB) */
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed) && !is_active) {
continue;
/* Skip active resources if we already displayed them by node */
} else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
if (is_active) {
continue;
}
/* Skip primitives already counted in a brief summary */
} else if (pcmk_is_set(show_opts, pcmk_show_brief)
&& (rsc->variant == pcmk_rsc_variant_primitive)) {
continue;
/* Skip resources that aren't at least partially active,
* unless we're displaying inactive resources
*/
} else if (!partially_active && !pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
continue;
} else if (partially_active && !pe__rsc_running_on_any(rsc, only_node)) {
continue;
}
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
/* Print this resource */
x = out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc,
only_node, only_rsc);
if (x == pcmk_rc_ok) {
rc = pcmk_rc_ok;
}
}
if (print_summary && rc != pcmk_rc_ok) {
if (!printed_header) {
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
print_resource_header(out, show_opts);
printed_header = true;
}
if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) {
out->list_item(out, NULL, "No inactive resources");
} else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) {
out->list_item(out, NULL, "No resources");
} else {
out->list_item(out, NULL, "No active resources");
}
}
if (printed_header) {
out->end_list(out);
}
return rc;
}
PCMK__OUTPUT_ARGS("resource-operation-list", "pcmk_scheduler_t *",
"pcmk_resource_t *", "pcmk_node_t *", "GList *", "uint32_t")
static int
resource_operation_list(pcmk__output_t *out, va_list args)
{
pcmk_scheduler_t *scheduler G_GNUC_UNUSED = va_arg(args,
pcmk_scheduler_t *);
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
GList *op_list = va_arg(args, GList *);
uint32_t show_opts = va_arg(args, uint32_t);
GList *gIter = NULL;
int rc = pcmk_rc_no_output;
/* Print each operation */
for (gIter = op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *xml_op = (xmlNode *) gIter->data;
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
const char *interval_ms_s = crm_element_value(xml_op,
PCMK_META_INTERVAL);
const char *op_rc = crm_element_value(xml_op, XML_LRM_ATTR_RC);
int op_rc_i;
pcmk__scan_min_int(op_rc, &op_rc_i, 0);
/* Display 0-interval monitors as "probe" */
if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) {
task = "probe";
}
/* If this is the first printed operation, print heading for resource */
if (rc == pcmk_rc_no_output) {
time_t last_failure = 0;
int failcount = pe_get_failcount(node, rsc, &last_failure,
pcmk__fc_default, NULL);
out->message(out, "resource-history", rsc, rsc_printable_id(rsc), true,
failcount, last_failure, true);
rc = pcmk_rc_ok;
}
/* Print the operation */
out->message(out, "op-history", xml_op, task, interval_ms_s,
op_rc_i, show_opts);
}
/* Free the list we created (no need to free the individual items) */
g_list_free(op_list);
PCMK__OUTPUT_LIST_FOOTER(out, rc);
return rc;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *fn = va_arg(args, const char *);
char *dump_text = crm_strdup_printf("%s: %s utilization on %s:",
fn, rsc->id, pe__node_name(node));
g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text);
out->list_item(out, NULL, "%s", dump_text);
free(dump_text);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *",
"const char *")
static int
resource_util_xml(pcmk__output_t *out, va_list args)
{
pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *);
pcmk_node_t *node = va_arg(args, pcmk_node_t *);
const char *fn = va_arg(args, const char *);
xmlNodePtr xml_node = pcmk__output_create_xml_node(out, "utilization",
"resource", rsc->id,
"node", node->details->uname,
"function", fn,
NULL);
g_hash_table_foreach(rsc->utilization, add_dump_node, xml_node);
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_html(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
if (ticket->last_granted > -1) {
char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0);
out->list_item(out, NULL, "%s:\t%s%s %s=\"%s\"", ticket->id,
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "",
"last-granted", pcmk__s(epoch_str, ""));
free(epoch_str);
} else {
out->list_item(out, NULL, "%s:\t%s%s", ticket->id,
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_text(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
if (ticket->last_granted > -1) {
char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0);
out->list_item(out, ticket->id, "%s%s %s=\"%s\"",
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "",
"last-granted", pcmk__s(epoch_str, ""));
free(epoch_str);
} else {
out->list_item(out, ticket->id, "%s%s",
ticket->granted ? "granted" : "revoked",
ticket->standby ? " [standby]" : "");
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *")
static int
ticket_xml(pcmk__output_t *out, va_list args) {
pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *);
xmlNodePtr node = NULL;
node = pcmk__output_create_xml_node(out, "ticket",
PCMK_XA_ID, ticket->id,
"status", ticket->granted ? "granted" : "revoked",
"standby", pcmk__btoa(ticket->standby),
NULL);
if (ticket->last_granted > -1) {
char *buf = pcmk__epoch2str(&ticket->last_granted, 0);
crm_xml_add(node, "last-granted", buf);
free(buf);
}
return pcmk_rc_ok;
}
PCMK__OUTPUT_ARGS("ticket-list", "pcmk_scheduler_t *", "bool")
static int
ticket_list(pcmk__output_t *out, va_list args) {
pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *);
bool print_spacer = va_arg(args, int);
GHashTableIter iter;
gpointer key, value;
if (g_hash_table_size(scheduler->tickets) == 0) {
return pcmk_rc_no_output;
}
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
/* Print section heading */
out->begin_list(out, NULL, NULL, "Tickets");
/* Print each ticket */
g_hash_table_iter_init(&iter, scheduler->tickets);
while (g_hash_table_iter_next(&iter, &key, &value)) {
pcmk_ticket_t *ticket = (pcmk_ticket_t *) value;
out->message(out, "ticket", ticket);
}
/* Close section */
out->end_list(out);
return pcmk_rc_ok;
}
static pcmk__message_entry_t fmt_functions[] = {
{ "ban", "default", ban_text },
{ "ban", "html", ban_html },
{ "ban", "xml", ban_xml },
{ "ban-list", "default", ban_list },
{ "bundle", "default", pe__bundle_text },
{ "bundle", "xml", pe__bundle_xml },
{ "bundle", "html", pe__bundle_html },
{ "clone", "default", pe__clone_default },
{ "clone", "xml", pe__clone_xml },
{ "cluster-counts", "default", cluster_counts_text },
{ "cluster-counts", "html", cluster_counts_html },
{ "cluster-counts", "xml", cluster_counts_xml },
{ "cluster-dc", "default", cluster_dc_text },
{ "cluster-dc", "html", cluster_dc_html },
{ "cluster-dc", "xml", cluster_dc_xml },
{ "cluster-options", "default", cluster_options_text },
{ "cluster-options", "html", cluster_options_html },
{ "cluster-options", "log", cluster_options_log },
{ "cluster-options", "xml", cluster_options_xml },
{ "cluster-summary", "default", cluster_summary },
{ "cluster-summary", "html", cluster_summary_html },
{ "cluster-stack", "default", cluster_stack_text },
{ "cluster-stack", "html", cluster_stack_html },
{ "cluster-stack", "xml", cluster_stack_xml },
{ "cluster-times", "default", cluster_times_text },
{ "cluster-times", "html", cluster_times_html },
{ "cluster-times", "xml", cluster_times_xml },
{ "failed-action", "default", failed_action_default },
{ "failed-action", "xml", failed_action_xml },
{ "failed-action-list", "default", failed_action_list },
{ "group", "default", pe__group_default},
{ "group", "xml", pe__group_xml },
{ "maint-mode", "text", cluster_maint_mode_text },
{ "node", "default", node_text },
{ "node", "html", node_html },
{ "node", "xml", node_xml },
{ "node-and-op", "default", node_and_op },
{ "node-and-op", "xml", node_and_op_xml },
{ "node-capacity", "default", node_capacity },
{ "node-capacity", "xml", node_capacity_xml },
{ "node-history-list", "default", node_history_list },
{ "node-list", "default", node_list_text },
{ "node-list", "html", node_list_html },
{ "node-list", "xml", node_list_xml },
{ "node-weight", "default", node_weight },
{ "node-weight", "xml", node_weight_xml },
{ "node-attribute", "default", node_attribute_text },
{ "node-attribute", "html", node_attribute_html },
{ "node-attribute", "xml", node_attribute_xml },
{ "node-attribute-list", "default", node_attribute_list },
{ "node-summary", "default", node_summary },
{ "op-history", "default", op_history_text },
{ "op-history", "xml", op_history_xml },
{ "primitive", "default", pe__resource_text },
{ "primitive", "xml", pe__resource_xml },
{ "primitive", "html", pe__resource_html },
{ "promotion-score", "default", promotion_score },
{ "promotion-score", "xml", promotion_score_xml },
{ "resource-config", "default", resource_config },
{ "resource-config", "text", resource_config_text },
{ "resource-history", "default", resource_history_text },
{ "resource-history", "xml", resource_history_xml },
{ "resource-list", "default", resource_list },
{ "resource-operation-list", "default", resource_operation_list },
{ "resource-util", "default", resource_util },
{ "resource-util", "xml", resource_util_xml },
{ "ticket", "default", ticket_text },
{ "ticket", "html", ticket_html },
{ "ticket", "xml", ticket_xml },
{ "ticket-list", "default", ticket_list },
{ NULL, NULL, NULL }
};
void
pe__register_messages(pcmk__output_t *out) {
pcmk__register_messages(out, fmt_functions);
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index 690a70bc09..636a527ea8 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5128 +1,5127 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/services.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/util.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include <pe_status_private.h>
CRM_TRACE_INIT_DATA(pe_status);
// A (parsed) resource action history entry
struct action_history {
pcmk_resource_t *rsc; // Resource that history is for
pcmk_node_t *node; // Node that history is for
xmlNode *xml; // History entry XML
// Parsed from entry XML
const char *id; // XML ID of history entry
const char *key; // Operation key of action
const char *task; // Action name
const char *exit_reason; // Exit reason given for result
guint interval_ms; // Action interval
int call_id; // Call ID of action
int expected_exit_status; // Expected exit status of action
int exit_status; // Actual exit status of action
int execution_status; // Execution status of action
};
/* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than
* use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the
* flag is stringified more readably in log messages.
*/
#define set_config_flag(scheduler, option, flag) do { \
const char *scf_value = pe_pref((scheduler)->config_hash, (option)); \
if (scf_value != NULL) { \
if (crm_is_true(scf_value)) { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} else { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", \
crm_system_name, (scheduler)->flags, \
(flag), #flag); \
} \
} \
} while(0)
static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node,
xmlNode *xml_op, xmlNode **last_failure,
enum action_fail_response *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);
static gboolean
is_dangling_guest_node(pcmk_node_t *node)
{
/* we are looking for a remote-node that was supposed to be mapped to a
* container resource, but all traces of that container have disappeared
* from both the config and the status section. */
if (pe__is_guest_or_remote_node(node) &&
node->details->remote_rsc &&
node->details->remote_rsc->container == NULL &&
pcmk_is_set(node->details->remote_rsc->flags,
pcmk_rsc_removed_filler)) {
return TRUE;
}
return FALSE;
}
/*!
* \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);
/* A guest node is fenced by marking its container as failed */
if (pe__is_guest_node(node)) {
pcmk_resource_t *rsc = node->details->remote_rsc->container;
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",
pe__node_name(node), reason, rsc->id);
} else {
pcmk__sched_warn("Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pe__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.
*/
node->details->remote_requires_reset = TRUE;
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",
pe__node_name(node), reason);
pcmk__set_rsc_flags(node->details->remote_rsc,
pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
} else if (pe__is_remote_node(node)) {
pcmk_resource_t *rsc = node->details->remote_rsc;
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",
pe__node_name(node), reason);
} else if(node->details->remote_requires_reset == FALSE) {
node->details->remote_requires_reset = TRUE;
pcmk__sched_warn("Remote node %s %s: %s",
pe__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",
pe__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
pcmk__sched_warn("Cluster node %s %s: %s",
pe__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 XML_CIB_TAG_NVPAIR \
"[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \
"or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \
"and @" PCMK_XA_VALUE "='" PCMK__VALUE_UNFENCING "']"
// unfencing in rsc_defaults or any resource
#define XPATH_ENABLE_UNFENCING \
"/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \
"//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \
"|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \
"/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
xmlXPathObjectPtr result = NULL;
if (!pcmk_is_set(scheduler->flags, flag)) {
result = xpath_search(scheduler->input, xpath);
if (result && (numXpathResults(result) > 0)) {
pcmk__set_scheduler_flags(scheduler, flag);
}
freeXpathObject(result);
}
}
gboolean
unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler)
{
const char *value = NULL;
guint interval_ms = 0U;
GHashTable *config_hash = pcmk__strkey_table(free, free);
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
scheduler->config_hash = config_hash;
pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash,
CIB_OPTIONS_FIRST, FALSE, scheduler);
verify_pe_options(scheduler->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 = pe_pref(scheduler->config_hash, PCMK_OPT_HAVE_WATCHDOG);
if (value && crm_is_true(value)) {
crm_info("Watchdog-based self-fencing will be performed via SBD if "
"fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" is nonzero");
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing);
}
/* Set certain flags via xpath here, so they can be used before the relevant
* configuration sections are unpacked.
*/
set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING,
scheduler);
value = pe_pref(scheduler->config_hash, PCMK_OPT_STONITH_TIMEOUT);
pcmk_parse_interval_spec(value, &interval_ms);
if (interval_ms >= INT_MAX) {
scheduler->stonith_timeout = INT_MAX;
} else {
scheduler->stonith_timeout = (int) interval_ms;
}
crm_debug("STONITH timeout: %d", scheduler->stonith_timeout);
set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED,
pcmk_sched_fencing_enabled);
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
crm_debug("STONITH of failed nodes is enabled");
} else {
crm_debug("STONITH of failed nodes is disabled");
}
scheduler->stonith_action = pe_pref(scheduler->config_hash,
PCMK_OPT_STONITH_ACTION);
if (!strcmp(scheduler->stonith_action, "poweroff")) {
pcmk__warn_once(pcmk__wo_poweroff,
"Support for " PCMK_OPT_STONITH_ACTION " of "
"'poweroff' is deprecated and will be removed in a "
"future release (use 'off' instead)");
scheduler->stonith_action = PCMK_ACTION_OFF;
}
crm_trace("STONITH will %s nodes", scheduler->stonith_action);
set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING,
pcmk_sched_concurrent_fencing);
if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) {
crm_debug("Concurrent fencing is enabled");
} else {
crm_debug("Concurrent fencing is disabled");
}
value = pe_pref(scheduler->config_hash,
PCMK_OPT_PRIORITY_FENCING_DELAY);
if (value) {
pcmk_parse_interval_spec(value, &interval_ms);
scheduler->priority_fencing_delay = (int) (interval_ms / 1000);
crm_trace("Priority fencing delay is %ds",
scheduler->priority_fencing_delay);
}
set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES,
pcmk_sched_stop_all);
crm_debug("Stop all active resources: %s",
pcmk__btoa(pcmk_is_set(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 = pe_pref(scheduler->config_hash, PCMK_OPT_NO_QUORUM_POLICY);
if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_ignore;
} else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_freeze;
} else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) {
scheduler->no_quorum_policy = pcmk_no_quorum_demote;
} else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) {
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
int do_panic = 0;
crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC,
&do_panic);
if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
scheduler->no_quorum_policy = pcmk_no_quorum_fence;
} else {
crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop': cluster has never had quorum");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY
" to 'stop' because fencing is disabled");
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
} else {
scheduler->no_quorum_policy = pcmk_no_quorum_stop;
}
switch (scheduler->no_quorum_policy) {
case pcmk_no_quorum_freeze:
crm_debug("On loss of quorum: Freeze resources");
break;
case pcmk_no_quorum_stop:
crm_debug("On loss of quorum: Stop ALL resources");
break;
case pcmk_no_quorum_demote:
crm_debug("On loss of quorum: "
"Demote promotable resources and stop other resources");
break;
case pcmk_no_quorum_fence:
crm_notice("On loss of quorum: Fence all remaining nodes");
break;
case pcmk_no_quorum_ignore:
crm_notice("On loss of quorum: Ignore");
break;
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES,
pcmk_sched_stop_removed_resources);
if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
crm_trace("Orphan resources are stopped");
} else {
crm_trace("Orphan resources are ignored");
}
set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS,
pcmk_sched_cancel_removed_actions);
if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) {
crm_trace("Orphan resource actions are stopped");
} else {
crm_trace("Orphan resource actions are ignored");
}
value = pe_pref(scheduler->config_hash, PCMK__OPT_REMOVE_AFTER_STOP);
if (value != NULL) {
if (crm_is_true(value)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop);
#ifndef PCMK__COMPAT_2_0
pcmk__warn_once(pcmk__wo_remove_after,
"Support for the " PCMK__OPT_REMOVE_AFTER_STOP
" cluster property is deprecated and will be "
"removed in a future release");
#endif
} else {
pcmk__clear_scheduler_flags(scheduler,
pcmk_sched_remove_after_stop);
}
}
set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE,
pcmk_sched_in_maintenance);
crm_trace("Maintenance mode: %s",
pcmk__btoa(pcmk_is_set(scheduler->flags,
pcmk_sched_in_maintenance)));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk_sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk_sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->placement_strategy = pe_pref(scheduler->config_hash,
PCMK_OPT_PLACEMENT_STRATEGY);
crm_trace("Placement strategy: %s", scheduler->placement_strategy);
set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK,
pcmk_sched_shutdown_lock);
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
value = pe_pref(scheduler->config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT);
pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock));
scheduler->shutdown_lock /= 1000;
crm_trace("Resources will be locked to nodes that were cleanly "
"shut down (locks expire after %s)",
pcmk__readable_interval(scheduler->shutdown_lock));
} else {
crm_trace("Resources will not be locked to nodes that were cleanly "
"shut down");
}
value = pe_pref(scheduler->config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT);
pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout));
scheduler->node_pending_timeout /= 1000;
if (scheduler->node_pending_timeout == 0) {
crm_trace("Do not fence pending nodes");
} else {
crm_trace("Fence pending nodes after %s",
pcmk__readable_interval(scheduler->node_pending_timeout
* 1000));
}
return TRUE;
}
pcmk_node_t *
pe_create_node(const char *id, const char *uname, const char *type,
const char *score, pcmk_scheduler_t *scheduler)
{
pcmk_node_t *new_node = NULL;
if (pe_find_node(scheduler->nodes, uname) != NULL) {
pcmk__config_warn("More than one node entry has name '%s'", uname);
}
new_node = calloc(1, sizeof(pcmk_node_t));
if (new_node == NULL) {
pcmk__sched_err("Could not allocate memory for node %s", uname);
return NULL;
}
new_node->weight = char2score(score);
new_node->details = calloc(1, sizeof(struct pe_node_shared_s));
if (new_node->details == NULL) {
free(new_node);
pcmk__sched_err("Could not allocate memory for node %s", uname);
return NULL;
}
crm_trace("Creating node for entry %s/%s", uname, id);
new_node->details->id = id;
new_node->details->uname = uname;
new_node->details->online = FALSE;
new_node->details->shutdown = FALSE;
new_node->details->rsc_discovery_enabled = TRUE;
new_node->details->running_rsc = NULL;
new_node->details->data_set = scheduler;
if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) {
new_node->details->type = pcmk_node_variant_cluster;
} else if (pcmk__str_eq(type, "remote", pcmk__str_casei)) {
new_node->details->type = pcmk_node_variant_remote;
pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes);
} else {
/* @COMPAT 'ping' is the default for backward compatibility, but it
* should be changed to 'member' at a compatibility break
*/
if (!pcmk__str_eq(type, "ping", pcmk__str_casei)) {
pcmk__config_warn("Node %s has unrecognized type '%s', "
"assuming 'ping'", pcmk__s(uname, "without name"),
type);
}
pcmk__warn_once(pcmk__wo_ping_node,
"Support for nodes of type 'ping' (such as %s) is "
"deprecated and will be removed in a future release",
pcmk__s(uname, "unnamed node"));
new_node->details->type = node_ping;
}
new_node->details->attrs = pcmk__strkey_table(free, free);
if (pe__is_guest_or_remote_node(new_node)) {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("remote"));
} else {
g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("cluster"));
}
new_node->details->utilization = pcmk__strkey_table(free, free);
new_node->details->digest_cache = pcmk__strkey_table(free,
pe__free_digests);
scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node,
pe__cmp_node_name);
return new_node;
}
static const char *
expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data)
{
xmlNode *attr_set = NULL;
xmlNode *attr = NULL;
const char *container_id = ID(xml_obj);
const char *remote_name = NULL;
const char *remote_server = NULL;
const char *remote_port = NULL;
const char *connect_timeout = "60s";
const char *remote_allow_migrate=NULL;
const char *is_managed = NULL;
for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL;
attr_set = pcmk__xe_next(attr_set)) {
if (!pcmk__str_eq((const char *)attr_set->name, XML_TAG_META_SETS,
pcmk__str_casei)) {
continue;
}
for (attr = pcmk__xe_first_child(attr_set); attr != NULL;
attr = pcmk__xe_next(attr)) {
const char *value = crm_element_value(attr, PCMK_XA_VALUE);
const char *name = crm_element_value(attr, PCMK_XA_NAME);
if (name == NULL) { // Sanity
continue;
}
if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) {
remote_name = value;
} else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) {
remote_server = value;
} else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) {
remote_port = value;
} else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) {
connect_timeout = value;
} else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) {
remote_allow_migrate = value;
} else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) {
is_managed = value;
}
}
}
if (remote_name == NULL) {
return NULL;
}
if (pe_find_resource(data->resources, remote_name) != NULL) {
return NULL;
}
pe_create_remote_xml(parent, remote_name, container_id,
remote_allow_migrate, is_managed,
connect_timeout, remote_server, remote_port);
return remote_name;
}
static void
handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node)
{
if ((new_node->details->type == pcmk_node_variant_remote)
&& (new_node->details->remote_rsc == 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;
}
/* We need to be able to determine if a node's status section
* exists or not separate from whether the node is unclean. */
new_node->details->unseen = TRUE;
}
gboolean
unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
pcmk_node_t *new_node = NULL;
const char *id = NULL;
const char *uname = NULL;
const char *type = NULL;
const char *score = NULL;
for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) {
new_node = NULL;
id = crm_element_value(xml_obj, PCMK_XA_ID);
uname = crm_element_value(xml_obj, PCMK_XA_UNAME);
type = crm_element_value(xml_obj, PCMK_XA_TYPE);
score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE
"> entry in configuration without id");
continue;
}
new_node = pe_create_node(id, uname, type, score, scheduler);
if (new_node == NULL) {
return FALSE;
}
handle_startup_fencing(scheduler, new_node);
add_node_attrs(xml_obj, new_node, FALSE, scheduler);
crm_trace("Done with node %s",
crm_element_value(xml_obj, PCMK_XA_UNAME));
}
}
if (scheduler->localhost
&& (pe_find_node(scheduler->nodes, scheduler->localhost) == NULL)) {
crm_info("Creating a fake local node");
pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0,
scheduler);
}
return TRUE;
}
static void
setup_container(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler)
{
const char *container_id = NULL;
if (rsc->children) {
g_list_foreach(rsc->children, (GFunc) setup_container, scheduler);
return;
}
container_id = g_hash_table_lookup(rsc->meta, PCMK__META_CONTAINER);
if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) {
pcmk_resource_t *container = pe_find_resource(scheduler->resources,
container_id);
if (container) {
rsc->container = container;
pcmk__set_rsc_flags(container, pcmk_rsc_has_filler);
container->fillers = g_list_append(container->fillers, rsc);
pcmk__rsc_trace(rsc, "Resource %s's container is %s",
rsc->id, container_id);
} else {
pcmk__config_err("Resource %s: Unknown resource container (%s)",
rsc->id, container_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); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
const char *new_node_id = NULL;
/* Check for remote nodes, which are defined by ocf:pacemaker:remote
* primitives.
*/
if (xml_contains_remote_node(xml_obj)) {
new_node_id = ID(xml_obj);
/* The "pe_find_node" check is here to make sure we don't iterate over
* an expanded node that has already been added to the node list. */
if (new_node_id
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found remote node %s defined by resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
continue;
}
/* Check for guest nodes, which are defined by special meta-attributes
* of a primitive of any type (for example, VirtualDomain or Xen).
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) {
/* 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
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s",
new_node_id, ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
continue;
}
/* Check for guest nodes inside a group. Clones are currently not
* supported as guest nodes.
*/
if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) {
xmlNode *xml_obj2 = NULL;
for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL;
xml_obj2 = pcmk__xe_next(xml_obj2)) {
new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources,
scheduler);
if (new_node_id
&& (pe_find_node(scheduler->nodes, new_node_id) == NULL)) {
crm_trace("Found guest node %s in resource %s inside group %s",
new_node_id, ID(xml_obj2), ID(xml_obj));
pe_create_node(new_node_id, new_node_id, "remote", NULL,
scheduler);
}
}
}
}
return TRUE;
}
/* Call this after all the nodes and resources have been
* unpacked, but before the status section is read.
*
* A remote node's online status is reflected by the state
* of the remote node's connection resource. We need to link
* the remote node to this connection resource so we can have
* easy access to the connection resource during the scheduler calculations.
*/
static void
link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc)
{
pcmk_node_t *remote_node = NULL;
if (new_rsc->is_remote_node == FALSE) {
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 = pe_find_node(scheduler->nodes, new_rsc->id);
CRM_CHECK(remote_node != NULL, return);
pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s",
new_rsc->id, pe__node_name(remote_node));
remote_node->details->remote_rsc = new_rsc;
if (new_rsc->container == 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.
*/
g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND),
strdup("container"));
}
}
static void
destroy_tag(gpointer data)
{
pcmk_tag_t *tag = data;
if (tag) {
free(tag->id);
g_list_free_full(tag->refs, free);
free(tag);
}
}
/*!
* \internal
* \brief Parse configuration XML for resource information
*
* \param[in] xml_resources Top of resource configuration XML
* \param[in,out] scheduler Scheduler data
*
* \return TRUE
*
* \note unpack_remote_nodes() MUST be called before this, so that the nodes can
* be used when pe__unpack_resource() calls resource_location()
*/
gboolean
unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
GList *gIter = NULL;
scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag);
for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
pcmk_resource_t *new_rsc = NULL;
const char *id = ID(xml_obj);
if (pcmk__str_empty(id)) {
pcmk__config_err("Ignoring <%s> resource without ID",
xml_obj->name);
continue;
}
if (pcmk__str_eq((const char *) xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE,
pcmk__str_none)) {
if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id,
NULL, NULL) == FALSE) {
/* Record the template's ID for the knowledge of its existence anyway. */
g_hash_table_insert(scheduler->template_rsc_sets, strdup(id),
NULL);
}
continue;
}
crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id);
if (pe__unpack_resource(xml_obj, &new_rsc, NULL,
scheduler) == pcmk_rc_ok) {
scheduler->resources = g_list_append(scheduler->resources, new_rsc);
pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id);
} else {
pcmk__config_err("Ignoring <%s> resource '%s' "
"because configuration is invalid",
xml_obj->name, id);
}
}
for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
setup_container(rsc, scheduler);
link_rsc2remotenode(scheduler, rsc);
}
scheduler->resources = g_list_sort(scheduler->resources,
pe__cmp_rsc_priority);
if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) {
/* Ignore */
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) {
pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined");
pcmk__config_err("Either configure some or disable STONITH with the "
PCMK_OPT_STONITH_ENABLED " option");
pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity");
}
return TRUE;
}
gboolean
unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_tag = NULL;
scheduler->tags = pcmk__strkey_table(free, destroy_tag);
for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL;
xml_tag = pcmk__xe_next(xml_tag)) {
xmlNode *xml_obj_ref = NULL;
const char *tag_id = ID(xml_tag);
if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) {
continue;
}
if (tag_id == NULL) {
pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID,
(const char *) xml_tag->name);
continue;
}
for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL;
xml_obj_ref = pcmk__xe_next(xml_obj_ref)) {
const char *obj_ref = ID(xml_obj_ref);
if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) {
continue;
}
if (obj_ref == NULL) {
pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID,
xml_obj_ref->name, tag_id);
continue;
}
if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) {
return FALSE;
}
}
}
return TRUE;
}
/* The ticket state section:
* "/cib/status/tickets/ticket_state" */
static gboolean
unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler)
{
const char *ticket_id = NULL;
const char *granted = NULL;
const char *last_granted = NULL;
const char *standby = NULL;
xmlAttrPtr xIter = NULL;
pcmk_ticket_t *ticket = NULL;
ticket_id = ID(xml_ticket);
if (pcmk__str_empty(ticket_id)) {
return FALSE;
}
crm_trace("Processing ticket state for %s", ticket_id);
ticket = g_hash_table_lookup(scheduler->tickets, ticket_id);
if (ticket == NULL) {
ticket = ticket_new(ticket_id, scheduler);
if (ticket == NULL) {
return FALSE;
}
}
for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
const char *prop_name = (const char *)xIter->name;
const char *prop_value = pcmk__xml_attr_value(xIter);
if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) {
continue;
}
g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value));
}
granted = g_hash_table_lookup(ticket->state, "granted");
if (granted && crm_is_true(granted)) {
ticket->granted = TRUE;
crm_info("We have ticket '%s'", ticket->id);
} else {
ticket->granted = FALSE;
crm_info("We do not have ticket '%s'", ticket->id);
}
last_granted = g_hash_table_lookup(ticket->state, "last-granted");
if (last_granted) {
long long last_granted_ll;
pcmk__scan_ll(last_granted, &last_granted_ll, 0LL);
ticket->last_granted = (time_t) last_granted_ll;
}
standby = g_hash_table_lookup(ticket->state, "standby");
if (standby && crm_is_true(standby)) {
ticket->standby = TRUE;
if (ticket->granted) {
crm_info("Granted ticket '%s' is in standby-mode", ticket->id);
}
} else {
ticket->standby = FALSE;
}
crm_trace("Done with ticket state for %s", ticket_id);
return TRUE;
}
static gboolean
unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler)
{
xmlNode *xml_obj = NULL;
for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL;
xml_obj = pcmk__xe_next(xml_obj)) {
if (!pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) {
continue;
}
unpack_ticket_state(xml_obj, scheduler);
}
return TRUE;
}
static void
unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state,
pcmk_scheduler_t *scheduler)
{
const char *resource_discovery_enabled = NULL;
const xmlNode *attrs = NULL;
pcmk_resource_t *rsc = NULL;
if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
return;
}
if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) {
return;
}
crm_trace("Processing Pacemaker Remote node %s", pe__node_name(this_node));
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_MAINTENANCE),
&(this_node->details->remote_maintenance), 0);
rsc = this_node->details->remote_rsc;
if (this_node->details->remote_requires_reset == FALSE) {
this_node->details->unclean = FALSE;
this_node->details->unseen = FALSE;
}
attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pe__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) {
crm_info("%s is in standby mode", pe__node_name(this_node));
this_node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) ||
((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed))) {
crm_info("%s is in maintenance mode", pe__node_name(this_node));
this_node->details->maintenance = TRUE;
}
resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY);
if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) {
if (pe__is_remote_node(this_node)
&& !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
pcmk__config_warn("Ignoring " XML_NODE_ATTR_RSC_DISCOVERY
" attribute on Pacemaker Remote node %s"
" because fencing is disabled",
pe__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",
pe__node_name(this_node));
this_node->details->rsc_discovery_enabled = FALSE;
}
}
}
/*!
* \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 = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS,
FALSE);
add_node_attrs(attrs, node, TRUE, scheduler);
if (crm_is_true(pe_node_attribute_raw(node, "standby"))) {
crm_info("%s is in standby mode", pe__node_name(node));
node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) {
crm_info("%s is in maintenance mode", pe__node_name(node));
node->details->maintenance = TRUE;
}
discovery = pe_node_attribute_raw(node, XML_NODE_ATTR_RSC_DISCOVERY);
if ((discovery != NULL) && !crm_is_true(discovery)) {
pcmk__config_warn("Ignoring " XML_NODE_ATTR_RSC_DISCOVERY " attribute "
"for %s because disabling resource discovery is "
"not allowed for cluster nodes", pe__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* node_state element itself and node attributes inside it, but not the
* resource history inside it. Multiple passes through the status are needed to
* fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = crm_element_value(state, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " XML_CIB_TAG_STATE " entry without "
PCMK_XA_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = crm_element_value(state, PCMK_XA_UNAME);
if (uname == NULL) {
/* If a joining peer makes the cluster acquire the quorum from corosync
* meanwhile it has not joined CPG membership of pacemaker-controld yet,
* it's possible that the created node_state entry doesn't have an uname
* yet. We should recognize the node as `pending` and wait for it to
* join CPG.
*/
crm_trace("Handling " XML_CIB_TAG_STATE " entry with id=\"%s\" without "
PCMK_XA_UNAME, id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
pcmk__config_warn("Ignoring recorded node state for id=\"%s\" (%s) "
"because it is no longer in the configuration",
id, pcmk__s(uname, "uname unknown"));
return;
}
if (pe__is_guest_or_remote_node(this_node)) {
/* We can't determine the online status of Pacemaker Remote nodes until
* after all resource history has been unpacked. In this first pass, we
* do need to mark whether the node has been fenced, as this plays a
* role during unpacking cluster node resource state.
*/
pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_FENCED),
&(this_node->details->remote_was_fenced), 0);
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;
this_node->details->unseen = FALSE;
crm_trace("Determining online status of cluster node %s (id %s)",
pe__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 node_state entries in CIB status
for (const xmlNode *state = first_named_child(status, XML_CIB_TAG_STATE);
state != NULL; state = crm_next_same_xml(state)) {
const char *id = ID(state);
const char *uname = crm_element_value(state, PCMK_XA_UNAME);
pcmk_node_t *this_node = NULL;
if ((id == NULL) || (uname == NULL)) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history from malformed "
XML_CIB_TAG_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 (this_node->details->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 (pe__is_guest_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.
*/
pcmk_resource_t *rsc = this_node->details->remote_rsc;
if ((rsc == NULL) || (rsc->role != pcmk_role_started)
|| (rsc->container->role != pcmk_role_started)) {
crm_trace("Not unpacking resource history for guest node %s "
"because container and connection are not known to "
"be up", id);
continue;
}
} else if (pe__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->details->remote_rsc;
if ((rsc == NULL)
|| (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)
&& (rsc->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 (pe__is_guest_or_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);
this_node->details->unpacked = TRUE;
unpack_node_lrm(this_node, state, scheduler);
rc = EAGAIN; // Other node histories might depend on this one
}
return rc;
}
/* remove nodes that are down, stopping */
/* create positive rsc_to_node constraints between resources and the nodes they are running on */
/* anything else? */
gboolean
unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler)
{
xmlNode *state = NULL;
crm_trace("Beginning unpack");
if (scheduler->tickets == NULL) {
scheduler->tickets = pcmk__strkey_table(free, destroy_ticket);
}
for (state = pcmk__xe_first_child(status); state != NULL;
state = pcmk__xe_next(state)) {
if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) {
unpack_tickets_state((xmlNode *) state, scheduler);
} else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
unpack_node_state(state, scheduler);
}
}
while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) {
crm_trace("Another pass through node resource histories is needed");
}
// Now catch any nodes we didn't see
unpack_node_history(status,
pcmk_is_set(scheduler->flags,
pcmk_sched_fencing_enabled),
scheduler);
/* Now that we know where resources are, we can schedule stops of containers
* with failed bundle connections
*/
if (scheduler->stop_needed != NULL) {
for (GList *item = scheduler->stop_needed; item; item = item->next) {
pcmk_resource_t *container = item->data;
pcmk_node_t *node = pe__current_node(container);
if (node) {
stop_action(container, node, FALSE);
}
}
g_list_free(scheduler->stop_needed);
scheduler->stop_needed = NULL;
}
/* Now that we know status of all Pacemaker Remote connections and nodes,
* we can stop connections for node shutdowns, and check the online status
* of remote/guest nodes that didn't have any node history to unpack.
*/
for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *this_node = gIter->data;
if (!pe__is_guest_or_remote_node(this_node)) {
continue;
}
if (this_node->details->shutdown
&& (this_node->details->remote_rsc != NULL)) {
pe__set_next_role(this_node->details->remote_rsc, pcmk_role_stopped,
"remote shutdown");
}
if (!this_node->details->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 node_state entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM);
int member = 0;
if (member_time == NULL) {
return -1LL;
} else if (crm_str_to_boolean(member_time, &member) == 1) {
/* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was
* recorded as a boolean for a DC < 2.1.7, or the node is pending
* shutdown and has left the CPG, in which case it was set to 1 to avoid
* fencing for PCMK_OPT_NODE_PENDING_TIMEOUT.
*
* We return the effective time for in_ccm=1 because what's important to
* avoid fencing is that effective time minus this value is less than
* the pending node timeout.
*/
return member? (long long) get_effective_time(scheduler) : 0LL;
} else {
long long when_member = 0LL;
if ((pcmk__scan_ll(member_time, &when_member,
0LL) != pcmk_rc_ok) || (when_member < 0LL)) {
crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM
" in " XML_CIB_TAG_STATE " entry", member_time);
return -1LL;
}
return when_member;
}
}
/*!
* \internal
* \brief Unpack node's time when it became online in process group
*
* \param[in] node_state Node's node_state entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = crm_element_value(node_state, PCMK__XA_CRMD);
// @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline"
if (pcmk__str_eq(peer_time, OFFLINESTATUS,
pcmk__str_casei|pcmk__str_null_matches)) {
return 0LL;
} else if (pcmk__str_eq(peer_time, ONLINESTATUS, 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 "
XML_CIB_TAG_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 node_state entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
long long value = 0LL;
int value_i = 0;
const char *value_s = pe_node_attribute_raw(node, PCMK_NODE_ATTR_TERMINATE);
// Value may be boolean or an epoch time
if (crm_str_to_boolean(value_s, &value_i) == 1) {
return (value_i != 0);
}
if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) {
return (value > 0);
}
crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE
"node attribute for %s", value_s, pe__node_name(node));
return false;
}
static gboolean
determine_online_status_no_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
gboolean online = FALSE;
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
if (when_member <= 0) {
crm_trace("Node %s is %sdown", pe__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",
pe__node_name(this_node), join);
}
} else if (this_node->details->expected_up == FALSE) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pe__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",
pe__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->node_pending_timeout > 0)
&& (when_member > 0) && (when_online <= 0)) {
// There is a timeout on pending nodes, and node is pending
time_t timeout = when_member + scheduler->node_pending_timeout;
if (get_effective_time(node->details->data_set) >= timeout) {
return true; // Node has timed out
}
// Node is pending, but still has time
pe__update_recheck_time(timeout, scheduler, "pending node timeout");
}
return false;
}
static bool
determine_online_status_fencing(pcmk_scheduler_t *scheduler,
const xmlNode *node_state,
pcmk_node_t *this_node)
{
bool termination_requested = unpack_node_terminate(this_node, node_state);
const char *join = crm_element_value(node_state, PCMK__XA_JOIN);
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
long long when_member = unpack_node_member(node_state, scheduler);
long long when_online = unpack_node_online(node_state);
/*
- PCMK__XA_JOIN ::= member|down|pending|banned
- PCMK__XA_EXPECTED ::= member|down
@COMPAT with entries recorded for DCs < 2.1.7
- PCMK__XA_IN_CCM ::= true|false
- PCMK__XA_CRMD ::= online|offline
Since crm_feature_set 3.18.0 (pacemaker-2.1.7):
- PCMK__XA_IN_CCM ::= <timestamp>|0
Since when node has been a cluster member. A value 0 of means the node is not
a cluster member.
- PCMK__XA_CRMD ::= <timestamp>|0
Since when peer has been online in CPG. A value 0 means the peer is offline
in CPG.
*/
crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s",
pe__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", pe__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", pe__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",
pe__node_name(this_node));
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pe__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", pe__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", pe__node_name(this_node));
this_node->details->standby = TRUE;
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->details->remote_rsc;
pcmk_resource_t *container = NULL;
pcmk_node_t *host = NULL;
/* If there is a node state entry for a (former) Pacemaker Remote node
* but no resource creating that node, the node's connection resource will
* be NULL. Consider it an offline remote node in that case.
*/
if (rsc == NULL) {
this_node->details->online = FALSE;
goto remote_online_done;
}
container = rsc->container;
if (container && pcmk__list_of_1(rsc->running_on)) {
host = rsc->running_on->data;
}
/* If the resource is currently started, mark it online. */
if (rsc->role == pcmk_role_started) {
crm_trace("%s node %s presumed ONLINE because connection resource is started",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = TRUE;
}
/* consider this node shutting down if transitioning start->stop */
if ((rsc->role == pcmk_role_started)
&& (rsc->next_role == pcmk_role_stopped)) {
crm_trace("%s node %s shutting down because connection resource is stopping",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->shutdown = TRUE;
}
/* Now check all the failure conditions. */
if(container && pcmk_is_set(container->flags, pcmk_rsc_failed)) {
crm_trace("Guest node %s UNCLEAN because guest resource failed",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
crm_trace("%s node %s OFFLINE because connection resource failed",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
} else if ((rsc->role == pcmk_role_stopped)
|| ((container != NULL)
&& (container->role == pcmk_role_stopped))) {
crm_trace("%s node %s OFFLINE because its resource is stopped",
(container? "Guest" : "Remote"), this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = FALSE;
} else if (host && (host->details->online == FALSE)
&& host->details->unclean) {
crm_trace("Guest node %s UNCLEAN because host is unclean",
this_node->details->id);
this_node->details->online = FALSE;
this_node->details->remote_requires_reset = TRUE;
}
remote_online_done:
crm_trace("Remote node %s online=%s",
this_node->details->id, this_node->details->online ? "TRUE" : "FALSE");
}
static void
determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node,
pcmk_scheduler_t *scheduler)
{
gboolean online = FALSE;
const char *exp_state = crm_element_value(node_state, PCMK__XA_EXPECTED);
CRM_CHECK(this_node != NULL, return);
this_node->details->shutdown = FALSE;
this_node->details->expected_up = 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)) {
this_node->details->expected_up = TRUE;
}
if (this_node->details->type == node_ping) {
this_node->details->unclean = FALSE;
online = FALSE; /* As far as resource management is concerned,
* the node is safely offline.
* Anyone caught abusing this logic will be shot
*/
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
online = determine_online_status_no_fencing(scheduler, node_state,
this_node);
} else {
online = determine_online_status_fencing(scheduler, node_state,
this_node);
}
if (online) {
this_node->details->online = TRUE;
} else {
/* remove node from contention */
this_node->fixed = TRUE; // @COMPAT deprecated and unused
this_node->weight = -INFINITY;
}
if (online && this_node->details->shutdown) {
/* don't run resources here */
this_node->fixed = TRUE; // @COMPAT deprecated and unused
this_node->weight = -INFINITY;
}
if (this_node->details->type == node_ping) {
crm_info("%s is not a Pacemaker node", pe__node_name(this_node));
} else if (this_node->details->unclean) {
pcmk__sched_warn("%s is unclean", pe__node_name(this_node));
} else if (this_node->details->online) {
crm_info("%s is %s", pe__node_name(this_node),
this_node->details->shutdown ? "shutting down" :
this_node->details->pending ? "pending" :
this_node->details->standby ? "standby" :
this_node->details->maintenance ? "maintenance" : "online");
} else {
crm_trace("%s is offline", pe__node_name(this_node));
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
CRM_ASSERT(end);
basename = strndup(last_rsc_id, end - last_rsc_id + 1);
CRM_ASSERT(basename);
return basename;
}
/*!
* \internal
* \brief Get the name of the first instance of a cloned resource
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name plus :0
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_zero(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
size_t base_name_len = end - last_rsc_id + 1;
char *zero = NULL;
CRM_ASSERT(end);
zero = calloc(base_name_len + 3, sizeof(char));
CRM_ASSERT(zero);
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 = create_xml_node(NULL, XML_CIB_TAG_RESOURCE);
copy_in_properties(xml_rsc, rsc_entry);
crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id);
crm_log_xml_debug(xml_rsc, "Orphan resource");
if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) {
return NULL;
}
if (xml_contains_remote_node(xml_rsc)) {
pcmk_node_t *node;
crm_debug("Detected orphaned remote node %s", rsc_id);
node = pe_find_node(scheduler->nodes, rsc_id);
if (node == NULL) {
node = pe_create_node(rsc_id, rsc_id, "remote", NULL, scheduler);
}
link_rsc2remotenode(scheduler, rsc);
if (node) {
crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id);
node->details->shutdown = TRUE;
}
}
if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) {
/* This orphaned rsc needs to be mapped to a container. */
crm_trace("Detected orphaned container filler %s", rsc_id);
pcmk__set_rsc_flags(rsc, pcmk_rsc_removed_filler);
}
pcmk__set_rsc_flags(rsc, pcmk_rsc_removed);
scheduler->resources = g_list_append(scheduler->resources, rsc);
return rsc;
}
/*!
* \internal
* \brief Create orphan instance for anonymous clone resource history
*
* \param[in,out] parent Clone resource that orphan will be added to
* \param[in] rsc_id Orphan's resource ID
* \param[in] node Where orphan is active (for logging only)
* \param[in,out] scheduler Scheduler data
*
* \return Newly added orphaned instance of \p parent
*/
static pcmk_resource_t *
create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *top = pe__create_clone_child(parent, scheduler);
// find_rsc() because we might be a cloned group
pcmk_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL,
pcmk_rsc_match_clone_only);
pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s",
top->id, parent->id, rsc_id, pe__node_name(node));
return orphan;
}
/*!
* \internal
* \brief Check a node for an instance of an anonymous clone
*
* Return a child instance of the specified anonymous clone, in order of
* preference: (1) the instance running on the specified node, if any;
* (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX
* instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX
* instances are already active).
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node on which to check for instance
* \param[in,out] parent Clone to check
* \param[in] rsc_id Name of cloned resource in history (no instance)
*/
static pcmk_resource_t *
find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
pcmk_resource_t *parent, const char *rsc_id)
{
GList *rIter = NULL;
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *inactive_instance = NULL;
gboolean skip_inactive = FALSE;
CRM_ASSERT(parent != NULL);
CRM_ASSERT(pe_rsc_is_clone(parent));
CRM_ASSERT(!pcmk_is_set(parent->flags, pcmk_rsc_unique));
// Check for active (or partially active, for cloned groups) instance
pcmk__rsc_trace(parent, "Looking for %s on %s in %s",
rsc_id, pe__node_name(node), parent->id);
for (rIter = parent->children; rsc == NULL && rIter; 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->fns->location(child, &locations, 2);
if (locations) {
/* We should never associate the same numbered anonymous clone
* instance with multiple nodes, and clone instances can't migrate,
* so there must be only one location, regardless of history.
*/
CRM_LOG_ASSERT(locations->next == NULL);
if (((pcmk_node_t *) locations->data)->details == node->details) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry is orphaned, rsc will be NULL.
*/
rsc = parent->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
if (rsc) {
/* If there are multiple instance history entries for an
* anonymous clone in a single node's history (which can
* happen if PCMK_META_GLOBALLY_UNIQUE is switched from true
* to false), we want to consider the instances beyond the
* first as orphans, even if there are inactive instance
* numbers available.
*/
if (rsc->running_on) {
crm_notice("Active (now-)anonymous clone %s has "
"multiple (orphan) instance histories on %s",
parent->id, pe__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->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 && inactive_instance->pending_node
&& (inactive_instance->pending_node->details != node->details)) {
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 "requires" set to "quorum" or "nothing", and we don't
* have a clone instance for every node, we don't want to consume a valid
* instance number for unclean nodes. Such instances may appear to be active
* according to the history, but should be considered inactive, so we can
* start an instance elsewhere. Treat such instances as orphans.
*
* An exception is instances running on guest nodes -- since guest node
* "fencing" is actually just a resource stop, requires shouldn't apply.
*
* @TODO Ideally, we'd use an inactive instance number if it is not needed
* for any clean instances. However, we don't know that at this point.
*/
if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)
&& (!node->details->online || node->details->unclean)
&& !pe__is_guest_node(node)
&& !pe__is_universal_clone(parent, scheduler)) {
rsc = NULL;
}
if (rsc == NULL) {
rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler);
pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id);
}
return rsc;
}
static pcmk_resource_t *
unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
const char *rsc_id)
{
pcmk_resource_t *rsc = NULL;
pcmk_resource_t *parent = NULL;
crm_trace("looking for %s", rsc_id);
rsc = pe_find_resource(scheduler->resources, rsc_id);
if (rsc == NULL) {
/* If we didn't find the resource by its name in the operation history,
* check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0,
* we create a single :0 orphan to match against here.
*/
char *clone0_id = clone_zero(rsc_id);
pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources,
clone0_id);
if (clone0 && !pcmk_is_set(clone0->flags, pcmk_rsc_unique)) {
rsc = clone0;
parent = uber_parent(clone0);
crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id);
} else {
crm_trace("%s is not known as %s either (orphan)",
rsc_id, clone0_id);
}
free(clone0_id);
} else if (rsc->variant > pcmk_rsc_variant_primitive) {
crm_trace("Resource history for %s is orphaned because it is no longer primitive",
rsc_id);
return NULL;
} else {
parent = uber_parent(rsc);
}
if (pe_rsc_is_anon_clone(parent)) {
if (pe_rsc_is_bundled(parent)) {
rsc = pe__find_bundle_replica(parent->parent, node);
} else {
char *base = clone_strip(rsc_id);
rsc = find_anonymous_clone(scheduler, node, parent, base);
free(base);
CRM_ASSERT(rsc != NULL);
}
}
if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei)
&& !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) {
pcmk__str_update(&rsc->clone_name, rsc_id);
pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s",
rsc_id, pe__node_name(node), rsc->id,
pcmk_is_set(rsc->flags, pcmk_rsc_removed)? " (ORPHAN)" : "");
}
return rsc;
}
static pcmk_resource_t *
process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node,
pcmk_scheduler_t *scheduler)
{
pcmk_resource_t *rsc = NULL;
const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
crm_debug("Detected orphan resource %s on %s", rsc_id, pe__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id);
resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__",
scheduler);
}
return rsc;
}
static void
process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node,
enum action_fail_response on_fail)
{
pcmk_node_t *tmpnode = NULL;
char *reason = NULL;
enum action_fail_response save_on_fail = pcmk_on_fail_ignore;
CRM_ASSERT(rsc);
pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s",
rsc->id, role2text(rsc->role), pe__node_name(node),
fail2text(on_fail));
/* process current state */
if (rsc->role != pcmk_role_unknown) {
pcmk_resource_t *iter = rsc;
while (iter) {
if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) {
pcmk_node_t *n = pe__copy_node(node);
pcmk__rsc_trace(rsc, "%s%s%s known on %s",
rsc->id,
((rsc->clone_name == NULL)? "" : " also known as "),
((rsc->clone_name == NULL)? "" : rsc->clone_name),
pe__node_name(n));
g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n);
}
if (pcmk_is_set(iter->flags, pcmk_rsc_unique)) {
break;
}
iter = iter->parent;
}
}
/* If a managed resource is believed to be running, but node is down ... */
if ((rsc->role > pcmk_role_stopped)
&& node->details->online == FALSE
&& node->details->maintenance == FALSE
&& 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 (pe__is_guest_node(node)) {
pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed);
should_fence = TRUE;
} else if (pcmk_is_set(rsc->cluster->flags,
pcmk_sched_fencing_enabled)) {
if (pe__is_remote_node(node) && node->details->remote_rsc
&& !pcmk_is_set(node->details->remote_rsc->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.
*/
node->details->unseen = TRUE;
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(rsc->cluster, 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(rsc->cluster, node, reason, FALSE);
free(reason);
break;
case pcmk_on_fail_standby_node:
node->details->standby = TRUE;
node->details->standby_onfail = TRUE;
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, -INFINITY, "__action_migration_auto__",
rsc->cluster);
break;
case pcmk_on_fail_stop:
pe__set_next_role(rsc, pcmk_role_stopped, "on-fail=stop");
break;
case pcmk_on_fail_restart:
if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
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->container && pe_rsc_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.
*/
rsc->cluster->stop_needed =
g_list_prepend(rsc->cluster->stop_needed, rsc->container);
} else if (rsc->container) {
stop_action(rsc->container, node, FALSE);
} else if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
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(rsc->cluster->flags, pcmk_sched_fencing_enabled)) {
tmpnode = NULL;
if (rsc->is_remote_node) {
tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id);
}
if (tmpnode &&
pe__is_remote_node(tmpnode) &&
tmpnode->details->remote_was_fenced == 0) {
/* The remote connection resource failed in a way that
* should result in fencing the remote node.
*/
pe_fence_node(rsc->cluster, tmpnode,
"remote connection is unrecoverable", FALSE);
}
}
/* require the stop action regardless if fencing is occurring or not. */
if (rsc->role > pcmk_role_stopped) {
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->remote_reconnect_ms) {
pe__set_next_role(rsc, pcmk_role_stopped, "remote reset");
}
break;
}
/* ensure a remote-node connection failure forces an unclean remote-node
* to be fenced. By setting unseen = FALSE, the remote-node failure will
* result in a fencing operation regardless if we're going to attempt to
* reconnect to the remote-node in this transition or not. */
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed) && rsc->is_remote_node) {
tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id);
if (tmpnode && tmpnode->details->unclean) {
tmpnode->details->unseen = FALSE;
}
}
if ((rsc->role != pcmk_role_stopped)
&& (rsc->role != pcmk_role_unknown)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__config_warn("Detected active orphan %s running on %s",
rsc->id, pe__node_name(node));
} else {
pcmk__config_warn("Resource '%s' must be stopped manually on "
"%s because cluster is configured not to "
"stop active orphans",
rsc->id, pe__node_name(node));
}
}
native_add_running(rsc, node, rsc->cluster,
(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->clone_name && strchr(rsc->clone_name, ':') != NULL) {
/* Only do this for older status sections that included instance numbers
* Otherwise stopped instances will appear as orphans
*/
pcmk__rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)",
rsc->clone_name, rsc->id);
free(rsc->clone_name);
rsc->clone_name = 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->role == pcmk_role_stopped)
&& rsc->partial_migration_source
&& rsc->partial_migration_source->details == node->details
&& rsc->partial_migration_target
&& rsc->running_on) {
rsc->role = pcmk_role_started;
}
}
/* create active recurring operations as optional */
static void
process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc,
int start_index, int stop_index,
GList *sorted_op_list, pcmk_scheduler_t *scheduler)
{
int counter = -1;
const char *task = NULL;
const char *status = NULL;
GList *gIter = sorted_op_list;
CRM_ASSERT(rsc);
pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d",
rsc->id, start_index, stop_index);
for (; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
guint interval_ms = 0;
char *key = NULL;
const char *id = ID(rsc_op);
counter++;
if (node->details->online == FALSE) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline",
rsc->id, pe__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, pe__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pe__node_name(node), counter);
continue;
}
crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms);
if (interval_ms == 0) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring",
id, pe__node_name(node));
continue;
}
- status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
+ status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(status, "-1", pcmk__str_casei)) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: status",
id, pe__node_name(node));
continue;
}
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
/* create the action */
key = pcmk__op_key(rsc->id, task, interval_ms);
pcmk__rsc_trace(rsc, "Creating %s on %s", key, pe__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
- status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS);
+ status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS);
if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei)
&& pcmk__str_eq(status, "0", pcmk__str_casei)) {
*stop_index = counter;
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
*start_index = counter;
} else if ((implied_monitor_start <= *stop_index)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK,
&lock_time) == pcmk_ok) && (lock_time != 0)) {
if ((scheduler->shutdown_lock > 0)
&& (get_effective_time(scheduler)
> (lock_time + scheduler->shutdown_lock))) {
pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired",
rsc->id, pe__node_name(node));
pe__clear_resource_history(rsc, node);
} else {
/* @COMPAT I don't like breaking const signatures, but
* rsc->lock_node should really be const -- we just can't change it
* until the next API compatibility break.
*/
rsc->lock_node = (pcmk_node_t *) node;
rsc->lock_time = lock_time;
}
}
}
/*!
* \internal
* \brief Unpack one lrm_resource entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry 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 = 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 action_fail_response on_fail = pcmk_on_fail_ignore;
enum rsc_role_e saved_role = pcmk_role_unknown;
if (rsc_id == NULL) {
pcmk__config_err("Ignoring invalid " XML_LRM_TAG_RESOURCE
" entry: No " PCMK_XA_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " XML_LRM_TAG_RESOURCE " for %s on %s",
rsc_id, pe__node_name(node));
// Build a list of individual lrm_rsc_op entries, so we can sort them
for (rsc_op = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP);
rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) {
op_list = g_list_prepend(op_list, rsc_op);
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
}
}
/* find the resource */
rsc = unpack_find_resource(scheduler, node, rsc_id);
if (rsc == NULL) {
if (op_list == NULL) {
// If there are no operations, there is nothing to do
return NULL;
} else {
rsc = process_orphan_resource(lrm_resource, node, scheduler);
}
}
CRM_ASSERT(rsc != NULL);
// Check whether the resource is "shutdown-locked" to this node
if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) {
unpack_shutdown_lock(lrm_resource, rsc, node, scheduler);
}
/* process operations */
saved_role = rsc->role;
rsc->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->next_role == pcmk_role_unknown)
|| (req_role < rsc->next_role)) {
pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE);
} else if (req_role > rsc->next_role) {
pcmk__rsc_info(rsc,
"%s: Not overwriting calculated next role %s"
" with requested next role %s",
rsc->id, role2text(rsc->next_role),
role2text(req_role));
}
}
if (saved_role > rsc->role) {
rsc->role = saved_role;
}
return rsc;
}
static void
handle_orphaned_container_fillers(const xmlNode *lrm_rsc_list,
pcmk_scheduler_t *scheduler)
{
for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list);
rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) {
pcmk_resource_t *rsc;
pcmk_resource_t *container;
const char *rsc_id;
const char *container_id;
if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) {
continue;
}
container_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER);
rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID);
if (container_id == NULL || rsc_id == NULL) {
continue;
}
container = pe_find_resource(scheduler->resources, container_id);
if (container == NULL) {
continue;
}
rsc = pe_find_resource(scheduler->resources, rsc_id);
if ((rsc == NULL) || (rsc->container != NULL)
|| !pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) {
continue;
}
pcmk__rsc_trace(rsc, "Mapped container of orphaned resource %s to %s",
rsc->id, container_id);
rsc->container = container;
container->fillers = g_list_append(container->fillers, 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_orphaned_container_filler = false;
// Drill down to lrm_resources section
xml = find_xml_node(xml, XML_CIB_TAG_LRM, FALSE);
if (xml == NULL) {
return;
}
xml = find_xml_node(xml, XML_LRM_TAG_RESOURCES, FALSE);
if (xml == NULL) {
return;
}
// Unpack each lrm_resource entry
for (const xmlNode *rsc_entry = first_named_child(xml, XML_LRM_TAG_RESOURCE);
rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) {
pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler);
if ((rsc != NULL)
&& pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) {
found_orphaned_container_filler = true;
}
}
/* Now that all resource state has been unpacked for this node, map any
* orphaned container fillers to their container resource.
*/
if (found_orphaned_container_filler) {
handle_orphaned_container_fillers(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->role = pcmk_role_unpromoted;
} else {
rsc->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->weight = *score;
}
#define XPATH_NODE_STATE "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE
#define SUB_XPATH_LRM_RESOURCE "/" XML_CIB_TAG_LRM \
"/" XML_LRM_TAG_RESOURCES \
"/" XML_LRM_TAG_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" XML_LRM_TAG_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 @" XML_LRM_ATTR_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" XML_LRM_ATTR_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
if (xml && target_rc >= 0) {
int rc = PCMK_OCF_UNKNOWN_ERROR;
int status = PCMK_EXEC_ERROR;
crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc);
- crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status);
+ crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status);
if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) {
return NULL;
}
}
return xml;
}
static xmlNode *
find_lrm_resource(const char *rsc_id, const char *node_name,
pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']",
NULL);
xml = get_xpath_object((const char *) xpath->str, scheduler->input,
LOG_DEBUG);
g_string_free(xpath, TRUE);
return xml;
}
/*!
* \internal
* \brief Check whether a resource has no completed action history on a node
*
* \param[in,out] rsc Resource to check
* \param[in] node_name Node to check
*
* \return true if \p rsc_id is unknown on \p node_name, otherwise false
*/
static bool
unknown_on_node(pcmk_resource_t *rsc, const char *node_name)
{
bool result = false;
xmlXPathObjectPtr search;
GString *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, "']"
SUB_XPATH_LRM_RSC_OP "[@" XML_LRM_ATTR_RC "!='193']",
NULL);
search = xpath_search(rsc->cluster->input, (const char *) xpath->str);
result = (numXpathResults(search) == 0);
freeXpathObject(search);
g_string_free(xpath, TRUE);
return result;
}
/*!
* \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] same_node Whether the operations are on the same node
* \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, bool same_node,
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 && pe__is_newer_op(monitor, xml_op, same_node) > 0);
}
/*!
* \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] same_node Whether the operations are on the same node
* \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, bool same_node,
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 = first_named_child(lrm_resource, XML_LRM_TAG_RSC_OP);
op != NULL; op = crm_next_same_xml(op)) {
const char * task = NULL;
if (op == xml_op) {
continue;
}
task = crm_element_value(op, PCMK_XA_OPERATION);
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
NULL)
&& pe__is_newer_op(op, xml_op, same_node) > 0) {
return true;
}
}
return false;
}
/*!
* \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_to;
const char *source = NULL;
const char *target = NULL;
bool same_node = false;
if (migrate_from) {
xml_op = migrate_from;
}
source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE);
target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET);
/* It's preferred to compare to the migrate event on the same node if
* existing, since call ids are more reliable.
*/
if (pcmk__str_eq(node_name, target, pcmk__str_casei)) {
if (migrate_from) {
xml_op = migrate_from;
same_node = true;
} else {
xml_op = migrate_to;
}
} else if (pcmk__str_eq(node_name, source, pcmk__str_casei)) {
if (migrate_to) {
xml_op = migrate_to;
same_node = true;
} else {
xml_op = migrate_from;
}
}
/* 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, same_node, scheduler)
|| monitor_not_running_after(rsc_id, node_name, xml_op, same_node,
scheduler);
}
/*!
* \internal
* \brief Parse migration source and target node names from history entry
*
* \param[in] entry Resource history entry for a migration action
* \param[in] source_node If not NULL, source must match this node
* \param[in] target_node If not NULL, target must match this node
* \param[out] source_name Where to store migration source node name
* \param[out] target_name Where to store migration target node name
*
* \return Standard Pacemaker return code
*/
static int
get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node,
const pcmk_node_t *target_node,
const char **source_name, const char **target_name)
{
*source_name = crm_element_value(entry, XML_LRM_ATTR_MIGRATE_SOURCE);
*target_name = crm_element_value(entry, XML_LRM_ATTR_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
XML_LRM_ATTR_MIGRATE_SOURCE " and "
XML_LRM_ATTR_MIGRATE_TARGET, ID(entry));
return pcmk_rc_unpack_error;
}
if ((source_node != NULL)
&& !pcmk__str_eq(*source_name, source_node->details->uname,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
XML_LRM_ATTR_MIGRATE_SOURCE "='%s' does not match %s",
ID(entry), *source_name, pe__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->details->uname,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
XML_LRM_ATTR_MIGRATE_TARGET "='%s' does not match %s",
ID(entry), *target_name, pe__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, pe__node_name(node));
rsc->role = pcmk_role_stopped;
rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations,
(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;
// 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,
true, history->rsc->cluster);
// 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, history->rsc->cluster);
if (migrate_from != NULL) {
if (source_newer_op) {
/* There's a newer non-monitor operation on the source and a
* migrate_from on the target, so this migrate_to is irrelevant to
* the resource's state.
*/
return;
}
crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc);
- crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS,
- &from_status);
+ crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status);
}
/* If the resource has newer state on both the source and target after the
* migration events, this migrate_to is irrelevant to the resource's state.
*/
target_newer_state = newer_state_after_migrate(history->rsc->id, target,
history->xml, migrate_from,
history->rsc->cluster);
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->role = pcmk_role_started;
target_node = pe_find_node(history->rsc->cluster->nodes, 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, history->rsc->cluster,
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 = pe_find_node(history->rsc->cluster->nodes,
source);
native_add_running(history->rsc, target_node, history->rsc->cluster,
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->partial_migration_target = target_node;
history->rsc->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;
// 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->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, history->rsc->cluster);
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,
history->rsc->cluster)) {
/* 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 = pe_find_node(history->rsc->cluster->nodes,
target);
if (target_node && target_node->details->online) {
native_add_running(history->rsc, target_node, history->rsc->cluster,
FALSE);
}
} else if (!non_monitor_after(history->rsc->id, source, history->xml, true,
history->rsc->cluster)) {
/* 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->dangling_migrations =
g_list_prepend(history->rsc->dangling_migrations,
(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;
// 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->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,
history->rsc->cluster);
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,
history->rsc->cluster)) {
/* 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 = pe_find_node(history->rsc->cluster->nodes,
source);
if (source_node && source_node->details->online) {
native_add_running(history->rsc, source_node, history->rsc->cluster,
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)
{
if (!(history->node->details->online)) {
return;
}
for (const xmlNode *xIter = history->rsc->cluster->failed->children;
xIter != NULL; xIter = xIter->next) {
const char *key = pe__xe_history_key(xIter);
const char *uname = crm_element_value(xIter, PCMK_XA_UNAME);
if (pcmk__str_eq(history->key, key, pcmk__str_none)
&& pcmk__str_eq(uname, history->node->details->uname,
pcmk__str_casei)) {
crm_trace("Skipping duplicate entry %s on %s",
history->key, pe__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pe__node_name(history->node));
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname);
crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id);
add_node_copy(history->rsc->cluster->failed, history->xml);
}
static char *
last_change_str(const xmlNode *xml_op)
{
time_t when;
char *result = NULL;
if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE,
&when) == pcmk_ok) {
char *when_s = pcmk__epoch2str(&when, 0);
const char *p = strchr(when_s, ' ');
// Skip day of week to make message shorter
if ((p != NULL) && (*(++p) != '\0')) {
result = strdup(p);
CRM_ASSERT(result != NULL);
}
free(when_s);
}
if (result == NULL) {
result = strdup("unknown time");
CRM_ASSERT(result != NULL);
}
return result;
}
/*!
* \internal
* \brief Compare two on-fail values
*
* \param[in] first One on-fail value to compare
* \param[in] second The other on-fail value to compare
*
* \return A negative number if second is more severe than first, zero if they
* are equal, or a positive number if first is more severe than second.
* \note This is only needed until the action_fail_response values can be
* renumbered at the next API compatibility break.
*/
static int
cmp_on_fail(enum action_fail_response first, enum action_fail_response second)
{
switch (first) {
case pcmk_on_fail_demote:
switch (second) {
case pcmk_on_fail_ignore:
return 1;
case pcmk_on_fail_demote:
return 0;
default:
return -1;
}
break;
case pcmk_on_fail_reset_remote:
switch (second) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
return 1;
case pcmk_on_fail_reset_remote:
return 0;
default:
return -1;
}
break;
case pcmk_on_fail_restart_container:
switch (second) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
case pcmk_on_fail_reset_remote:
return 1;
case pcmk_on_fail_restart_container:
return 0;
default:
return -1;
}
break;
default:
break;
}
switch (second) {
case pcmk_on_fail_demote:
return (first == pcmk_on_fail_ignore)? -1 : 1;
case pcmk_on_fail_reset_remote:
switch (first) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
return -1;
default:
return 1;
}
break;
case pcmk_on_fail_restart_container:
switch (first) {
case pcmk_on_fail_ignore:
case pcmk_on_fail_demote:
case pcmk_on_fail_restart:
case pcmk_on_fail_reset_remote:
return -1;
default:
return 1;
}
break;
default:
break;
}
return first - second;
}
/*!
* \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 = -INFINITY;
pcmk_resource_t *fail_rsc = rsc;
if (fail_rsc->parent != NULL) {
pcmk_resource_t *parent = uber_parent(fail_rsc);
if (pe_rsc_is_anon_clone(parent)) {
/* For anonymous clones, if an operation with on-fail=stop fails for
* any instance, the entire clone must stop.
*/
fail_rsc = parent;
}
}
// Ban the resource from all nodes
crm_notice("%s will not be started under current conditions", fail_rsc->id);
if (fail_rsc->allowed_nodes != NULL) {
g_hash_table_destroy(fail_rsc->allowed_nodes);
}
fail_rsc->allowed_nodes = pe__node_list2table(rsc->cluster->nodes);
g_hash_table_foreach(fail_rsc->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 action_fail_response *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 action_fail_response config_on_fail,
enum rsc_role_e fail_role, xmlNode **last_failure,
enum action_fail_response *on_fail)
{
bool is_probe = false;
char *last_change_s = NULL;
*last_failure = history->xml;
is_probe = pcmk_xe_is_probe(history->xml);
last_change_s = last_change_str(history->xml);
if (!pcmk_is_set(history->rsc->cluster->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 " CRM_XS " exit-status=%d id=%s",
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pe__node_name(history->node), last_change_s,
history->exit_status, history->id);
} else {
pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of "
"%s on %s at %s " CRM_XS " exit-status=%d id=%s",
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
(is_probe? "probe" : history->task), history->rsc->id,
pe__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 resource-discovery option for location constraints",
history->rsc->id, pe__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (cmp_on_fail(*on_fail, config_on_fail) < 0) {
pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s",
fail2text(*on_fail), fail2text(config_on_fail),
history->key);
*on_fail = config_on_fail;
}
if (strcmp(history->task, PCMK_ACTION_STOP) == 0) {
resource_location(history->rsc, history->node, -INFINITY,
"__stop_fail__", history->rsc->cluster);
} 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->role = pcmk_role_promoted;
} else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) {
if (config_on_fail == pcmk_on_fail_block) {
history->rsc->role = pcmk_role_promoted;
pe__set_next_role(history->rsc, pcmk_role_stopped,
"demote with on-fail=block");
} else if (history->exit_status == PCMK_OCF_NOT_RUNNING) {
history->rsc->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->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->role = pcmk_role_stopped;
} else if (history->rsc->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, role2text(history->rsc->role),
pcmk__btoa(history->node->details->unclean),
fail2text(config_on_fail), role2text(fail_role));
if ((fail_role != pcmk_role_started)
&& (history->rsc->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->details->data_set, history->node)) {
return; // Failed stops are recoverable via fencing
}
last_change_s = last_change_str(history->xml);
pcmk__sched_err("No further recovery can be attempted for %s "
"because %s on %s failed (%s%s%s) at %s "
CRM_XS " rc=%d id=%s",
history->rsc->id, history->task,
pe__node_name(history->node),
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""),
last_change_s, history->exit_status, history->id);
free(last_change_s);
pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_managed);
pcmk__set_rsc_flags(history->rsc, pcmk_rsc_blocked);
}
/*!
* \internal
* \brief Update action history's execution status and why
*
* \param[in,out] history Parsed action history entry
* \param[out] why Where to store reason for update
* \param[in] value New value
* \param[in] reason Description of why value was changed
*/
static inline void
remap_because(struct action_history *history, const char **why, int value,
const char *reason)
{
if (history->execution_status != value) {
history->execution_status = value;
*why = reason;
}
}
/*!
* \internal
* \brief Remap informational monitor results and operation status
*
* For the monitor results, certain OCF codes are for providing extended information
* to the user about services that aren't yet failed but not entirely healthy either.
* These must be treated as the "normal" result by Pacemaker.
*
* For operation status, the action result can be used to determine an appropriate
* status for the purposes of responding to the action. The status provided by the
* executor is not directly usable since the executor does not know what was expected.
*
* \param[in,out] history Parsed action history entry
* \param[in,out] on_fail What should be done about the result
* \param[in] expired Whether result is expired
*
* \note If the result is remapped and the node is not shutting down or failed,
* the operation will be recorded in the scheduler data's list of failed
* operations to highlight it for the user.
*
* \note This may update the resource's current and next role.
*/
static void
remap_operation(struct action_history *history,
enum action_fail_response *on_fail, bool expired)
{
bool is_probe = false;
int orig_exit_status = history->exit_status;
int orig_exec_status = history->execution_status;
const char *why = NULL;
const char *task = history->task;
// Remap degraded results to their successful counterparts
history->exit_status = pcmk__effective_rc(history->exit_status);
if (history->exit_status != orig_exit_status) {
why = "degraded result";
if (!expired && (!history->node->details->shutdown
|| history->node->details->online)) {
record_failed_op(history);
}
}
if (!pe_rsc_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, pe__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, pe__node_name(history->node),
history->expected_exit_status,
services_ocf_exitcode_str(history->expected_exit_status),
history->exit_status,
services_ocf_exitcode_str(history->exit_status),
(pcmk__str_empty(history->exit_reason)? "" : ": "),
pcmk__s(history->exit_reason, ""));
}
switch (history->exit_status) {
case PCMK_OCF_OK:
if (is_probe
&& (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active on %s at %s",
history->rsc->id, pe__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->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, pe__node_name(history->node),
last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->role = pcmk_role_promoted;
}
break;
case PCMK_OCF_FAILED_PROMOTED:
if (!expired) {
history->rsc->role = pcmk_role_promoted;
}
remap_because(history, &why, PCMK_EXEC_ERROR, "exit status");
break;
case PCMK_OCF_NOT_CONFIGURED:
remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status");
break;
case PCMK_OCF_UNIMPLEMENT_FEATURE:
{
guint interval_ms = 0;
crm_element_value_ms(history->xml, PCMK_META_INTERVAL,
&interval_ms);
if (interval_ms == 0) {
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD,
"exit status");
} else {
remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED,
"exit status");
}
}
break;
case PCMK_OCF_NOT_INSTALLED:
case PCMK_OCF_INVALID_PARAM:
case PCMK_OCF_INSUFFICIENT_PRIV:
if (!expired) {
block_if_unrecoverable(history);
}
remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status");
break;
default:
if (history->execution_status == PCMK_EXEC_DONE) {
char *last_change_s = last_change_str(history->xml);
crm_info("Treating unknown exit status %d from %s of %s "
"on %s at %s as failure",
history->exit_status, task, history->rsc->id,
pe__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure,
rsc->cluster);
} else {
pcmk__op_digest_t *digest_data = NULL;
digest_data = rsc_action_digest_cmp(rsc, xml_op, node,
rsc->cluster);
switch (digest_data->rc) {
case pcmk__digest_unknown:
crm_trace("Resource %s history entry %s on %s"
" has no digest to compare",
rsc->id, pe__xe_history_key(xml_op),
node->details->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 = pe_find_node(scheduler->nodes, 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->remote_reconnect_ms
&& pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& (interval_ms != 0)
&& pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
pcmk_node_t *remote_node = pe_find_node(rsc->cluster->nodes, rsc->id);
if (remote_node && !remote_node->details->remote_was_fenced) {
if (is_last_failure) {
crm_info("Waiting to clear monitor failure for remote node %s"
" until fencing has occurred", rsc->id);
}
return TRUE;
}
}
return FALSE;
}
/*!
* \internal
* \brief Check operation age and schedule failure clearing when appropriate
*
* This function has two distinct purposes. The first is to check whether an
* operation history entry is expired (i.e. the resource has a failure timeout,
* the entry is older than the timeout, and the resource either has no fail
* count or its fail count is entirely older than the timeout). The second is to
* schedule fail count clearing when appropriate (i.e. the operation is expired
* and either the resource has an expired fail count or the operation is a
* last_failure for a remote connection resource with a reconnect interval,
* or the operation is a last_failure for a start or monitor operation and the
* resource's parameters have changed since the operation).
*
* \param[in,out] history Parsed action result history
*
* \return true if operation history entry is expired, otherwise false
*/
static bool
check_operation_expiry(struct action_history *history)
{
bool expired = false;
bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0");
time_t last_run = 0;
int unexpired_fail_count = 0;
const char *clear_reason = NULL;
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, pe__node_name(history->node));
return false; // "Not installed" must always be cleared manually
}
if ((history->rsc->failure_timeout > 0)
&& (crm_element_value_epoch(history->xml, XML_RSC_OP_LAST_CHANGE,
&last_run) == 0)) {
/* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a
* timestamp
*/
time_t now = get_effective_time(history->rsc->cluster);
time_t last_failure = 0;
// Is this particular operation history older than the failure timeout?
if ((now >= (last_run + history->rsc->failure_timeout))
&& !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 timeout=%ds"
" last-failure@%lld",
history->id, (long long) last_run, (expired? "" : "not "),
(long long) now, unexpired_fail_count,
history->rsc->failure_timeout, (long long) last_failure);
last_failure += history->rsc->failure_timeout + 1;
if (unexpired_fail_count && (now < last_failure)) {
pe__update_recheck_time(last_failure, history->rsc->cluster,
"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, pe__node_name(history->node));
expired = false;
}
} else if (is_last_failure
&& (history->rsc->remote_reconnect_ms != 0)) {
/* 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, history->rsc->cluster);
if (pcmk_is_set(history->rsc->cluster->flags,
pcmk_sched_fencing_enabled)
&& (history->rsc->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,
history->rsc->cluster);
}
}
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, pe__node_name(history->node));
expired = false;
break;
}
}
return expired;
}
int
pe__target_rc_from_xml(const xmlNode *xml_op)
{
int target_rc = 0;
const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY);
if (key == NULL) {
return -1;
}
decode_transition_key(key, NULL, NULL, NULL, &target_rc);
return target_rc;
}
/*!
* \internal
* \brief Update a resource's state for an action result
*
* \param[in,out] history Parsed action history entry
* \param[in] exit_status Exit status to base new state on
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
update_resource_state(struct action_history *history, int exit_status,
const xmlNode *last_failure,
enum action_fail_response *on_fail)
{
bool clear_past_failure = false;
if ((exit_status == PCMK_OCF_NOT_INSTALLED)
|| (!pe_rsc_is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml))) {
history->rsc->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, pe__xe_history_key(last_failure),
pcmk__str_none)) {
clear_past_failure = true;
}
if (history->rsc->role < pcmk_role_started) {
set_active(history->rsc);
}
} else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) {
history->rsc->role = pcmk_role_started;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) {
history->rsc->role = pcmk_role_stopped;
clear_past_failure = true;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE,
pcmk__str_none)) {
history->rsc->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 on-fail=demote
clear_past_failure = true;
}
history->rsc->role = pcmk_role_unpromoted;
} else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM,
pcmk__str_none)) {
history->rsc->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->role < pcmk_role_started) {
pcmk__rsc_trace(history->rsc, "%s active on %s",
history->rsc->id, pe__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, fail2text(*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->remote_reconnect_ms == 0) {
/* With no reconnect interval, the connection is allowed to
* start again after the remote node is fenced and
* completely stopped. (With a reconnect interval, we wait
* for the failure to be cleared entirely before attempting
* to reconnect.)
*/
*on_fail = pcmk_on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"clear past failures and reset remote");
}
break;
}
}
/*!
* \internal
* \brief Check whether a given history entry matters for resource state
*
* \param[in] history Parsed action history entry
*
* \return true if action can affect resource state, otherwise false
*/
static inline bool
can_affect_state(struct action_history *history)
{
#if 0
/* @COMPAT It might be better to parse only actions we know we're interested
* in, rather than exclude a couple we don't. However that would be a
* behavioral change that should be done at a major or minor series release.
* Currently, unknown operations can affect whether a resource is considered
* active and/or failed.
*/
return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR,
PCMK_ACTION_START, PCMK_ACTION_STOP,
PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE,
PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM,
"asyncmon", NULL);
#else
return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY,
PCMK_ACTION_META_DATA, NULL);
#endif
}
/*!
* \internal
* \brief Unpack execution/exit status and exit reason from a history entry
*
* \param[in,out] history Action history entry to unpack
*
* \return Standard Pacemaker return code
*/
static int
unpack_action_result(struct action_history *history)
{
- if ((crm_element_value_int(history->xml, XML_LRM_ATTR_OPSTATUS,
+ if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS,
&(history->execution_status)) < 0)
|| (history->execution_status < PCMK_EXEC_PENDING)
|| (history->execution_status > PCMK_EXEC_MAX)
|| (history->execution_status == PCMK_EXEC_CANCELLED)) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
- "with invalid " XML_LRM_ATTR_OPSTATUS " '%s'",
+ "with invalid " PCMK__XA_OP_STATUS " '%s'",
history->id, history->rsc->id,
pe__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
- XML_LRM_ATTR_OPSTATUS),
+ PCMK__XA_OP_STATUS),
""));
return pcmk_rc_unpack_error;
}
if ((crm_element_value_int(history->xml, XML_LRM_ATTR_RC,
&(history->exit_status)) < 0)
|| (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) {
#if 0
/* @COMPAT We should ignore malformed entries, but since that would
* change behavior, it should be done at a major or minor series
* release.
*/
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"with invalid " XML_LRM_ATTR_RC " '%s'",
history->id, history->rsc->id,
pe__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
XML_LRM_ATTR_RC),
""));
return pcmk_rc_unpack_error;
#else
history->exit_status = CRM_EX_ERROR;
#endif
}
history->exit_reason = crm_element_value(history->xml,
XML_LRM_ATTR_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 (!pe_rsc_is_bundled(history->rsc)
&& pcmk_xe_mask_probe_failure(history->xml)
&& (orig_exit_status != history->expected_exit_status)) {
if (history->rsc->role <= pcmk_role_stopped) {
history->rsc->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,
pe__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,
pe__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 an 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, pe__node_name(history->node));
crm_xml_add(history->xml, XML_LRM_ATTR_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 action_fail_response *on_fail)
{
pcmk_resource_t *ban_rsc = history->rsc;
if (!pcmk_is_set(history->rsc->flags, pcmk_rsc_unique)) {
ban_rsc = uber_parent(history->rsc);
}
crm_notice("Treating probe result '%s' for %s on %s as 'not running'",
services_ocf_exitcode_str(orig_exit_status), history->rsc->id,
pe__node_name(history->node));
update_resource_state(history, history->expected_exit_status, last_failure,
on_fail);
crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname);
record_failed_op(history);
resource_location(ban_rsc, history->node, -INFINITY, "masked-probe-failure",
history->rsc->cluster);
}
/*!
* \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
* lrm_resource block, as node and resource are assumed to be the same.
*/
static bool
failure_is_newer(const struct action_history *history,
const xmlNode *last_failure)
{
guint failure_interval_ms = 0U;
long long failure_change = 0LL;
long long this_change = 0LL;
if (last_failure == NULL) {
return false; // Resource has no last_failure entry
}
if (!pcmk__str_eq(history->task,
crm_element_value(last_failure, PCMK_XA_OPERATION),
pcmk__str_none)) {
return false; // last_failure is for different action
}
if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL,
&failure_interval_ms) != pcmk_ok)
|| (history->interval_ms != failure_interval_ms)) {
return false; // last_failure is for action with different interval
}
if ((pcmk__scan_ll(crm_element_value(history->xml, XML_RSC_OP_LAST_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(crm_element_value(last_failure,
XML_RSC_OP_LAST_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->role = pcmk_role_promoted;
} else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0)
&& history->node->details->unclean) {
/* A migrate_to action is pending on a unclean source, so force a stop
* on the target.
*/
const char *migrate_target = NULL;
pcmk_node_t *target = NULL;
migrate_target = crm_element_value(history->xml,
XML_LRM_ATTR_MIGRATE_TARGET);
target = pe_find_node(history->rsc->cluster->nodes, migrate_target);
if (target != NULL) {
stop_action(history->rsc, target, FALSE);
}
}
if (history->rsc->pending_task != 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_task().
*/
#if 0
history->rsc->pending_task = strdup("probe");
history->rsc->pending_node = history->node;
#endif
} else {
history->rsc->pending_task = strdup(history->task);
history->rsc->pending_node = history->node;
}
}
static void
unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op,
xmlNode **last_failure, enum action_fail_response *on_fail)
{
int old_rc = 0;
bool expired = false;
pcmk_resource_t *parent = rsc;
enum rsc_role_e fail_role = pcmk_role_unknown;
enum action_fail_response 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 = ID(xml_op);
if (history.id == NULL) {
pcmk__config_err("Ignoring resource history entry for %s on %s "
"without ID", rsc->id, pe__node_name(node));
return;
}
// Task and interval
history.task = crm_element_value(xml_op, PCMK_XA_OPERATION);
if (history.task == NULL) {
pcmk__config_err("Ignoring resource history entry %s for %s on %s "
"without " PCMK_XA_OPERATION,
history.id, rsc->id, pe__node_name(node));
return;
}
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms));
if (!can_affect_state(&history)) {
pcmk__rsc_trace(rsc,
"Ignoring resource history entry %s for %s on %s "
"with irrelevant action '%s'",
history.id, rsc->id, pe__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 = pe__xe_history_key(xml_op);
crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pe__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, pe__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 (!pe_rsc_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 "
CRM_XS " status=%d rc=%d id=%s",
history.task, rsc->id, pe__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, -INFINITY, "hard-error",
rsc->cluster);
unpack_rsc_op_failure(&history, failure_strategy, fail_role,
last_failure, on_fail);
goto done;
case PCMK_EXEC_NOT_CONNECTED:
if (pe__is_guest_or_remote_node(node)
&& pcmk_is_set(node->details->remote_rsc->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->details->remote_rsc,
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 "
CRM_XS " %s",
history.task, services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), rsc->id, pe__node_name(node),
last_change_s, history.id);
free(last_change_s);
update_resource_state(&history, history.expected_exit_status,
*last_failure, on_fail);
crm_xml_add(xml_op, PCMK_XA_UNAME, node->details->uname);
pcmk__set_rsc_flags(rsc, pcmk_rsc_ignore_failure);
record_failed_op(&history);
if ((failure_strategy == pcmk_on_fail_restart_container)
&& cmp_on_fail(*on_fail, pcmk_on_fail_restart) <= 0) {
*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) " CRM_XS " %s",
parent->id, pe__node_name(node),
services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, node, -INFINITY, "hard-error",
rsc->cluster);
} else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) {
pcmk__sched_err("Preventing %s from restarting anywhere because "
"of fatal failure (%s%s%s) " CRM_XS " %s",
parent->id,
services_ocf_exitcode_str(history.exit_status),
(pcmk__str_empty(history.exit_reason)? "" : ": "),
pcmk__s(history.exit_reason, ""), history.id);
resource_location(parent, NULL, -INFINITY, "fatal-error",
rsc->cluster);
}
}
done:
pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)",
rsc->id, pe__node_name(node), history.id,
role2text(rsc->role), role2text(rsc->next_role));
}
static void
add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite,
pcmk_scheduler_t *scheduler)
{
const char *cluster_name = NULL;
pe_rule_eval_data_t rule_data = {
.node_hash = NULL,
.role = pcmk_role_unknown,
.now = scheduler->now,
.match_data = NULL,
.rsc_data = NULL,
.op_data = NULL
};
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_UNAME), strdup(node->details->uname));
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID),
strdup(node->details->id));
if (pcmk__str_eq(node->details->id, scheduler->dc_uuid, pcmk__str_casei)) {
scheduler->dc_node = node;
node->details->is_dc = TRUE;
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(PCMK_VALUE_TRUE));
} else {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_IS_DC), strdup(PCMK_VALUE_FALSE));
}
cluster_name = g_hash_table_lookup(scheduler->config_hash,
PCMK_OPT_CLUSTER_NAME);
if (cluster_name) {
g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME),
strdup(cluster_name));
}
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data,
node->details->attrs, NULL, overwrite,
scheduler);
pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data,
node->details->utilization, NULL,
FALSE, scheduler);
if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) {
const char *site_name = pe_node_attribute_raw(node, "site-name");
if (site_name) {
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(site_name));
} else if (cluster_name) {
/* Default to cluster-name if unset */
g_hash_table_insert(node->details->attrs,
strdup(CRM_ATTR_SITE_NAME),
strdup(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);
rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) {
if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP,
pcmk__str_none)) {
crm_xml_add(rsc_op, "resource", rsc);
crm_xml_add(rsc_op, PCMK_XA_UNAME, node);
op_list = g_list_prepend(op_list, rsc_op);
}
}
if (op_list == NULL) {
/* if there are no operations, there is nothing to do */
return NULL;
}
sorted_op_list = g_list_sort(op_list, sort_op_by_callid);
/* create active recurring operations as optional */
if (active_filter == FALSE) {
return sorted_op_list;
}
op_list = NULL;
calculate_active_ops(sorted_op_list, &start_index, &stop_index);
for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
xmlNode *rsc_op = (xmlNode *) gIter->data;
counter++;
if (start_index < stop_index) {
crm_trace("Skipping %s: not active", ID(rsc_entry));
break;
} else if (counter < start_index) {
crm_trace("Skipping %s: old", 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 = find_xml_node(scheduler->input, XML_CIB_TAG_STATUS, TRUE);
pcmk_node_t *this_node = NULL;
xmlNode *node_state = NULL;
for (node_state = pcmk__xe_first_child(status); node_state != NULL;
node_state = pcmk__xe_next(node_state)) {
if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) {
const char *uname = crm_element_value(node_state, PCMK_XA_UNAME);
if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) {
continue;
}
this_node = pe_find_node(scheduler->nodes, uname);
if(this_node == NULL) {
CRM_LOG_ASSERT(this_node != NULL);
continue;
} else if (pe__is_guest_or_remote_node(this_node)) {
determine_remote_online_status(scheduler, this_node);
} else {
determine_online_status(node_state, this_node, scheduler);
}
if (this_node->details->online
|| pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
/* offline nodes run no resources...
* unless stonith is enabled in which case we need to
* make sure rsc start events happen after the stonith
*/
xmlNode *lrm_rsc = NULL;
tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE);
tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE);
for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL;
lrm_rsc = pcmk__xe_next(lrm_rsc)) {
if (pcmk__str_eq((const char *)lrm_rsc->name,
XML_LRM_TAG_RESOURCE, pcmk__str_none)) {
const char *rsc_id = crm_element_value(lrm_rsc,
PCMK_XA_ID);
if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) {
continue;
}
intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
output = g_list_concat(output, intermediate);
}
}
}
}
}
return output;
}