static gboolean
update_failcount(xmlNode * event, const char *event_node_uuid, int rc, int target_rc, gboolean do_update)
{
int interval = 0;
char *task = NULL;
char *rsc_id = NULL;
char *attr_name = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
const char *on_uname = get_uname_from_event(event);
const char *origin = crm_element_value(event, XML_ATTR_ORIGIN);
if (rc == 99) {
/* this is an internal code for "we're busy, try again" */
return FALSE;
} else if (rc == target_rc) {
return FALSE;
}
if (safe_str_eq(origin, "build_active_RAs")) {
crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh",
id, rc, on_uname);
return FALSE;
}
CRM_LOG_ASSERT(on_uname != NULL);
if (on_uname == NULL) {
return TRUE;
}
if (failed_stop_offset == NULL) {
failed_stop_offset = strdup(INFINITY_S);
}
if (failed_start_offset == NULL) {
failed_start_offset = strdup(INFINITY_S);
}
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval), crm_err("Couldn't parse: %s", ID(event));
goto bail);
CRM_CHECK(task != NULL, goto bail);
CRM_CHECK(rsc_id != NULL, goto bail);
if (do_update || interval > 0) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
do_update = TRUE;
value = failed_start_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
value = failed_stop_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
value = failed_stop_offset;
} else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_DEMOTE)) {
do_update = TRUE;
}
if (value == NULL || safe_str_neq(value, INFINITY_S)) {
value = XML_NVPAIR_ATTR_VALUE "++";
}
if (do_update) {
char *now = crm_itoa(time(NULL));
gboolean is_remote_node = get_is_remote_from_event(event);
crm_warn("Updating failcount for %s on %s after failed %s:"
" rc=%d (update=%s, time=%s)", rsc_id, on_uname, task, rc, value, now);
attr_name = crm_concat("fail-count", rsc_id, '-');
update_attrd(on_uname, attr_name, value, NULL, is_remote_node);
free(attr_name);
attr_name = crm_concat("last-failure", rsc_id, '-');
update_attrd(on_uname, attr_name, now, NULL, is_remote_node);
free(attr_name);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
/*!
* \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(xmlNode * event, const char *event_node_uuid, int rc,
int target_rc, gboolean do_update, gboolean ignore_failures)
{
int interval = 0;
char *task = NULL;
char *rsc_id = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
const char *on_uname = crm_peer_uname(event_node_uuid);
const char *origin = crm_element_value(event, XML_ATTR_ORIGIN);
/* Nothing needs to be done for success, lrm status refresh,
* or direct nack (internal code for "busy, try again")
*/
if ((rc == CRM_DIRECT_NACK_RC) || (rc == target_rc)) {
return FALSE;
} else if (safe_str_eq(origin, "build_active_RAs")) {
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),
crm_err("Couldn't parse: %s", ID(event)); goto bail);
CRM_CHECK(task != NULL, goto bail);
CRM_CHECK(rsc_id != NULL, goto bail);
/* Decide whether update is necessary and what value to use */
if ((interval > 0) || safe_str_eq(task, CRMD_ACTION_PROMOTE)
|| safe_str_eq(task, CRMD_ACTION_DEMOTE)) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
do_update = TRUE;
if (failed_start_offset == NULL) {
failed_start_offset = strdup(INFINITY_S);
}
value = failed_start_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
if (failed_stop_offset == NULL) {
failed_stop_offset = strdup(INFINITY_S);
}
value = failed_stop_offset;
}
/* Fail count will be either incremented or set to infinity */
if (value == NULL || safe_str_neq(value, INFINITY_S)) {
value = XML_NVPAIR_ATTR_VALUE "++";
}
if (do_update) {
char *now = crm_itoa(time(NULL));
char *attr_name = NULL;
gboolean is_remote_node = FALSE;
if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) {
is_remote_node = TRUE;
}
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) {
attr_name = crm_concat("fail-count", rsc_id, '-');
update_attrd(on_uname, attr_name, value, NULL, is_remote_node);
free(attr_name);
}
/* 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)
*/
attr_name = crm_concat("last-failure", rsc_id, '-');
update_attrd(on_uname, attr_name, now, NULL, is_remote_node);
free(attr_name);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
static char *
convert_non_atomic_uuid(char *old_uuid, resource_t * rsc, gboolean allow_notify,
gboolean free_original)
{
int interval = 0;
char *uuid = NULL;
char *rid = NULL;
char *raw_task = NULL;
int task = no_action;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "Processing %s", old_uuid);
if (old_uuid == NULL) {
return NULL;
} else if (strstr(old_uuid, "notify") != NULL) {
goto done; /* no conversion */
} else if (rsc->variant < pe_group) {
goto done; /* no conversion */
}
CRM_ASSERT(parse_op_key(old_uuid, &rid, &raw_task, &interval));
if (interval > 0) {
goto done; /* no conversion */
}
task = text2task(raw_task);
switch (task) {
case stop_rsc:
case start_rsc:
case action_notify:
case action_promote:
case action_demote:
break;
case stopped_rsc:
case started_rsc:
case action_notified:
case action_promoted:
case action_demoted:
task--;
break;
case monitor_rsc:
case shutdown_crm:
case stonith_node:
task = no_action;
break;
default:
crm_err("Unknown action: %s", raw_task);
task = no_action;
break;
}
if (task != no_action) {
if (is_set(rsc->flags, pe_rsc_notify) && allow_notify) {
uuid = generate_notify_key(rid, "confirmed-post", task2text(task + 1));
} else {
uuid = generate_op_key(rid, task2text(task + 1), 0);
}
pe_rsc_trace(rsc, "Converted %s -> %s", old_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(old_uuid);
}
if (free_original) {
free(old_uuid);
}
free(raw_task);
free(rid);
return uuid;
}
if (failed_stop_offset == NULL) {
failed_stop_offset = strdup(INFINITY_S);
}
if (failed_start_offset == NULL) {
failed_start_offset = strdup(INFINITY_S);
}
if (on_uname == NULL) {
/* uname not in event, check cache */
on_uname = get_uname(event_node_uuid);
CRM_CHECK(on_uname != NULL, return TRUE);
}
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval), crm_err("Couldn't parse: %s", ID(event));
goto bail);
CRM_CHECK(task != NULL, goto bail);
CRM_CHECK(rsc_id != NULL, goto bail);
if (do_update || interval > 0) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
do_update = TRUE;
value = failed_start_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
value = failed_stop_offset;
