34 #define set_config_flag(data_set, option, flag) do { \ 35 const char *tmp = pe_pref(data_set->config_hash, option); \ 37 if(crm_is_true(tmp)) { \ 38 set_bit(data_set->flags, flag); \ 40 clear_bit(data_set->flags, flag); \ 53 is_dangling_container_remote_node(
node_t *node)
88 "(otherwise would because %s): " 89 "its guest resource %s is unmanaged",
92 crm_warn(
"Guest node %s will be fenced " 93 "(by recovering its guest resource %s): %s",
105 }
else if (is_dangling_container_remote_node(node)) {
106 crm_info(
"Cleaning up dangling connection for guest node %s: " 107 "fencing was already done because %s, " 108 "and guest resource no longer exists",
117 "(otherwise would because %s): connection is unmanaged",
123 pe_can_fence(data_set, node)?
"will be fenced" :
"is unclean",
130 crm_trace(
"Cluster node %s %s because %s",
132 pe_can_fence(data_set, node)?
"would also be fenced" :
"also is unclean",
138 pe_can_fence(data_set, node)?
"will be fenced" :
"is unclean",
148 #define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \ 149 "[(@" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_PROVIDES "'" \ 150 "or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \ 151 "and @" XML_NVPAIR_ATTR_VALUE "='unfencing']" 154 #define XPATH_ENABLE_UNFENCING \ 155 "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \ 156 "//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \ 157 "|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \ 158 "/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR 161 void set_if_xpath(
unsigned long long flag,
const char *xpath,
164 xmlXPathObjectPtr result = NULL;
166 if (is_not_set(data_set->
flags, flag)) {
168 if (result && (numXpathResults(result) > 0)) {
178 const char *value = NULL;
179 GHashTable *config_hash = crm_str_table_new();
190 crm_info(
"Startup probes: disabled (dangerous)");
195 crm_notice(
"Watchdog will be used via SBD if fencing is required");
209 crm_debug(
"STONITH of failed nodes is %s",
220 crm_debug(
"Stop all active resources: %s",
225 crm_debug(
"Cluster is symmetric" " - resources can run anywhere by default");
231 "Support for 'default-resource-stickiness' cluster property" 232 " is deprecated and will be removed in a future release" 233 " (use resource-stickiness in rsc_defaults instead)");
255 crm_notice(
"Resetting no-quorum-policy to 'stop': cluster has never had quorum");
259 crm_config_err(
"Resetting no-quorum-policy to 'stop': stonith is not configured");
269 crm_debug(
"On loss of CCM Quorum: Freeze resources");
272 crm_debug(
"On loss of CCM Quorum: Stop ALL resources");
275 crm_notice(
"On loss of CCM Quorum: Fence all remaining nodes");
287 crm_trace(
"Orphan resource actions are %s",
291 crm_trace(
"Stopped resources are removed from the status section: %s",
303 "Support for 'is-managed-default' cluster property" 304 " is deprecated and will be removed in a future release" 305 " (use is-managed in rsc_defaults instead)");
307 crm_trace(
"By default resources are %smanaged",
312 is_set(data_set->
flags,
319 crm_trace(
"Unseen nodes will be fenced");
328 crm_debug(
"Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s",
340 destroy_digest_cache(gpointer ptr)
363 " - this is rarely intended", uname);
366 new_node = calloc(1,
sizeof(
node_t));
367 if (new_node == NULL) {
372 new_node->
fixed = FALSE;
375 if (new_node->
details == NULL) {
380 crm_trace(
"Creating node for entry %s/%s", uname,
id);
392 }
else if (type == NULL ||
safe_str_eq(type,
"member")
411 destroy_digest_cache);
424 if (data->name_check == NULL) {
425 data->name_check = g_hash_table_new(
crm_str_hash, g_str_equal);
426 for (xml_rsc = __xml_first_child(parent); xml_rsc != NULL; xml_rsc = __xml_next_element(xml_rsc)) {
427 const char *
id =
ID(xml_rsc);
430 g_hash_table_insert(data->name_check, (
char *)
id, (
char *)
id);
433 if (g_hash_table_lookup(data->name_check, remote_name)) {
438 crm_err(
"Invalid remote-node name, a resource called '%s' already exists.", remote_name);
449 xmlNode *attr_set = NULL;
450 xmlNode *attr = NULL;
452 const char *container_id =
ID(xml_obj);
453 const char *remote_name = NULL;
454 const char *remote_server = NULL;
455 const char *remote_port = NULL;
456 const char *connect_timeout =
"60s";
457 const char *remote_allow_migrate=NULL;
458 const char *container_managed = NULL;
460 for (attr_set = __xml_first_child(xml_obj); attr_set != NULL; attr_set = __xml_next_element(attr_set)) {
465 for (attr = __xml_first_child(attr_set); attr != NULL; attr = __xml_next_element(attr)) {
472 remote_server = value;
475 }
else if (
safe_str_eq(name,
"remote-connect-timeout")) {
476 connect_timeout = value;
477 }
else if (
safe_str_eq(name,
"remote-allow-migrate")) {
478 remote_allow_migrate=value;
480 container_managed = value;
485 if (remote_name == NULL) {
494 remote_allow_migrate, container_managed,
"30s",
"30s",
495 connect_timeout, remote_server, remote_port);
527 xmlNode *xml_obj = NULL;
529 const char *
id = NULL;
530 const char *
uname = NULL;
531 const char *
type = NULL;
532 const char *score = NULL;
534 for (xml_obj = __xml_first_child(xml_nodes); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) {
542 crm_trace(
"Processing node %s/%s", uname,
id);
550 if (new_node == NULL) {
560 handle_startup_fencing(data_set, new_node);
571 crm_info(
"Creating a fake local node");
582 const char *container_id = NULL;
587 for (; gIter != NULL; gIter = gIter->next) {
590 setup_container(child_rsc, data_set);
603 pe_rsc_trace(rsc,
"Resource %s's container is %s", rsc->
id, container_id);
605 pe_err(
"Resource %s: Unknown resource container (%s)", rsc->
id, container_id);
613 xmlNode *xml_obj = NULL;
616 for (xml_obj = __xml_first_child(xml_resources); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) {
617 const char *new_node_id = NULL;
622 new_node_id =
ID(xml_obj);
626 crm_trace(
"Found baremetal remote node %s in container resource %s", new_node_id,
ID(xml_obj));
643 new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, data_set);
646 crm_trace(
"Found guest remote node %s in container resource %s", new_node_id,
ID(xml_obj));
653 xmlNode *xml_obj2 = NULL;
655 for (xml_obj2 = __xml_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = __xml_next_element(xml_obj2)) {
657 new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, data_set);
660 crm_trace(
"Found guest remote node %s in container resource %s which is in group %s", new_node_id,
ID(xml_obj2),
ID(xml_obj));
682 node_t *remote_node = NULL;
703 handle_startup_fencing(data_set, remote_node);
708 strdup(
"container"));
713 destroy_tag(gpointer data)
719 g_list_free_full(tag->
refs, free);
739 xmlNode *xml_obj = NULL;
746 for (xml_obj = __xml_first_child(xml_resources); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) {
750 const char *template_id =
ID(xml_obj);
753 template_id, NULL, NULL) == FALSE) {
760 crm_trace(
"Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj),
ID(xml_obj));
768 if (new_rsc != NULL && new_rsc->
fns != NULL) {
774 for (gIter = data_set->
resources; gIter != NULL; gIter = gIter->next) {
777 setup_container(rsc, data_set);
778 link_rsc2remotenode(data_set, rsc);
788 crm_config_err(
"Resource start-up disabled since no STONITH resources have been defined");
789 crm_config_err(
"Either configure some or disable STONITH with the stonith-enabled option");
790 crm_config_err(
"NOTE: Clusters with shared data need STONITH to ensure data integrity");
799 xmlNode *xml_tag = NULL;
804 for (xml_tag = __xml_first_child(xml_tags); xml_tag != NULL; xml_tag = __xml_next_element(xml_tag)) {
805 xmlNode *xml_obj_ref = NULL;
806 const char *tag_id =
ID(xml_tag);
812 if (tag_id == NULL) {
818 for (xml_obj_ref = __xml_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = __xml_next_element(xml_obj_ref)) {
819 const char *obj_ref =
ID(xml_obj_ref);
825 if (obj_ref == NULL) {
826 crm_config_err(
"Failed unpacking %s for tag %s: %s should be specified",
827 crm_element_name(xml_obj_ref), tag_id,
XML_ATTR_ID);
845 const char *ticket_id = NULL;
846 const char *granted = NULL;
847 const char *last_granted = NULL;
848 const char *standby = NULL;
849 xmlAttrPtr xIter = NULL;
853 ticket_id =
ID(xml_ticket);
854 if (ticket_id == NULL || strlen(ticket_id) == 0) {
858 crm_trace(
"Processing ticket state for %s", ticket_id);
860 ticket = g_hash_table_lookup(data_set->
tickets, ticket_id);
861 if (ticket == NULL) {
863 if (ticket == NULL) {
868 for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) {
869 const char *prop_name = (
const char *)xIter->name;
875 g_hash_table_replace(ticket->
state, strdup(prop_name), strdup(prop_value));
878 granted = g_hash_table_lookup(ticket->
state,
"granted");
884 crm_info(
"We do not have ticket '%s'", ticket->
id);
887 last_granted = g_hash_table_lookup(ticket->
state,
"last-granted");
892 standby = g_hash_table_lookup(ticket->
state,
"standby");
896 crm_info(
"Granted ticket '%s' is in standby-mode", ticket->
id);
902 crm_trace(
"Done with ticket state for %s", ticket_id);
910 xmlNode *xml_obj = NULL;
912 for (xml_obj = __xml_first_child(xml_tickets); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) {
916 unpack_ticket_state(xml_obj, data_set);
925 get_ticket_state_legacy(gpointer key, gpointer value, gpointer user_data)
927 const char *long_key = key;
928 char *state_key = NULL;
930 const char *granted_prefix =
"granted-ticket-";
931 const char *last_granted_prefix =
"last-granted-";
932 static int granted_prefix_strlen = 0;
933 static int last_granted_prefix_strlen = 0;
935 const char *ticket_id = NULL;
936 const char *is_granted = NULL;
937 const char *last_granted = NULL;
938 const char *sep = NULL;
943 if (granted_prefix_strlen == 0) {
944 granted_prefix_strlen = strlen(granted_prefix);
947 if (last_granted_prefix_strlen == 0) {
948 last_granted_prefix_strlen = strlen(last_granted_prefix);
951 if (strstr(long_key, granted_prefix) == long_key) {
952 ticket_id = long_key + granted_prefix_strlen;
953 if (strlen(ticket_id)) {
954 state_key = strdup(
"granted");
957 }
else if (strstr(long_key, last_granted_prefix) == long_key) {
958 ticket_id = long_key + last_granted_prefix_strlen;
959 if (strlen(ticket_id)) {
960 state_key = strdup(
"last-granted");
961 last_granted = value;
963 }
else if ((sep = strrchr(long_key,
'-'))) {
965 state_key =
strndup(long_key, strlen(long_key) - strlen(sep));
968 if (ticket_id == NULL || strlen(ticket_id) == 0) {
973 if (state_key == NULL || strlen(state_key) == 0) {
978 ticket = g_hash_table_lookup(data_set->
tickets, ticket_id);
979 if (ticket == NULL) {
981 if (ticket == NULL) {
987 g_hash_table_replace(ticket->
state, state_key, strdup(value));
995 crm_info(
"We do not have ticket '%s'", ticket->
id);
998 }
else if (last_granted) {
1006 const char *resource_discovery_enabled = NULL;
1007 xmlNode *attrs = NULL;
1009 const char *shutdown = NULL;
1015 if ((this_node == NULL) || (
is_remote_node(this_node) == FALSE)) {
1052 if (resource_discovery_enabled && !
crm_is_true(resource_discovery_enabled)) {
1054 crm_warn(
"ignoring %s attribute on baremetal remote node %s, disabling resource discovery requires stonith to be enabled.",
1068 unpack_node_loop(xmlNode * status,
bool fence,
pe_working_set_t * data_set)
1070 bool changed =
false;
1071 xmlNode *lrm_rsc = NULL;
1073 for (xmlNode *state = __xml_first_child(status); state != NULL; state = __xml_next_element(state)) {
1074 const char *
id = NULL;
1075 const char *
uname = NULL;
1076 node_t *this_node = NULL;
1077 bool process = FALSE;
1087 if (this_node == NULL) {
1088 crm_info(
"Node %s is unknown",
id);
1092 crm_info(
"Node %s is already processed",
id);
1106 }
else if(rsc == NULL) {
1126 determine_remote_online_status(data_set, this_node);
1127 unpack_handle_remote_attrs(this_node, state, data_set);
1139 crm_trace(
"Processing lrm resource entries on %shealthy%s node: %s",
1159 const char *
id = NULL;
1160 const char *
uname = NULL;
1162 xmlNode *state = NULL;
1163 node_t *this_node = NULL;
1167 if (data_set->
tickets == NULL) {
1172 for (state = __xml_first_child(status); state != NULL; state = __xml_next_element(state)) {
1174 xmlNode *xml_tickets = state;
1175 GHashTable *state_hash = NULL;
1179 state_hash = crm_str_table_new();
1182 state_hash, NULL, TRUE, data_set->
now);
1184 g_hash_table_foreach(state_hash, get_ticket_state_legacy, data_set);
1187 g_hash_table_destroy(state_hash);
1191 unpack_tickets_state(xml_tickets, data_set);
1195 xmlNode *attrs = NULL;
1196 const char *resource_discovery_enabled = NULL;
1202 if (uname == NULL) {
1206 }
else if (this_node == NULL) {
1220 crm_trace(
"Processing node id=%s, uname=%s",
id, uname);
1241 if (resource_discovery_enabled && !
crm_is_true(resource_discovery_enabled)) {
1242 crm_warn(
"ignoring %s attribute on node %s, disabling resource discovery is not allowed on cluster nodes",
1255 pe_fence_node(data_set, this_node,
"cluster does not have quorum");
1261 while(unpack_node_loop(status, FALSE, data_set)) {
1268 for (
GListPtr gIter = data_set->
nodes; gIter != NULL; gIter = gIter->next) {
1269 node_t *this_node = gIter->data;
1271 if (this_node == NULL) {
1278 determine_remote_online_status(data_set, this_node);
1285 determine_online_status_no_fencing(
pe_working_set_t * data_set, xmlNode * node_state,
1288 gboolean online = FALSE;
1301 crm_debug(
"Node is not ready to run resources: %s", join);
1306 crm_trace(
"\tis_peer=%s, join=%s, expected=%s",
1311 pe_fence_node(data_set, this_node,
"peer is unexpectedly down");
1312 crm_info(
"\tin_cluster=%s, is_peer=%s, join=%s, expected=%s",
1319 determine_online_status_fencing(
pe_working_set_t * data_set, xmlNode * node_state,
1322 gboolean online = FALSE;
1323 gboolean do_terminate = FALSE;
1338 do_terminate = TRUE;
1340 }
else if (terminate != NULL && strlen(terminate) > 0) {
1342 char t = terminate[0];
1344 if (t !=
'0' && isdigit(t)) {
1345 do_terminate = TRUE;
1349 crm_trace(
"%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d",
1357 if (exp_state == NULL) {
1367 }
else if (in_cluster == NULL) {
1368 pe_fence_node(data_set, this_node,
"peer has not been seen by the cluster");
1371 pe_fence_node(data_set, this_node,
"peer failed the pacemaker membership criteria");
1390 pe_fence_node(data_set, this_node,
"peer is no longer part of the cluster");
1393 pe_fence_node(data_set, this_node,
"peer process is no longer available");
1396 }
else if (do_terminate) {
1397 pe_fence_node(data_set, this_node,
"termination was requested");
1409 pe_fence_node(data_set, this_node,
"peer was in an unknown state");
1410 crm_warn(
"%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d",
1431 goto remote_online_done;
1436 if (container && (g_list_length(rsc->
running_on) == 1)) {
1442 crm_trace(
"%s node %s presumed ONLINE because connection resource is started",
1443 (container?
"Guest" :
"Remote"), this_node->
details->
id);
1449 crm_trace(
"%s node %s shutting down because connection resource is stopping",
1450 (container?
"Guest" :
"Remote"), this_node->
details->
id);
1456 crm_trace(
"Guest node %s UNCLEAN because guest resource failed",
1462 crm_trace(
"%s node %s OFFLINE because connection resource failed",
1463 (container?
"Guest" :
"Remote"), this_node->
details->
id);
1469 crm_trace(
"%s node %s OFFLINE because its resource is stopped",
1470 (container?
"Guest" :
"Remote"), this_node->
details->
id);
1476 crm_trace(
"Guest node %s UNCLEAN because host is unclean",
1491 gboolean online = FALSE;
1492 const char *shutdown = NULL;
1495 if (this_node == NULL) {
1519 online = determine_online_status_no_fencing(data_set, node_state, this_node);
1522 online = determine_online_status_fencing(data_set, node_state, this_node);
1530 this_node->
fixed = TRUE;
1536 this_node->
fixed = TRUE;
1566 CRM_CHECK(last_rsc_id != NULL,
return NULL);
1567 lpc = strlen(last_rsc_id);
1569 switch (last_rsc_id[lpc]) {
1571 crm_err(
"Empty string: %s", last_rsc_id);
1586 zero = calloc(1, lpc + 1);
1587 memcpy(zero, last_rsc_id, lpc);
1595 zero = strdup(last_rsc_id);
1605 CRM_CHECK(last_rsc_id != NULL,
return NULL);
1606 if (last_rsc_id != NULL) {
1607 lpc = strlen(last_rsc_id);
1611 switch (last_rsc_id[lpc]) {
1627 zero = calloc(1, lpc + 3);
1628 memcpy(zero, last_rsc_id, lpc);
1630 zero[lpc + 1] =
'0';
1638 lpc = strlen(last_rsc_id);
1639 zero = calloc(1, lpc + 3);
1640 memcpy(zero, last_rsc_id, lpc);
1642 zero[lpc + 1] =
'0';
1644 crm_trace(
"%s -> %s", last_rsc_id, zero);
1649 create_fake_resource(
const char *rsc_id, xmlNode * rsc_entry,
pe_working_set_t * data_set)
1665 crm_debug(
"Detected orphaned remote node %s", rsc_id);
1670 link_rsc2remotenode(data_set, rsc);
1673 crm_trace(
"Setting node %s as shutting down due to orphaned connection resource", rsc_id);
1680 crm_trace(
"Detected orphaned container filler %s", rsc_id);
1696 gboolean skip_inactive = FALSE;
1704 for (rIter = parent->
children; rsc == NULL && rIter; rIter = rIter->next) {
1710 if (locations == NULL) {
1715 for (nIter = locations; nIter && rsc == NULL; nIter = nIter->next) {
1716 node_t *childnode = nIter->data;
1729 if (rsc && rsc->running_on) {
1730 crm_notice(
"/Anonymous/ clone %s is already running on %s",
1732 skip_inactive = TRUE;
1737 g_list_free(locations);
1741 if (skip_inactive == FALSE) {
1742 pe_rsc_trace(parent,
"Looking for %s anywhere", rsc_id);
1743 for (rIter = parent->
children; rsc == NULL && rIter; rIter = rIter->next) {
1753 if (locations == NULL) {
1758 g_list_free(locations);
1770 pe_rsc_debug(parent,
"Created orphan %s for %s: %s on %s", top->
id, parent->
id, rsc_id,
1775 pe_rsc_debug(rsc,
"Internally renamed %s on %s to %s%s",
1785 xmlNode * rsc_entry)
1802 crm_trace(
"%s is not known as %s either", rsc_id, tmp);
1808 crm_trace(
"%s not found: %s", rsc_id, parent ? parent->
id :
"orphan");
1811 crm_trace(
"%s is no longer a primitive resource, the lrm_resource entry is obsolete",
1819 if(parent && parent->
parent) {
1822 }
else if (pe_rsc_is_clone(parent)) {
1826 rsc = find_anonymous_clone(data_set, node, parent, base);
1847 rsc = create_fake_resource(rsc_id, rsc_entry, data_set);
1867 char *reason = NULL;
1870 pe_rsc_trace(rsc,
"Resource %s is %s on %s: on_fail=%s",
1898 gboolean should_fence = FALSE;
1909 should_fence = TRUE;
1922 " revoked if remote connection can " 1923 "be re-established elsewhere)",
1926 should_fence = TRUE;
1930 if (reason == NULL) {
2014 "remote connection is unrecoverable");
2049 " %s must be stopped manually on %s",
2072 for (; gIter != NULL; gIter = gIter->next) {
2078 g_list_free(possible_matches);
2086 int start_index,
int stop_index,
2090 const char *task = NULL;
2091 const char *status = NULL;
2095 pe_rsc_trace(rsc,
"%s: Start index %d, stop index = %d", rsc->
id, start_index, stop_index);
2097 for (; gIter != NULL; gIter = gIter->next) {
2098 xmlNode *rsc_op = (xmlNode *) gIter->data;
2102 const char *
id =
ID(rsc_op);
2103 const char *interval_s = NULL;
2112 }
else if (start_index < stop_index && counter <= stop_index) {
2116 }
else if (counter < start_index) {
2123 if (interval == 0) {
2145 int implied_monitor_start = -1;
2146 int implied_master_start = -1;
2147 const char *task = NULL;
2148 const char *status = NULL;
2154 for (; gIter != NULL; gIter = gIter->next) {
2155 xmlNode *rsc_op = (xmlNode *) gIter->data;
2173 implied_monitor_start =
counter;
2176 implied_master_start =
counter;
2180 if (*start_index == -1) {
2181 if (implied_master_start != -1) {
2182 *start_index = implied_master_start;
2183 }
else if (implied_monitor_start != -1) {
2184 *start_index = implied_monitor_start;
2193 int stop_index = -1;
2194 int start_index = -1;
2197 const char *task = NULL;
2204 xmlNode *migrate_op = NULL;
2205 xmlNode *rsc_op = NULL;
2206 xmlNode *last_failure = NULL;
2212 crm_element_name(rsc_entry), rsc_id, node->
details->
uname);
2216 sorted_op_list = NULL;
2218 for (rsc_op = __xml_first_child(rsc_entry); rsc_op != NULL; rsc_op = __xml_next_element(rsc_op)) {
2220 op_list = g_list_prepend(op_list, rsc_op);
2224 if (op_list == NULL) {
2230 rsc = unpack_find_resource(data_set, node, rsc_id, rsc_entry);
2232 rsc = process_orphan_resource(rsc_entry, node, data_set);
2237 saved_role = rsc->
role;
2242 for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
2243 xmlNode *rsc_op = (xmlNode *) gIter->data;
2247 migrate_op = rsc_op;
2250 unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set);
2255 process_recurring(node, rsc, start_index, stop_index, sorted_op_list, data_set);
2258 g_list_free(sorted_op_list);
2260 process_rsc_state(rsc, node, on_fail, migrate_op, data_set);
2264 pe_rsc_debug(rsc,
"%s: Overwriting calculated next role %s" 2265 " with requested next role %s",
2270 pe_rsc_info(rsc,
"%s: Not overwriting calculated next role %s" 2271 " with requested next role %s",
2276 if (saved_role > rsc->
role) {
2277 rsc->
role = saved_role;
2284 handle_orphaned_container_fillers(xmlNode * lrm_rsc_list,
pe_working_set_t * data_set)
2286 xmlNode *rsc_entry = NULL;
2287 for (rsc_entry = __xml_first_child(lrm_rsc_list); rsc_entry != NULL;
2288 rsc_entry = __xml_next_element(rsc_entry)) {
2293 const char *container_id;
2301 if (container_id == NULL || rsc_id == NULL) {
2306 if (container == NULL) {
2317 pe_rsc_trace(rsc,
"Mapped orphaned rsc %s's container to %s", rsc->
id, container_id);
2326 xmlNode *rsc_entry = NULL;
2327 gboolean found_orphaned_container_filler = FALSE;
2333 for (rsc_entry = __xml_first_child(lrm_rsc_list); rsc_entry != NULL;
2334 rsc_entry = __xml_next_element(rsc_entry)) {
2337 resource_t *rsc = unpack_lrm_rsc_state(node, rsc_entry, data_set);
2342 found_orphaned_container_filler = TRUE;
2350 if (found_orphaned_container_filler) {
2351 handle_orphaned_container_fillers(lrm_rsc_list, data_set);
2369 set_node_score(gpointer key, gpointer value, gpointer user_data)
2372 int *score = user_data;
2377 #define STATUS_PATH_MAX 1024 2379 find_lrm_op(
const char *resource,
const char *op,
const char *node,
const char *source,
2385 offset += snprintf(xpath + offset,
STATUS_PATH_MAX - offset,
"//node_state[@uname='%s']", node);
2439 if (stop_op == NULL || stop_id < task_id) {
2440 int from_rc = 0, from_status = 0;
2441 const char *migrate_source =
2443 const char *migrate_target =
2448 xmlNode *migrate_from =
2456 pe_rsc_trace(rsc,
"%s op on %s exited with status=%d, rc=%d",
2457 ID(migrate_from), migrate_target, from_status, from_rc);
2462 pe_rsc_trace(rsc,
"Detected dangling migration op: %s on %s",
ID(xml_op),
2472 }
else if (migrate_from) {
2474 pe_rsc_trace(rsc,
"Marking active on %s %p %d", migrate_target, target,
2481 pe_rsc_trace(rsc,
"Marking active on %s %p %d", migrate_target, target,
2517 xmlNode *migrate_op =
2531 if (stop_op == NULL || stop_id < migrate_id) {
2547 xmlNode *migrate_op =
2561 if (stop_op == NULL || stop_id < migrate_id) {
2564 pe_rsc_trace(rsc,
"Stop: %p %d, Migrated: %p %d", stop_op, stop_id, migrate_op,
2570 }
else if (migrate_op == NULL) {
2580 xmlNode *xIter = NULL;
2587 for (xIter = data_set->
failed->children; xIter; xIter = xIter->next) {
2603 static const char *get_op_key(xmlNode *xml_op)
2613 unpack_rsc_op_failure(
resource_t * rsc,
node_t * node,
int rc, xmlNode * xml_op, xmlNode ** last_failure,
2617 bool is_probe = FALSE;
2620 const char *key = get_op_key(xml_op);
2626 *last_failure = xml_op;
2635 crm_warn(
"Processing failed op %s for %s on %s: %s (%d)",
2636 task, rsc->
id, node->
details->
uname, services_ocf_exitcode_str(rc),
2639 record_failed_op(xml_op, node, rsc, data_set);
2642 crm_trace(
"Processing failed op %s for %s on %s: %s (%d)",
2643 task, rsc->
id, node->
details->
uname, services_ocf_exitcode_str(rc),
2647 action =
custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set);
2661 unpack_rsc_migration_failure(rsc, node, xml_op, data_set);
2680 crm_warn(
"Forcing %s to stop after a failed demote action", rsc->
id);
2700 pe_rsc_trace(rsc,
"Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s",
2717 if (pe_rsc_is_clone(parent)
2725 crm_warn(
"Making sure %s doesn't come up again", fail_rsc->
id);
2729 g_hash_table_foreach(fail_rsc->
allowed_nodes, set_node_score, &score);
2736 determine_op_status(
2742 const char *key = get_op_key(xml_op);
2745 bool is_probe = FALSE;
2753 if (target_rc >= 0 && target_rc != rc) {
2755 pe_rsc_debug(rsc,
"%s on %s returned '%s' (%d) instead of the expected value: '%s' (%d)",
2757 services_ocf_exitcode_str(rc), rc,
2758 services_ocf_exitcode_str(target_rc), target_rc);
2766 if (is_probe && target_rc == 7) {
2768 pe_rsc_info(rsc,
"Operation %s found resource %s active on %s",
2797 pe_rsc_info(rsc,
"Operation %s found resource %s active in master mode on %s",
2800 }
else if (target_rc == rc) {
2803 }
else if (target_rc >= 0) {
2811 crm_err(
"%s reported %s in master mode on %s",
2839 pe_proc_err(
"No further recovery can be attempted for %s: %s action failed with '%s' (%d)",
2840 rsc->
id, task, services_ocf_exitcode_str(rc), rc);
2849 crm_info(
"Treating %s (rc=%d) on %s as an ERROR",
2860 bool expired = FALSE;
2861 time_t last_failure = 0;
2864 const char *key = get_op_key(xml_op);
2866 const char *clear_reason = NULL;
2887 if (strstr(
ID(xml_op),
"last_failure")) {
2888 crm_info(
"Waiting to clear monitor failure for remote node %s until fencing has occurred", rsc->
id);
2892 failure_timeout = 0;
2897 if (failure_timeout > 0) {
2903 if (now > (last_run + failure_timeout)) {
2910 if (failure_timeout > 0) {
2912 xml_op, data_set)) {
2915 xml_op, data_set) == 0) {
2916 clear_reason =
"it expired";
2923 clear_reason =
"reconnect interval is set";
2927 }
else if (strstr(
ID(xml_op),
"last_failure") &&
2928 ((strcmp(task,
"start") == 0) || (strcmp(task,
"monitor") == 0))) {
2935 crm_trace(
"rsc op %s/%s on node %s does not have a op digest to compare against", rsc->
id,
2942 clear_reason =
"resource parameters have changed";
2946 if (clear_reason != NULL) {
2949 node, FALSE, TRUE, data_set);
2978 char *dummy_string = NULL;
3003 update_resource_state(
resource_t * rsc,
node_t * node, xmlNode * xml_op,
const char * task,
int rc,
3006 gboolean clear_past_failure = FALSE;
3012 clear_past_failure = TRUE;
3019 const char *op_key = get_op_key(xml_op);
3020 const char *last_failure_key = get_op_key(last_failure);
3023 clear_past_failure = TRUE;
3033 clear_past_failure = TRUE;
3037 clear_past_failure = TRUE;
3041 clear_past_failure = TRUE;
3049 clear_past_failure = TRUE;
3052 unpack_rsc_migration(rsc, node, xml_op, data_set);
3060 if (clear_past_failure) {
3066 pe_rsc_trace(rsc,
"%s.%s is not cleared by a completed stop",
3098 const char *key = NULL;
3099 const char *task = NULL;
3100 const char *task_key = NULL;
3107 gboolean expired = FALSE;
3113 CRM_CHECK(xml_op != NULL,
return FALSE);
3115 task_key = get_op_key(xml_op);
3139 pe_rsc_trace(rsc,
"Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)",
3143 pe_rsc_trace(rsc,
"Node %s (where %s is running) is unclean." 3144 " Further action depends on the value of the stop's on-fail attribute",
3154 expired = check_operation_expiry(rsc, node, rc, xml_op, data_set);
3164 record_failed_op(xml_op, node, rsc, data_set);
3173 record_failed_op(xml_op, node, rsc, data_set);
3177 if (expired && target_rc != rc) {
3180 pe_rsc_debug(rsc,
"Expired operation '%s' on %s returned '%s' (%d) instead of the expected value: '%s' (%d)",
3182 services_ocf_exitcode_str(rc), rc,
3183 services_ocf_exitcode_str(target_rc), target_rc);
3186 crm_notice(
"Ignoring expired calculated failure %s (rc=%d, magic=%s) on %s",
3191 crm_notice(
"Re-initiated expired calculated failure %s (rc=%d, magic=%s) on %s",
3200 status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set);
3207 pe_err(
"Don't know what to do for cancelled ops yet");
3245 update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set);
3249 failure_strategy = get_action_on_fail(rsc, task_key, task, data_set);
3251 crm_warn(
"Cannot ignore failed %s (status=%d, rc=%d) on %s: " 3252 "Resource agent doesn't exist",
3258 unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set);
3267 failure_strategy = get_action_on_fail(rsc, task_key, task, data_set);
3272 crm_warn(
"Pretending the failure of %s (rc=%d) on %s succeeded",
3275 update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set);
3279 record_failed_op(xml_op, node, rsc, data_set);
3282 *on_fail = failure_strategy;
3286 unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set);
3290 "Preventing %s from re-starting on %s: operation %s failed '%s' (%d)",
3292 task, services_ocf_exitcode_str(rc), rc);
3297 crm_err(
"Preventing %s from re-starting anywhere: operation %s failed '%s' (%d)",
3298 parent->
id, task, services_ocf_exitcode_str(rc), rc);
3314 const char *cluster_name = NULL;
3331 cluster_name = g_hash_table_lookup(data_set->
config_hash,
"cluster-name");
3334 strdup(cluster_name));
3348 }
else if (cluster_name) {
3352 strdup(cluster_name));
3359 extract_operations(
const char *node,
const char *rsc, xmlNode * rsc_entry, gboolean active_filter)
3362 int stop_index = -1;
3363 int start_index = -1;
3365 xmlNode *rsc_op = NULL;
3373 sorted_op_list = NULL;
3375 for (rsc_op = __xml_first_child(rsc_entry); rsc_op != NULL; rsc_op = __xml_next_element(rsc_op)) {
3379 op_list = g_list_prepend(op_list, rsc_op);
3383 if (op_list == NULL) {
3391 if (active_filter == FALSE) {
3392 return sorted_op_list;
3399 for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) {
3400 xmlNode *rsc_op = (xmlNode *) gIter->data;
3404 if (start_index < stop_index) {
3405 crm_trace(
"Skipping %s: not active",
ID(rsc_entry));
3408 }
else if (counter < start_index) {
3412 op_list = g_list_append(op_list, rsc_op);
3415 g_list_free(sorted_op_list);
3426 xmlNode *tmp = NULL;
3429 node_t *this_node = NULL;
3431 xmlNode *node_state = NULL;
3433 for (node_state = __xml_first_child(status); node_state != NULL;
3434 node_state = __xml_next_element(node_state)) {
3444 if(this_node == NULL) {
3449 determine_remote_online_status(data_set, this_node);
3460 xmlNode *lrm_rsc = NULL;
3465 for (lrm_rsc = __xml_first_child(tmp); lrm_rsc != NULL;
3466 lrm_rsc = __xml_next_element(lrm_rsc)) {
3475 intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter);
3476 output = g_list_concat(output, intermediate);
gboolean unpack_config(xmlNode *config, pe_working_set_t *data_set)
#define CRM_CHECK(expr, failure_action)
#define XML_RSC_OP_LAST_CHANGE
gboolean unpack_rsc_op(resource_t *rsc, node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed, pe_working_set_t *data_set)
void verify_pe_options(GHashTable *options)
xmlNode * find_xml_node(xmlNode *cib, const char *node_path, gboolean must_find)
#define crm_notice(fmt, args...)
#define CRMD_ACTION_MIGRATED
#define pe_rsc_debug(rsc, fmt, args...)
#define pe_flag_have_stonith_resource
node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t *data_set)
gboolean safe_str_neq(const char *a, const char *b)
gint sort_rsc_priority(gconstpointer a, gconstpointer b)
gboolean determine_online_status(xmlNode *node_state, node_t *this_node, pe_working_set_t *data_set)
gboolean get_target_role(resource_t *rsc, enum rsc_role_e *role)
#define XML_NODE_IS_FENCED
#define XML_ATTR_TRANSITION_MAGIC
node_t * node_copy(const node_t *this_node)
#define stop_action(rsc, node, optional)
#define pe_flag_enable_unfencing
#define pe_rsc_orphan_container_filler
int default_resource_stickiness
char * clone_strip(const char *last_rsc_id)
#define XML_ATTR_QUORUM_PANIC
bool pe_can_fence(pe_working_set_t *data_set, node_t *node)
#define XML_TAG_UTILIZATION
#define pe_flag_have_remote_nodes
void(* free)(resource_t *)
bool container_fix_remote_addr(resource_t *rsc)
#define XML_RULE_ATTR_SCORE
#define XML_BOOLEAN_FALSE
#define crm_config_err(fmt...)
int get_target_rc(xmlNode *xml_op)
action_t * pe_fence_op(node_t *node, const char *op, bool optional, const char *reason, pe_working_set_t *data_set)
enum action_fail_response on_fail
int char2score(const char *score)
#define pe_proc_warn(fmt...)
#define XML_TAG_TRANSIENT_NODEATTRS
#define CRMD_ACTION_NOTIFY
#define pe_flag_startup_probes
long long crm_get_msec(const char *input)
GListPtr find_actions(GListPtr input, const char *key, const node_t *on_node)
gboolean common_unpack(xmlNode *xml_obj, resource_t **rsc, resource_t *parent, pe_working_set_t *data_set)
enum pe_obj_types variant
#define XML_LRM_ATTR_INTERVAL
#define XML_LRM_TAG_RESOURCE
#define pe_flag_stop_rsc_orphans
node_t * partial_migration_source
#define CRMD_ACTION_PROMOTE
int crm_parse_int(const char *text, const char *default_text)
#define XML_NVPAIR_ATTR_NAME
#define XML_NODE_IS_MAINTENANCE
gint sort_op_by_callid(gconstpointer a, gconstpointer b)
#define XML_NODE_EXPECTED
node_t * pe_find_node(GListPtr node_list, const char *uname)
#define XML_CIB_TAG_RSC_TEMPLATE
resource_t * create_child_clone(resource_t *rsc, int sub_id, pe_working_set_t *data_set)
time_t get_effective_time(pe_working_set_t *data_set)
no_quorum_policy_t no_quorum_policy
#define CRM_LOG_ASSERT(expr)
const char * pe_pref(GHashTable *options, const char *name)
resource_t * uber_parent(resource_t *rsc)
#define clear_bit(word, bit)
void copy_in_properties(xmlNode *target, xmlNode *src)
#define CRMD_JOINSTATE_NACK
#define CRM_ATTR_CLUSTER_NAME
#define pe_rsc_allow_migrate
#define XPATH_ENABLE_UNFENCING
xmlNode * get_xpath_object(const char *xpath, xmlNode *xml_obj, int error_level)
#define pe_proc_err(fmt...)
char * strndup(const char *str, size_t len)
#define XML_CIB_TAG_PROPSET
gboolean decode_transition_key(const char *key, char **uuid, int *action, int *transition_id, int *target_rc)
#define XML_LRM_ATTR_RSCID
gboolean unpack_resources(xmlNode *xml_resources, pe_working_set_t *data_set)
#define CRMD_ACTION_START
#define XML_LRM_ATTR_TASK_KEY
#define XML_TAG_ATTR_SETS
#define XML_LRM_ATTR_TASK
const char * role2text(enum rsc_role_e role)
gboolean is_remote_node(node_t *node)
#define CRM_OP_CLEAR_FAILCOUNT
struct node_shared_s * details
gboolean unpack_status(xmlNode *status, pe_working_set_t *data_set)
#define CRMD_JOINSTATE_DOWN
#define crm_warn(fmt, args...)
resource_t * find_container_child(const char *stem, resource_t *rsc, node_t *node)
#define CRMD_ACTION_DEMOTE
#define set_bit(word, bit)
#define crm_atoi(text, default_text)
xmlNode * pe_create_remote_xml(xmlNode *parent, const char *uname, const char *container_id, const char *migrateable, const char *is_managed, const char *interval, const char *monitor_timeout, const char *start_timeout, const char *server, const char *port)
#define crm_debug(fmt, args...)
void native_add_running(resource_t *rsc, node_t *node, pe_working_set_t *data_set)
#define XML_CIB_ATTR_SHUTDOWN
#define XML_RSC_ATTR_CONTAINER
#define XML_CIB_TAG_RESOURCE
gboolean unpack_nodes(xmlNode *xml_nodes, pe_working_set_t *data_set)
#define XML_CIB_TAG_STATE
node_t * partial_migration_target
#define pe_flag_startup_fencing
resource_object_functions_t * fns
GHashTable * allowed_nodes
GHashTable * digest_cache
#define set_config_flag(data_set, option, flag)
#define crm_trace(fmt, args...)
#define CRMD_JOINSTATE_MEMBER
#define do_crm_log(level, fmt, args...)
Log a message.
enum rsc_digest_cmp_val rc
gboolean is_baremetal_remote_node(node_t *node)
#define pe_rsc_is_container
char * digest_secure_calc
gboolean unpack_remote_nodes(xmlNode *xml_resources, pe_working_set_t *data_set)
gboolean add_node_attrs(xmlNode *xml_obj, node_t *node, gboolean overwrite, pe_working_set_t *data_set)
gboolean is_container_remote_node(node_t *node)
xmlNode * add_node_copy(xmlNode *new_parent, xmlNode *xml_node)
const char * stonith_action
#define crm_log_xml_debug(xml, text)
#define XML_TAG_META_SETS
Wrappers for and extensions to libxml2.
#define XML_ATTR_TE_NOWAIT
gboolean add_tag_ref(GHashTable *tags, const char *tag_name, const char *obj_ref)
#define XML_RSC_ATTR_MANAGED
xmlNode * create_xml_node(xmlNode *parent, const char *name)
int crm_element_value_int(xmlNode *data, const char *name, int *dest)
char * clone_zero(const char *last_rsc_id)
const char * crm_element_value(xmlNode *data, const char *name)
action_t * custom_action(resource_t *rsc, char *key, const char *task, node_t *on_node, gboolean optional, gboolean foo, pe_working_set_t *data_set)
const char * pe_node_attribute_raw(pe_node_t *node, const char *name)
#define pe_flag_maintenance_mode
node_t *(* location)(resource_t *, GListPtr *, gboolean)
#define XML_LRM_ATTR_MIGRATE_TARGET
#define CIB_OPTIONS_FIRST
#define XML_RSC_ATTR_REMOTE_NODE
#define XML_LRM_ATTR_RESTART_DIGEST
GListPtr dangling_migrations
void free_xml(xmlNode *child)
#define pe_flag_stop_everything
gboolean crm_str_eq(const char *a, const char *b, gboolean use_case)
const char * placement_strategy
int remote_reconnect_interval
gboolean remote_maintenance
#define crm_config_warn(fmt...)
#define XML_ATTR_TRANSITION_KEY
const char * crm_xml_add(xmlNode *node, const char *name, const char *value)
GHashTable * node_hash_from_list(GListPtr list)
gboolean xml_contains_remote_node(xmlNode *xml)
node_t * pe_find_node_any(GListPtr node_list, const char *id, const char *uname)
const char * fail2text(enum action_fail_response fail)
void add_hash_param(GHashTable *hash, const char *name, const char *value)
#define pe_flag_quick_location
#define pe_rsc_start_pending
#define XML_LRM_TAG_RESOURCES
#define crm_err(fmt, args...)
resource_t *(* find_rsc)(resource_t *parent, const char *search, node_t *node, int flags)
#define XML_CIB_TAG_TICKET_STATE
void resource_location(resource_t *rsc, node_t *node, int score, const char *tag, pe_working_set_t *data_set)
xmlXPathObjectPtr xpath_search(xmlNode *xml_top, const char *path)
void pe_fence_node(pe_working_set_t *data_set, node_t *node, const char *reason)
Schedule a fence action for a node.
ticket_t * ticket_new(const char *ticket_id, pe_working_set_t *data_set)
bool remote_id_conflict(const char *remote_name, pe_working_set_t *data)
enum rsc_role_e next_role
#define XML_ATTR_HAVE_WATCHDOG
#define XML_NODE_ATTR_RSC_DISCOVERY
int compare_version(const char *version1, const char *version2)
gboolean rsc_discovery_enabled
#define pe_flag_remove_after_stop
#define CRMD_ACTION_METADATA
#define pe_rsc_failure_ignored
#define XML_LRM_ATTR_CALLID
#define CRMD_ACTION_MIGRATE
#define XML_NVPAIR_ATTR_VALUE
enum rsc_role_e fail_role
gboolean remote_requires_reset
#define XML_ATTR_CRM_VERSION
#define XML_LRM_ATTR_OPSTATUS
gboolean unpack_lrm_resources(node_t *node, xmlNode *lrm_rsc_list, pe_working_set_t *data_set)
int pe_get_failcount(node_t *node, resource_t *rsc, time_t *last_failure, uint32_t flags, xmlNode *xml_op, pe_working_set_t *data_set)
#define CRMD_JOINSTATE_PENDING
enum pe_action_flags flags
GListPtr find_operations(const char *rsc, const char *node, gboolean active_filter, pe_working_set_t *data_set)
#define XML_NODE_JOIN_STATE
void pe_free_action(action_t *action)
#define pe_flag_have_quorum
void destroy_ticket(gpointer data)
#define XML_CIB_TAG_STATUS
#define XML_CIB_TAG_OBJ_REF
void unpack_instance_attributes(xmlNode *top, xmlNode *xml_obj, const char *set_name, GHashTable *node_hash, GHashTable *hash, const char *always_first, gboolean overwrite, crm_time_t *now)
#define pe_flag_is_managed_default
gboolean remote_was_fenced
#define XML_NODE_IN_CLUSTER
#define pe_flag_stop_action_orphans
gboolean crm_is_true(const char *s)
#define CRM_ATTR_SITE_NAME
void calculate_active_ops(GListPtr sorted_op_list, int *start_index, int *stop_index)
#define XML_CIB_TAG_GROUP
CRM_TRACE_INIT_DATA(pe_status)
#define XML_LRM_TAG_RSC_OP
#define pe_rsc_trace(rsc, fmt, args...)
#define pe_flag_symmetric_cluster
char * generate_op_key(const char *rsc_id, const char *op_type, int interval)
Generate an operation key.
void print_resource(int log_level, const char *pre_text, resource_t *rsc, gboolean details)
gboolean unpack_tags(xmlNode *xml_tags, pe_working_set_t *data_set)
resource_t * pe_find_resource(GListPtr rsc_list, const char *id_rh)
#define safe_str_eq(a, b)
op_digest_cache_t * rsc_action_digest_cmp(resource_t *rsc, xmlNode *xml_op, node_t *node, pe_working_set_t *data_set)
char * crm_strdup_printf(char const *format,...) __attribute__((__format__(__printf__
#define XML_LRM_ATTR_MIGRATE_SOURCE
void freeXpathObject(xmlXPathObjectPtr xpathObj)
gint sort_node_uname(gconstpointer a, gconstpointer b)
#define XML_CIB_TAG_TICKETS
#define crm_info(fmt, args...)
char * digest_restart_calc
void g_hash_destroy_str(gpointer data)
GHashTable * template_rsc_sets
#define pe_flag_concurrent_fencing
#define pe_flag_start_failure_fatal
#define pe_flag_stonith_enabled
enum crm_ais_msg_types type
#define pe_warn_once(pe_wo_bit, fmt...)
#define pe_rsc_info(rsc, fmt, args...)
#define CRMD_ACTION_STATUS