diff --git a/daemons/attrd/attrd_cib.c b/daemons/attrd/attrd_cib.c
index 848ec77dac..359e07fd83 100644
--- a/daemons/attrd/attrd_cib.c
+++ b/daemons/attrd/attrd_cib.c
@@ -1,704 +1,704 @@
/*
* Copyright 2013-2024 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 <errno.h>
#include <inttypes.h> // PRIu32
#include <stdbool.h>
#include <stdlib.h>
#include <glib.h>
#include <crm/msg_xml.h>
#include <crm/common/logging.h>
#include <crm/common/results.h>
#include <crm/common/strings_internal.h>
#include <crm/common/xml.h>
#include <crm/cluster/internal.h> // pcmk__get_node()
#include "pacemaker-attrd.h"
static int last_cib_op_done = 0;
static void write_attribute(attribute_t *a, bool ignore_delay);
static void
attrd_cib_destroy_cb(gpointer user_data)
{
cib_t *cib = user_data;
cib->cmds->signoff(cib);
if (attrd_shutting_down(false)) {
crm_info("Disconnected from the CIB manager");
} else {
// @TODO This should trigger a reconnect, not a shutdown
crm_crit("Lost connection to the CIB manager, shutting down");
attrd_exit_status = CRM_EX_DISCONNECT;
attrd_shutdown(0);
}
}
static void
attrd_cib_updated_cb(const char *event, xmlNode *msg)
{
const xmlNode *patchset = NULL;
const char *client_name = NULL;
if (attrd_shutting_down(true)) {
return;
}
if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) {
return;
}
if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS)) {
mainloop_set_trigger(attrd_config_read);
}
if (!attrd_election_won()) {
// Don't write attributes if we're not the writer
return;
}
client_name = crm_element_value(msg, F_CIB_CLIENTNAME);
if (!cib__client_triggers_refresh(client_name)) {
// The CIB is still accurate
return;
}
if (cib__element_in_patchset(patchset, PCMK_XE_NODES)
|| cib__element_in_patchset(patchset, PCMK_XE_STATUS)) {
/* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS
* section. Write transient attributes to ensure they're up-to-date in
* the CIB.
*/
if (client_name == NULL) {
client_name = crm_element_value(msg, F_CIB_CLIENTID);
}
crm_notice("Updating all attributes after %s event triggered by %s",
event, pcmk__s(client_name, "(unidentified client)"));
attrd_write_attributes(attrd_write_all);
}
}
int
attrd_cib_connect(int max_retry)
{
static int attempts = 0;
int rc = -ENOTCONN;
the_cib = cib_new();
if (the_cib == NULL) {
return -ENOTCONN;
}
do {
if (attempts > 0) {
sleep(attempts);
}
attempts++;
crm_debug("Connection attempt %d to the CIB manager", attempts);
rc = the_cib->cmds->signon(the_cib, T_ATTRD, cib_command);
} while ((rc != pcmk_ok) && (attempts < max_retry));
if (rc != pcmk_ok) {
crm_err("Connection to the CIB manager failed: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
goto cleanup;
}
crm_debug("Connected to the CIB manager after %d attempts", attempts);
rc = the_cib->cmds->set_connection_dnotify(the_cib, attrd_cib_destroy_cb);
if (rc != pcmk_ok) {
crm_err("Could not set disconnection callback");
goto cleanup;
}
rc = the_cib->cmds->add_notify_callback(the_cib, T_CIB_DIFF_NOTIFY,
attrd_cib_updated_cb);
if (rc != pcmk_ok) {
crm_err("Could not set CIB notification callback");
goto cleanup;
}
return pcmk_ok;
cleanup:
cib__clean_up_connection(&the_cib);
return -ENOTCONN;
}
void
attrd_cib_disconnect(void)
{
CRM_CHECK(the_cib != NULL, return);
the_cib->cmds->del_notify_callback(the_cib, T_CIB_DIFF_NOTIFY,
attrd_cib_updated_cb);
cib__clean_up_connection(&the_cib);
}
static void
attrd_erase_cb(xmlNode *msg, int call_id, int rc, xmlNode *output,
void *user_data)
{
const char *node = pcmk__s((const char *) user_data, "a node");
if (rc == pcmk_ok) {
crm_info("Cleared transient node attributes for %s from CIB", node);
} else {
crm_err("Unable to clear transient node attributes for %s from CIB: %s",
node, pcmk_strerror(rc));
}
}
#define XPATH_TRANSIENT "//" PCMK__XE_NODE_STATE \
"[@" PCMK_XA_UNAME "='%s']" \
- "/" XML_TAG_TRANSIENT_NODEATTRS
+ "/" PCMK__XE_TRANSIENT_ATTRIBUTES
/*!
* \internal
* \brief Wipe all transient node attributes for a node from the CIB
*
* \param[in] node Node to clear attributes for
*/
void
attrd_cib_erase_transient_attrs(const char *node)
{
int call_id = 0;
char *xpath = NULL;
CRM_CHECK(node != NULL, return);
xpath = crm_strdup_printf(XPATH_TRANSIENT, node);
crm_debug("Clearing transient node attributes for %s from CIB using %s",
node, xpath);
call_id = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath);
free(xpath);
// strdup() is just for logging here, so ignore failure
the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE,
strdup(node), "attrd_erase_cb",
attrd_erase_cb, free);
}
/*!
* \internal
* \brief Prepare the CIB after cluster is connected
*/
void
attrd_cib_init(void)
{
/* We have no attribute values in memory, so wipe the CIB to match. This is
* normally done by the DC's controller when this node leaves the cluster, but
* this handles the case where the node restarted so quickly that the
* cluster layer didn't notice.
*
* \todo If pacemaker-attrd respawns after crashing (see PCMK_ENV_RESPAWNED),
* ideally we'd skip this and sync our attributes from the writer.
* However, currently we reject any values for us that the writer has, in
* attrd_peer_update().
*/
attrd_cib_erase_transient_attrs(attrd_cluster->uname);
// Set a trigger for reading the CIB (for the alerts section)
attrd_config_read = mainloop_add_trigger(G_PRIORITY_HIGH, attrd_read_options, NULL);
// Always read the CIB at start-up
mainloop_set_trigger(attrd_config_read);
}
static gboolean
attribute_timer_cb(gpointer data)
{
attribute_t *a = data;
crm_trace("Dampen interval expired for %s", a->id);
attrd_write_or_elect_attribute(a);
return FALSE;
}
static void
attrd_cib_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data)
{
int level = LOG_ERR;
GHashTableIter iter;
const char *peer = NULL;
attribute_value_t *v = NULL;
char *name = user_data;
attribute_t *a = g_hash_table_lookup(attributes, name);
if(a == NULL) {
crm_info("Attribute %s no longer exists", name);
return;
}
a->update = 0;
if (rc == pcmk_ok && call_id < 0) {
rc = call_id;
}
switch (rc) {
case pcmk_ok:
level = LOG_INFO;
last_cib_op_done = call_id;
if (a->timer && !a->timeout_ms) {
// Remove temporary dampening for failed writes
mainloop_timer_del(a->timer);
a->timer = NULL;
}
break;
case -pcmk_err_diff_failed: /* When an attr changes while the CIB is syncing */
case -ETIME: /* When an attr changes while there is a DC election */
case -ENXIO: /* When an attr changes while the CIB is syncing a
* newer config from a node that just came up
*/
level = LOG_WARNING;
break;
}
do_crm_log(level, "CIB update %d result for %s: %s " CRM_XS " rc=%d",
call_id, a->id, pcmk_strerror(rc), rc);
g_hash_table_iter_init(&iter, a->values);
while (g_hash_table_iter_next(&iter, (gpointer *) & peer, (gpointer *) & v)) {
do_crm_log(level, "* %s[%s]=%s",
a->id, peer, pcmk__s(v->requested, "(null)"));
if (rc == pcmk_ok) {
pcmk__str_update(&(v->requested), NULL);
} else {
a->changed = true; // Reattempt write below if we are still writer
}
}
if (a->changed && attrd_election_won()) {
if (rc == pcmk_ok) {
/* We deferred a write of a new update because this update was in
* progress. Write out the new value without additional delay.
*/
write_attribute(a, false);
/* We're re-attempting a write because the original failed; delay
* the next attempt so we don't potentially flood the CIB manager
* and logs with a zillion attempts per second.
*
* @TODO We could elect a new writer instead. However, we'd have to
* somehow downgrade our vote, and we'd still need something like this
* if all peers similarly fail to write this attribute (which may
* indicate a corrupted attribute entry rather than a CIB issue).
*/
} else if (a->timer) {
// Attribute has a dampening value, so use that as delay
if (!mainloop_timer_running(a->timer)) {
crm_trace("Delayed re-attempted write for %s by %s",
name, pcmk__readable_interval(a->timeout_ms));
mainloop_timer_start(a->timer);
}
} else {
/* Set a temporary dampening of 2 seconds (timer will continue
* to exist until the attribute's dampening gets set or the
* write succeeds).
*/
a->timer = attrd_add_timer(a->id, 2000, a);
mainloop_timer_start(a->timer);
}
}
}
/*!
* \internal
* \brief Add a set-attribute update request to the current CIB transaction
*
* \param[in] attr Attribute to update
* \param[in] attr_id ID of attribute to update
* \param[in] node_id ID of node for which to update attribute value
* \param[in] set_id ID of attribute set
* \param[in] value New value for attribute
*
* \return Standard Pacemaker return code
*/
static int
add_set_attr_update(const attribute_t *attr, const char *attr_id,
const char *node_id, const char *set_id, const char *value)
{
xmlNode *update = create_xml_node(NULL, PCMK__XE_NODE_STATE);
xmlNode *child = update;
int rc = ENOMEM;
if (child == NULL) {
goto done;
}
crm_xml_add(child, PCMK_XA_ID, node_id);
- child = create_xml_node(child, XML_TAG_TRANSIENT_NODEATTRS);
+ child = create_xml_node(child, PCMK__XE_TRANSIENT_ATTRIBUTES);
if (child == NULL) {
goto done;
}
crm_xml_add(child, PCMK_XA_ID, node_id);
child = create_xml_node(child, attr->set_type);
if (child == NULL) {
goto done;
}
crm_xml_add(child, PCMK_XA_ID, set_id);
child = create_xml_node(child, PCMK_XE_NVPAIR);
if (child == NULL) {
goto done;
}
crm_xml_add(child, PCMK_XA_ID, attr_id);
crm_xml_add(child, PCMK_XA_NAME, attr->id);
crm_xml_add(child, PCMK_XA_VALUE, value);
rc = the_cib->cmds->modify(the_cib, PCMK_XE_STATUS, update,
cib_can_create|cib_transaction);
rc = pcmk_legacy2rc(rc);
done:
free_xml(update);
return rc;
}
/*!
* \internal
* \brief Add an unset-attribute update request to the current CIB transaction
*
* \param[in] attr Attribute to update
* \param[in] attr_id ID of attribute to update
* \param[in] node_id ID of node for which to update attribute value
* \param[in] set_id ID of attribute set
*
* \return Standard Pacemaker return code
*/
static int
add_unset_attr_update(const attribute_t *attr, const char *attr_id,
const char *node_id, const char *set_id)
{
char *xpath = crm_strdup_printf("/" PCMK_XE_CIB
"/" PCMK_XE_STATUS
"/" PCMK__XE_NODE_STATE
"[@" PCMK_XA_ID "='%s']"
- "/" XML_TAG_TRANSIENT_NODEATTRS
+ "/" PCMK__XE_TRANSIENT_ATTRIBUTES
"[@" PCMK_XA_ID "='%s']"
"/%s[@" PCMK_XA_ID "='%s']"
"/" PCMK_XE_NVPAIR
"[@" PCMK_XA_ID "='%s' "
"and @" PCMK_XA_NAME "='%s']",
node_id, node_id, attr->set_type, set_id,
attr_id, attr->id);
int rc = the_cib->cmds->remove(the_cib, xpath, NULL,
cib_xpath|cib_transaction);
free(xpath);
return pcmk_legacy2rc(rc);
}
/*!
* \internal
* \brief Add an attribute update request to the current CIB transaction
*
* \param[in] attr Attribute to update
* \param[in] value New value for attribute
* \param[in] node_id ID of node for which to update attribute value
*
* \return Standard Pacemaker return code
*/
static int
add_attr_update(const attribute_t *attr, const char *value, const char *node_id)
{
char *set_id = NULL;
char *attr_id = NULL;
int rc = pcmk_rc_ok;
if (attr->set_id != NULL) {
pcmk__str_update(&set_id, attr->set_id);
} else {
set_id = crm_strdup_printf("%s-%s", PCMK_XE_STATUS, node_id);
}
crm_xml_sanitize_id(set_id);
if (attr->uuid != NULL) {
pcmk__str_update(&attr_id, attr->uuid);
} else {
attr_id = crm_strdup_printf("%s-%s", set_id, attr->id);
}
crm_xml_sanitize_id(attr_id);
if (value != NULL) {
rc = add_set_attr_update(attr, attr_id, node_id, set_id, value);
} else {
rc = add_unset_attr_update(attr, attr_id, node_id, set_id);
}
free(set_id);
free(attr_id);
return rc;
}
static void
send_alert_attributes_value(attribute_t *a, GHashTable *t)
{
int rc = 0;
attribute_value_t *at = NULL;
GHashTableIter vIter;
g_hash_table_iter_init(&vIter, t);
while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & at)) {
rc = attrd_send_attribute_alert(at->nodename, at->nodeid,
a->id, at->current);
crm_trace("Sent alerts for %s[%s]=%s: nodeid=%d rc=%d",
a->id, at->nodename, at->current, at->nodeid, rc);
}
}
static void
set_alert_attribute_value(GHashTable *t, attribute_value_t *v)
{
attribute_value_t *a_v = NULL;
a_v = calloc(1, sizeof(attribute_value_t));
CRM_ASSERT(a_v != NULL);
a_v->nodeid = v->nodeid;
a_v->nodename = strdup(v->nodename);
pcmk__str_update(&a_v->current, v->current);
g_hash_table_replace(t, a_v->nodename, a_v);
}
mainloop_timer_t *
attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr)
{
return mainloop_timer_add(id, timeout_ms, FALSE, attribute_timer_cb, attr);
}
/*!
* \internal
* \brief Write an attribute's values to the CIB if appropriate
*
* \param[in,out] a Attribute to write
* \param[in] ignore_delay If true, write attribute now regardless of any
* configured delay
*/
static void
write_attribute(attribute_t *a, bool ignore_delay)
{
int private_updates = 0, cib_updates = 0;
attribute_value_t *v = NULL;
GHashTableIter iter;
GHashTable *alert_attribute_value = NULL;
int rc = pcmk_ok;
if (a == NULL) {
return;
}
/* If this attribute will be written to the CIB ... */
if (!stand_alone && !a->is_private) {
/* Defer the write if now's not a good time */
if (a->update && (a->update < last_cib_op_done)) {
crm_info("Write out of '%s' continuing: update %d considered lost",
a->id, a->update);
a->update = 0; // Don't log this message again
} else if (a->update) {
crm_info("Write out of '%s' delayed: update %d in progress",
a->id, a->update);
goto done;
} else if (mainloop_timer_running(a->timer)) {
if (ignore_delay) {
mainloop_timer_stop(a->timer);
crm_debug("Overriding '%s' write delay", a->id);
} else {
crm_info("Delaying write of '%s'", a->id);
goto done;
}
}
// Initiate a transaction for all the peer value updates
CRM_CHECK(the_cib != NULL, goto done);
the_cib->cmds->set_user(the_cib, a->user);
rc = the_cib->cmds->init_transaction(the_cib);
if (rc != pcmk_ok) {
crm_err("Failed to write %s (id %s, set %s): Could not initiate "
"CIB transaction",
a->id, pcmk__s(a->uuid, "n/a"), pcmk__s(a->set_id, "n/a"));
goto done;
}
}
/* Attribute will be written shortly, so clear changed flag */
a->changed = false;
/* We will check all peers' uuids shortly, so initialize this to false */
a->unknown_peer_uuids = false;
/* Attribute will be written shortly, so clear forced write flag */
a->force_write = FALSE;
/* Make the table for the attribute trap */
alert_attribute_value = pcmk__strikey_table(NULL,
attrd_free_attribute_value);
/* Iterate over each peer value of this attribute */
g_hash_table_iter_init(&iter, a->values);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &v)) {
const char *uuid = NULL;
if (pcmk_is_set(v->flags, attrd_value_remote)) {
/* If this is a Pacemaker Remote node, the node's UUID is the same
* as its name, which we already have.
*/
uuid = v->nodename;
} else {
// This will create a cluster node cache entry if none exists
crm_node_t *peer = pcmk__get_node(v->nodeid, v->nodename, NULL,
pcmk__node_search_any);
uuid = peer->uuid;
// Remember peer's node ID if we're just now learning it
if ((peer->id != 0) && (v->nodeid == 0)) {
crm_trace("Learned ID %u for node %s", peer->id, v->nodename);
v->nodeid = peer->id;
}
}
/* If this is a private attribute, no update needs to be sent */
if (stand_alone || a->is_private) {
private_updates++;
continue;
}
// Defer write if this is a cluster node that's never been seen
if (uuid == NULL) {
a->unknown_peer_uuids = true;
crm_notice("Cannot update %s[%s]='%s' now because node's UUID is "
"unknown (will retry if learned)",
a->id, v->nodename, v->current);
continue;
}
// Update this value as part of the CIB transaction we're building
rc = add_attr_update(a, v->current, uuid);
if (rc != pcmk_rc_ok) {
crm_err("Failed to update %s[%s]='%s': %s "
CRM_XS " node uuid=%s id=%" PRIu32,
a->id, v->nodename, v->current, pcmk_rc_str(rc),
uuid, v->nodeid);
continue;
}
crm_debug("Updating %s[%s]=%s (node uuid=%s id=%" PRIu32 ")",
a->id, v->nodename, v->current, uuid, v->nodeid);
cib_updates++;
/* Preservation of the attribute to transmit alert */
set_alert_attribute_value(alert_attribute_value, v);
// Save this value so we can log it when write completes
pcmk__str_update(&(v->requested), v->current);
}
if (private_updates) {
crm_info("Processed %d private change%s for %s, id=%s, set=%s",
private_updates, pcmk__plural_s(private_updates),
a->id, pcmk__s(a->uuid, "n/a"), pcmk__s(a->set_id, "n/a"));
}
if (cib_updates > 0) {
char *id = NULL;
// Commit transaction
a->update = the_cib->cmds->end_transaction(the_cib, true, cib_none);
crm_info("Sent CIB request %d with %d change%s for %s (id %s, set %s)",
a->update, cib_updates, pcmk__plural_s(cib_updates),
a->id, pcmk__s(a->uuid, "n/a"), pcmk__s(a->set_id, "n/a"));
pcmk__str_update(&id, a->id);
if (the_cib->cmds->register_callback_full(the_cib, a->update,
CIB_OP_TIMEOUT_S, FALSE, id,
"attrd_cib_callback",
attrd_cib_callback, free)) {
// Transmit alert of the attribute
send_alert_attributes_value(a, alert_attribute_value);
}
}
done:
// Discard transaction (if any)
if (the_cib != NULL) {
the_cib->cmds->end_transaction(the_cib, false, cib_none);
the_cib->cmds->set_user(the_cib, NULL);
}
if (alert_attribute_value != NULL) {
g_hash_table_destroy(alert_attribute_value);
}
}
/*!
* \internal
* \brief Write out attributes
*
* \param[in] options Group of enum attrd_write_options
*/
void
attrd_write_attributes(uint32_t options)
{
GHashTableIter iter;
attribute_t *a = NULL;
crm_debug("Writing out %s attributes",
pcmk_is_set(options, attrd_write_all)? "all" : "changed");
g_hash_table_iter_init(&iter, attributes);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & a)) {
if (!pcmk_is_set(options, attrd_write_all) && a->unknown_peer_uuids) {
// Try writing this attribute again, in case peer ID was learned
a->changed = true;
} else if (a->force_write) {
/* If the force_write flag is set, write the attribute. */
a->changed = true;
}
if (pcmk_is_set(options, attrd_write_all) || a->changed) {
bool ignore_delay = pcmk_is_set(options, attrd_write_no_delay);
if (a->force_write) {
// Always ignore delay when forced write flag is set
ignore_delay = true;
}
write_attribute(a, ignore_delay);
} else {
crm_trace("Skipping unchanged attribute %s", a->id);
}
}
}
void
attrd_write_or_elect_attribute(attribute_t *a)
{
if (attrd_election_won()) {
write_attribute(a, false);
} else {
attrd_start_election_if_needed();
}
}
diff --git a/daemons/controld/controld_cib.c b/daemons/controld/controld_cib.c
index cc8489b9e7..d1a9b8f124 100644
--- a/daemons/controld/controld_cib.c
+++ b/daemons/controld/controld_cib.c
@@ -1,1070 +1,1070 @@
/*
* 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 <unistd.h> /* sleep */
#include <crm/common/alerts_internal.h>
#include <crm/common/xml.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-controld.h>
// Call ID of the most recent in-progress CIB resource update (or 0 if none)
static int pending_rsc_update = 0;
/*!
* \internal
* \brief Respond to a dropped CIB connection
*
* \param[in] user_data CIB connection that dropped
*/
static void
handle_cib_disconnect(gpointer user_data)
{
CRM_LOG_ASSERT(user_data == controld_globals.cib_conn);
controld_trigger_fsa();
controld_globals.cib_conn->state = cib_disconnected;
if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
// @TODO This should trigger a reconnect, not a shutdown
crm_crit("Lost connection to the CIB manager, shutting down");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
controld_clear_fsa_input_flags(R_CIB_CONNECTED);
} else { // Expected
crm_info("Disconnected from the CIB manager");
}
}
static void
do_cib_updated(const char *event, xmlNode * msg)
{
const xmlNode *patchset = NULL;
const char *client_name = NULL;
crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC));
if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) {
return;
}
if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS)
|| cib__element_in_patchset(patchset, PCMK_XE_CRM_CONFIG)) {
controld_trigger_config();
}
if (!AM_I_DC) {
// We're not in control of the join sequence
return;
}
client_name = crm_element_value(msg, F_CIB_CLIENTNAME);
if (!cib__client_triggers_refresh(client_name)) {
// The CIB is still accurate
return;
}
if (cib__element_in_patchset(patchset, PCMK_XE_NODES)
|| cib__element_in_patchset(patchset, PCMK_XE_STATUS)) {
/* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS
* section. Ensure the node list is up-to-date, and start the join
* process again so we get everyone's current resource history.
*/
if (client_name == NULL) {
client_name = crm_element_value(msg, F_CIB_CLIENTID);
}
crm_notice("Populating nodes and starting an election after %s event "
"triggered by %s",
event, pcmk__s(client_name, "(unidentified client)"));
populate_cib_nodes(node_update_quick|node_update_all, __func__);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
}
}
void
controld_disconnect_cib_manager(void)
{
cib_t *cib_conn = controld_globals.cib_conn;
CRM_ASSERT(cib_conn != NULL);
crm_debug("Disconnecting from the CIB manager");
controld_clear_fsa_input_flags(R_CIB_CONNECTED);
cib_conn->cmds->del_notify_callback(cib_conn, T_CIB_DIFF_NOTIFY,
do_cib_updated);
cib_free_callbacks(cib_conn);
if (cib_conn->state != cib_disconnected) {
cib_conn->cmds->set_secondary(cib_conn,
cib_scope_local|cib_discard_reply);
cib_conn->cmds->signoff(cib_conn);
}
}
/* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */
void
do_cib_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
static int cib_retries = 0;
cib_t *cib_conn = controld_globals.cib_conn;
void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect;
void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated;
int rc = pcmk_ok;
CRM_ASSERT(cib_conn != NULL);
if (pcmk_is_set(action, A_CIB_STOP)) {
if ((cib_conn->state != cib_disconnected)
&& (pending_rsc_update != 0)) {
crm_info("Waiting for resource update %d to complete",
pending_rsc_update);
crmd_fsa_stall(FALSE);
return;
}
controld_disconnect_cib_manager();
}
if (!pcmk_is_set(action, A_CIB_START)) {
return;
}
if (cur_state == S_STOPPING) {
crm_err("Ignoring request to connect to the CIB manager after "
"shutdown");
return;
}
rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
cib_command_nonblocking);
if (rc != pcmk_ok) {
// A short wait that usually avoids stalling the FSA
sleep(1);
rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
cib_command_nonblocking);
}
if (rc != pcmk_ok) {
crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc));
} else if (cib_conn->cmds->set_connection_dnotify(cib_conn,
dnotify_fn) != pcmk_ok) {
crm_err("Could not set dnotify callback");
} else if (cib_conn->cmds->add_notify_callback(cib_conn,
T_CIB_DIFF_NOTIFY,
update_cb) != pcmk_ok) {
crm_err("Could not set CIB notification callback (update)");
} else {
controld_set_fsa_input_flags(R_CIB_CONNECTED);
cib_retries = 0;
}
if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
cib_retries++;
if (cib_retries < 30) {
crm_warn("Couldn't complete CIB registration %d times... "
"pause and retry", cib_retries);
controld_start_wait_timer();
crmd_fsa_stall(FALSE);
} else {
crm_err("Could not complete CIB registration %d times... "
"hard error", cib_retries);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
}
#define MIN_CIB_OP_TIMEOUT (30)
/*!
* \internal
* \brief Get the timeout (in seconds) that should be used with CIB operations
*
* \return The maximum of 30 seconds, the value of the PCMK_cib_timeout
* environment variable, or 10 seconds times one more than the number of
* nodes in the cluster.
*/
unsigned int
cib_op_timeout(void)
{
// @COMPAT: Drop env_timeout at 3.0.0
static int env_timeout = -1;
unsigned int calculated_timeout = 0;
if (env_timeout == -1) {
const char *env = pcmk__env_option(PCMK__ENV_CIB_TIMEOUT);
pcmk__scan_min_int(env, &env_timeout, MIN_CIB_OP_TIMEOUT);
crm_trace("Minimum CIB op timeout: %ds (environment: %s)",
env_timeout, (env? env : "none"));
}
calculated_timeout = 1 + crm_active_peers();
if (crm_remote_peer_cache) {
calculated_timeout += g_hash_table_size(crm_remote_peer_cache);
}
calculated_timeout *= 10;
calculated_timeout = QB_MAX(calculated_timeout, env_timeout);
crm_trace("Calculated timeout: %us", calculated_timeout);
if (controld_globals.cib_conn) {
controld_globals.cib_conn->call_timeout = calculated_timeout;
}
return calculated_timeout;
}
/*!
* \internal
* \brief Get CIB call options to use local scope if primary is unavailable
*
* \return CIB call options
*/
int
crmd_cib_smart_opt(void)
{
int call_opt = cib_none;
if ((controld_globals.fsa_state == S_ELECTION)
|| (controld_globals.fsa_state == S_PENDING)) {
crm_info("Sending update to local CIB in state: %s",
fsa_state2string(controld_globals.fsa_state));
cib__set_call_options(call_opt, "update", cib_scope_local);
}
return call_opt;
}
static void
cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output,
void *user_data)
{
char *desc = user_data;
if (rc == 0) {
crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id);
} else {
crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d",
desc, call_id, pcmk_strerror(rc), rc);
}
}
// Searches for various portions of PCMK__XE_NODE_STATE to delete
// Match a particular node's PCMK__XE_NODE_STATE (takes node name 1x)
#define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
// Node's lrm section (name 1x)
#define XPATH_NODE_LRM XPATH_NODE_STATE "/" PCMK__XE_LRM
/* Node's PCMK__XE_LRM_RSC_OP entries and PCMK__XE_LRM_RESOURCE entries without
* unexpired lock
* (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x)
*/
#define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" PCMK__XE_LRM_RSC_OP \
"|" XPATH_NODE_STATE \
"//" PCMK__XE_LRM_RESOURCE \
"[not(@" PCMK_OPT_SHUTDOWN_LOCK ") " \
"or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]"
-// Node's transient_attributes section (name 1x)
-#define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" XML_TAG_TRANSIENT_NODEATTRS
+// Node's PCMK__XE_TRANSIENT_ATTRIBUTES section (name 1x)
+#define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" PCMK__XE_TRANSIENT_ATTRIBUTES
// Everything under PCMK__XE_NODE_STATE (name 1x)
#define XPATH_NODE_ALL XPATH_NODE_STATE "/*"
/* Unlocked history + transient attributes
* (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x)
*/
#define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS
/*!
* \internal
* \brief Get the XPath and description of a node state section to be deleted
*
* \param[in] uname Desired node
* \param[in] section Subsection of \c PCMK__XE_NODE_STATE to be deleted
* \param[out] xpath Where to store XPath of \p section
* \param[out] desc If not \c NULL, where to store description of \p section
*/
void
controld_node_state_deletion_strings(const char *uname,
enum controld_section_e section,
char **xpath, char **desc)
{
const char *desc_pre = NULL;
// Shutdown locks that started before this time are expired
long long expire = (long long) time(NULL)
- controld_globals.shutdown_lock_limit;
switch (section) {
case controld_section_lrm:
*xpath = crm_strdup_printf(XPATH_NODE_LRM, uname);
desc_pre = "resource history";
break;
case controld_section_lrm_unlocked:
*xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED,
uname, uname, expire);
desc_pre = "resource history (other than shutdown locks)";
break;
case controld_section_attrs:
*xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname);
desc_pre = "transient attributes";
break;
case controld_section_all:
*xpath = crm_strdup_printf(XPATH_NODE_ALL, uname);
desc_pre = "all state";
break;
case controld_section_all_unlocked:
*xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED,
uname, uname, expire, uname);
desc_pre = "all state (other than shutdown locks)";
break;
default:
// We called this function incorrectly
CRM_ASSERT(false);
break;
}
if (desc != NULL) {
*desc = crm_strdup_printf("%s for node %s", desc_pre, uname);
}
}
/*!
* \internal
* \brief Delete subsection of a node's CIB \c PCMK__XE_NODE_STATE
*
* \param[in] uname Desired node
* \param[in] section Subsection of \c PCMK__XE_NODE_STATE to delete
* \param[in] options CIB call options to use
*/
void
controld_delete_node_state(const char *uname, enum controld_section_e section,
int options)
{
cib_t *cib = controld_globals.cib_conn;
char *xpath = NULL;
char *desc = NULL;
int cib_rc = pcmk_ok;
CRM_ASSERT((uname != NULL) && (cib != NULL));
controld_node_state_deletion_strings(uname, section, &xpath, &desc);
cib__set_call_options(options, "node state deletion",
cib_xpath|cib_multiple);
cib_rc = cib->cmds->remove(cib, xpath, NULL, options);
fsa_register_cib_callback(cib_rc, desc, cib_delete_callback);
crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s",
desc, cib_rc, xpath);
// CIB library handles freeing desc
free(xpath);
}
// Takes node name and resource ID
#define XPATH_RESOURCE_HISTORY "//" PCMK__XE_NODE_STATE \
"[@" PCMK_XA_UNAME "='%s']/" \
PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE \
"[@" PCMK_XA_ID "='%s']"
// @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks
/*!
* \internal
* \brief Clear resource history from CIB for a given resource and node
*
* \param[in] rsc_id ID of resource to be cleared
* \param[in] node Node whose resource history should be cleared
* \param[in] user_name ACL user name to use
* \param[in] call_options CIB call options
*
* \return Standard Pacemaker return code
*/
int
controld_delete_resource_history(const char *rsc_id, const char *node,
const char *user_name, int call_options)
{
char *desc = NULL;
char *xpath = NULL;
int rc = pcmk_rc_ok;
cib_t *cib = controld_globals.cib_conn;
CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL);
desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node);
if (cib == NULL) {
crm_err("Unable to clear %s: no CIB connection", desc);
free(desc);
return ENOTCONN;
}
// Ask CIB to delete the entry
xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id);
cib->cmds->set_user(cib, user_name);
rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath);
cib->cmds->set_user(cib, NULL);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
crm_err("Could not delete resource status of %s on %s%s%s: %s "
CRM_XS " rc=%d", rsc_id, node,
(user_name? " for user " : ""), (user_name? user_name : ""),
pcmk_rc_str(rc), rc);
free(desc);
free(xpath);
return rc;
}
if (pcmk_is_set(call_options, cib_sync_call)) {
if (pcmk_is_set(call_options, cib_dryrun)) {
crm_debug("Deletion of %s would succeed", desc);
} else {
crm_debug("Deletion of %s succeeded", desc);
}
free(desc);
} else {
crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s",
desc, rc, xpath);
fsa_register_cib_callback(rc, desc, cib_delete_callback);
// CIB library handles freeing desc
}
free(xpath);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Build XML and string of parameters meeting some criteria, for digest
*
* \param[in] op Executor event with parameter table to use
* \param[in] metadata Parsed meta-data for executed resource agent
* \param[in] param_type Flag used for selection criteria
* \param[out] result Will be set to newly created XML with selected
* parameters as attributes
*
* \return Newly allocated space-separated string of parameter names
* \note Selection criteria varies by param_type: for the restart digest, we
* want parameters that are *not* marked reloadable (OCF 1.1) or that
* *are* marked unique (pre-1.1), for both string and XML results; for the
* secure digest, we want parameters that *are* marked private for the
* string, but parameters that are *not* marked private for the XML.
* \note It is the caller's responsibility to free the string return value with
* \p g_string_free() and the XML result with \p free_xml().
*/
static GString *
build_parameter_list(const lrmd_event_data_t *op,
const struct ra_metadata_s *metadata,
enum ra_param_flags_e param_type, xmlNode **result)
{
GString *list = NULL;
*result = create_xml_node(NULL, PCMK_XE_PARAMETERS);
/* Consider all parameters only except private ones to be consistent with
* what scheduler does with calculate_secure_digest().
*/
if (param_type == ra_param_private
&& compare_version(controld_globals.dc_version, "3.16.0") >= 0) {
g_hash_table_foreach(op->params, hash2field, *result);
pcmk__filter_op_for_digest(*result);
}
for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) {
struct ra_param_s *param = (struct ra_param_s *) iter->data;
bool accept_for_list = false;
bool accept_for_xml = false;
switch (param_type) {
case ra_param_reloadable:
accept_for_list = !pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = accept_for_list;
break;
case ra_param_unique:
accept_for_list = pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = accept_for_list;
break;
case ra_param_private:
accept_for_list = pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = !accept_for_list;
break;
}
if (accept_for_list) {
crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type));
if (list == NULL) {
// We will later search for " WORD ", so start list with a space
pcmk__add_word(&list, 256, " ");
}
pcmk__add_word(&list, 0, param->rap_name);
} else {
crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type));
}
if (accept_for_xml) {
const char *v = g_hash_table_lookup(op->params, param->rap_name);
if (v != NULL) {
crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v);
crm_xml_add(*result, param->rap_name, v);
}
} else {
crm_trace("Removing attr %s from the xml result", param->rap_name);
xml_remove_prop(*result, param->rap_name);
}
}
if (list != NULL) {
// We will later search for " WORD ", so end list with a space
pcmk__add_word(&list, 0, " ");
}
return list;
}
static void
append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
GString *list = NULL;
char *digest = NULL;
xmlNode *restart = NULL;
CRM_LOG_ASSERT(op->params != NULL);
if (op->interval_ms > 0) {
/* monitors are not reloadable */
return;
}
if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) {
/* Add parameters not marked reloadable to the PCMK__XA_OP_FORCE_RESTART
* list
*/
list = build_parameter_list(op, metadata, ra_param_reloadable,
&restart);
} else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) {
/* @COMPAT pre-OCF-1.1 resource agents
*
* Before OCF 1.1, Pacemaker abused "unique=0" to indicate
* reloadability. Add any parameters with unique="1" to the
* PCMK__XA_OP_FORCE_RESTART list.
*/
list = build_parameter_list(op, metadata, ra_param_unique, &restart);
} else {
// Resource does not support agent reloads
return;
}
digest = calculate_operation_digest(restart, version);
/* Add PCMK__XA_OP_FORCE_RESTART and PCMK__XA_OP_RESTART_DIGEST to indicate
* the resource supports reload, no matter if it actually supports any
* reloadable parameters
*/
crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART,
(list == NULL)? "" : (const char *) list->str);
crm_xml_add(update, PCMK__XA_OP_RESTART_DIGEST, digest);
if ((list != NULL) && (list->len > 0)) {
crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
} else {
crm_trace("%s: %s", op->rsc_id, digest);
}
if (list != NULL) {
g_string_free(list, TRUE);
}
free_xml(restart);
free(digest);
}
static void
append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
GString *list = NULL;
char *digest = NULL;
xmlNode *secure = NULL;
CRM_LOG_ASSERT(op->params != NULL);
/* To keep PCMK__XA_OP_SECURE_PARAMS short, we want it to contain the secure
* parameters but PCMK__XA_OP_SECURE_DIGEST to be based on the insecure ones
*/
list = build_parameter_list(op, metadata, ra_param_private, &secure);
if (list != NULL) {
digest = calculate_operation_digest(secure, version);
crm_xml_add(update, PCMK__XA_OP_SECURE_PARAMS,
(const char *) list->str);
crm_xml_add(update, PCMK__XA_OP_SECURE_DIGEST, digest);
crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
g_string_free(list, TRUE);
} else {
crm_trace("%s: no secure parameters", op->rsc_id);
}
free_xml(secure);
free(digest);
}
/*!
* \internal
* \brief Create XML for a resource history entry
*
* \param[in] func Function name of caller
* \param[in,out] parent XML to add entry to
* \param[in] rsc Affected resource
* \param[in,out] op Action to add an entry for (or NULL to do nothing)
* \param[in] node_name Node where action occurred
*/
void
controld_add_resource_history_xml_as(const char *func, xmlNode *parent,
const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op,
const char *node_name)
{
int target_rc = 0;
xmlNode *xml_op = NULL;
struct ra_metadata_s *metadata = NULL;
const char *caller_version = NULL;
lrm_state_t *lrm_state = NULL;
if (op == NULL) {
return;
}
target_rc = rsc_op_expected_rc(op);
caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET);
CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET);
xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc,
controld_globals.our_nodename, func);
if (xml_op == NULL) {
return;
}
if ((rsc == NULL) || (op->params == NULL)
|| !crm_op_needs_metadata(rsc->standard, op->op_type)) {
crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)",
op->op_type, op->rsc_id, op->params, rsc);
return;
}
lrm_state = lrm_state_find(node_name);
if (lrm_state == NULL) {
crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT
" because we have no connection to executor for %s",
op->rsc_id, op->op_type, op->interval_ms, node_name);
return;
}
/* Ideally the metadata is cached, and the agent is just a fallback.
*
* @TODO Go through all callers and ensure they get metadata asynchronously
* first.
*/
metadata = controld_get_rsc_metadata(lrm_state, rsc,
controld_metadata_from_agent
|controld_metadata_from_cache);
if (metadata == NULL) {
return;
}
crm_trace("Including additional digests for %s:%s:%s",
rsc->standard, rsc->provider, rsc->type);
append_restart_list(op, metadata, xml_op, caller_version);
append_secure_list(op, metadata, xml_op, caller_version);
return;
}
/*!
* \internal
* \brief Record an action as pending in the CIB, if appropriate
*
* \param[in] node_name Node where the action is pending
* \param[in] rsc Resource that action is for
* \param[in,out] op Pending action
*
* \return true if action was recorded in CIB, otherwise false
*/
bool
controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op)
{
const char *record_pending = NULL;
CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL),
return false);
// Never record certain operation types as pending
if ((op->op_type == NULL) || (op->params == NULL)
|| !controld_action_is_recordable(op->op_type)) {
return false;
}
// Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true)
record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING);
if ((record_pending != NULL) && !crm_is_true(record_pending)) {
return false;
}
op->call_id = -1;
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
crm_debug("Recording pending %s-interval %s for %s on %s in the CIB",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
node_name);
controld_update_resource_history(node_name, rsc, op, 0);
return true;
}
static void
cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
switch (rc) {
case pcmk_ok:
case -pcmk_err_diff_failed:
case -pcmk_err_diff_resync:
crm_trace("Resource history update completed (call=%d rc=%d)",
call_id, rc);
break;
default:
if (call_id > 0) {
crm_warn("Resource history update %d failed: %s "
CRM_XS " rc=%d", call_id, pcmk_strerror(rc), rc);
} else {
crm_warn("Resource history update failed: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
}
}
if (call_id == pending_rsc_update) {
pending_rsc_update = 0;
controld_trigger_fsa();
}
}
/* Only successful stops, and probes that found the resource inactive, get locks
* recorded in the history. This ensures the resource stays locked to the node
* until it is active there again after the node comes back up.
*/
static bool
should_preserve_lock(lrmd_event_data_t *op)
{
if (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) {
return false;
}
if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) {
return true;
}
if (!strcmp(op->op_type, PCMK_ACTION_MONITOR)
&& (op->rc == PCMK_OCF_NOT_RUNNING)) {
return true;
}
return false;
}
/*!
* \internal
* \brief Request a CIB update
*
* \param[in] section Section of CIB to update
* \param[in] data New XML of CIB section to update
* \param[in] options CIB call options
* \param[in] callback If not \c NULL, set this as the operation callback
*
* \return Standard Pacemaker return code
*
* \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is
* stored in \p pending_rsc_update on success.
*/
int
controld_update_cib(const char *section, xmlNode *data, int options,
void (*callback)(xmlNode *, int, int, xmlNode *, void *))
{
cib_t *cib = controld_globals.cib_conn;
int cib_rc = -ENOTCONN;
CRM_ASSERT(data != NULL);
if (cib != NULL) {
cib_rc = cib->cmds->modify(cib, section, data, options);
if (cib_rc >= 0) {
crm_debug("Submitted CIB update %d for %s section",
cib_rc, section);
}
}
if (callback == NULL) {
if (cib_rc < 0) {
crm_err("Failed to update CIB %s section: %s",
section, pcmk_rc_str(pcmk_legacy2rc(cib_rc)));
}
} else {
if ((cib_rc >= 0) && (callback == cib_rsc_callback)) {
/* Checking for a particular callback is a little hacky, but it
* didn't seem worth adding an output argument for cib_rc for just
* one use case.
*/
pending_rsc_update = cib_rc;
}
fsa_register_cib_callback(cib_rc, NULL, callback);
}
return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc);
}
/*!
* \internal
* \brief Update resource history entry in CIB
*
* \param[in] node_name Node where action occurred
* \param[in] rsc Resource that action is for
* \param[in,out] op Action to record
* \param[in] lock_time If nonzero, when resource was locked to node
*
* \note On success, the CIB update's call ID will be stored in
* pending_rsc_update.
*/
void
controld_update_resource_history(const char *node_name,
const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op, time_t lock_time)
{
xmlNode *update = NULL;
xmlNode *xml = NULL;
int call_opt = crmd_cib_smart_opt();
const char *node_id = NULL;
const char *container = NULL;
CRM_CHECK((node_name != NULL) && (op != NULL), return);
if (rsc == NULL) {
crm_warn("Resource %s no longer exists in the executor", op->rsc_id);
controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id);
return;
}
// <status>
update = create_xml_node(NULL, PCMK_XE_STATUS);
// <node_state ...>
xml = create_xml_node(update, PCMK__XE_NODE_STATE);
if (pcmk__str_eq(node_name, controld_globals.our_nodename,
pcmk__str_casei)) {
node_id = controld_globals.our_uuid;
} else {
node_id = node_name;
pcmk__xe_set_bool_attr(xml, XML_NODE_IS_REMOTE, true);
}
crm_xml_add(xml, PCMK_XA_ID, node_id);
crm_xml_add(xml, PCMK_XA_UNAME, node_name);
crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__);
// <lrm ...>
xml = create_xml_node(xml, PCMK__XE_LRM);
crm_xml_add(xml, PCMK_XA_ID, node_id);
// <lrm_resources>
xml = create_xml_node(xml, PCMK__XE_LRM_RESOURCES);
// <lrm_resource ...>
xml = create_xml_node(xml, PCMK__XE_LRM_RESOURCE);
crm_xml_add(xml, PCMK_XA_ID, op->rsc_id);
crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard);
crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider);
crm_xml_add(xml, PCMK_XA_TYPE, rsc->type);
if (lock_time != 0) {
/* Actions on a locked resource should either preserve the lock by
* recording it with the action result, or clear it.
*/
if (!should_preserve_lock(op)) {
lock_time = 0;
}
crm_xml_add_ll(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time);
}
if (op->params != NULL) {
container = g_hash_table_lookup(op->params,
CRM_META "_" PCMK__META_CONTAINER);
if (container != NULL) {
crm_trace("Resource %s is a part of container resource %s",
op->rsc_id, container);
crm_xml_add(xml, PCMK__META_CONTAINER, container);
}
}
// <lrm_resource_op ...> (possibly more than one)
controld_add_resource_history_xml(xml, rsc, op, node_name);
/* Update CIB asynchronously. Even if it fails, the resource state should be
* discovered during the next election. Worst case, the node is wrongly
* fenced for running a resource it isn't.
*/
crm_log_xml_trace(update, __func__);
controld_update_cib(PCMK_XE_STATUS, update, call_opt, cib_rsc_callback);
free_xml(update);
}
/*!
* \internal
* \brief Erase an LRM history entry from the CIB, given the operation data
*
* \param[in] op Operation whose history should be deleted
*/
void
controld_delete_action_history(const lrmd_event_data_t *op)
{
xmlNode *xml_top = NULL;
CRM_CHECK(op != NULL, return);
xml_top = create_xml_node(NULL, PCMK__XE_LRM_RSC_OP);
crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id);
crm_xml_add(xml_top, PCMK__XA_TRANSITION_KEY, op->user_data);
if (op->interval_ms > 0) {
char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
/* Avoid deleting last_failure too (if it was a result of this recurring op failing) */
crm_xml_add(xml_top, PCMK_XA_ID, op_id);
free(op_id);
}
crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)",
op->rsc_id, op->op_type, op->interval_ms, op->call_id);
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn,
PCMK_XE_STATUS, xml_top, cib_none);
crm_log_xml_trace(xml_top, "op:cancel");
free_xml(xml_top);
}
/* Define xpath to find LRM resource history entry by node and resource */
#define XPATH_HISTORY \
"/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \
"/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \
"/" PCMK__XE_LRM_RSC_OP
/* ... and also by operation key */
#define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']"
/* ... and also by operation key and operation call ID */
#define XPATH_HISTORY_CALL XPATH_HISTORY \
"[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']"
/* ... and also by operation key and original operation key */
#define XPATH_HISTORY_ORIG XPATH_HISTORY \
"[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']"
/*!
* \internal
* \brief Delete a last_failure resource history entry from the CIB
*
* \param[in] rsc_id Name of resource to clear history for
* \param[in] node Name of node to clear history for
* \param[in] action If specified, delete only if this was failed action
* \param[in] interval_ms If \p action is specified, it has this interval
*/
void
controld_cib_delete_last_failure(const char *rsc_id, const char *node,
const char *action, guint interval_ms)
{
char *xpath = NULL;
char *last_failure_key = NULL;
CRM_CHECK((rsc_id != NULL) && (node != NULL), return);
// Generate XPath to match desired entry
last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0);
if (action == NULL) {
xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id,
last_failure_key);
} else {
char *action_key = pcmk__op_key(rsc_id, action, interval_ms);
xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id,
last_failure_key, action_key);
free(action_key);
}
free(last_failure_key);
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
NULL, cib_xpath);
free(xpath);
}
/*!
* \internal
* \brief Delete resource history entry from the CIB, given operation key
*
* \param[in] rsc_id Name of resource to clear history for
* \param[in] node Name of node to clear history for
* \param[in] key Operation key of operation to clear history for
* \param[in] call_id If specified, delete entry only if it has this call ID
*/
void
controld_delete_action_history_by_key(const char *rsc_id, const char *node,
const char *key, int call_id)
{
char *xpath = NULL;
CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return);
if (call_id > 0) {
xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key,
call_id);
} else {
xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key);
}
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
NULL, cib_xpath);
free(xpath);
}
diff --git a/daemons/controld/controld_te_callbacks.c b/daemons/controld/controld_te_callbacks.c
index 0f376cf43c..71932964bd 100644
--- a/daemons/controld/controld_te_callbacks.c
+++ b/daemons/controld/controld_te_callbacks.c
@@ -1,699 +1,699 @@
/*
* 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/stat.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/msg_xml.h>
#include <crm/cluster.h> /* For ONLINESTATUS etc */
#include <pacemaker-controld.h>
void te_update_confirm(const char *event, xmlNode * msg);
#define RSC_OP_PREFIX "//" PCMK__XE_DIFF_ADDED "//" PCMK_XE_CIB \
"//" PCMK__XE_LRM_RSC_OP "[@" PCMK_XA_ID "='"
// An explicit PCMK_OPT_SHUTDOWN_LOCK of 0 means the lock has been cleared
static bool
shutdown_lock_cleared(xmlNode *lrm_resource)
{
time_t shutdown_lock = 0;
return (crm_element_value_epoch(lrm_resource, PCMK_OPT_SHUTDOWN_LOCK,
&shutdown_lock) == pcmk_ok)
&& (shutdown_lock == 0);
}
static void
te_update_diff_v1(const char *event, xmlNode *diff)
{
int lpc, max;
xmlXPathObject *xpathObj = NULL;
GString *rsc_op_xpath = NULL;
CRM_CHECK(diff != NULL, return);
pcmk__output_set_log_level(controld_globals.logger_out, LOG_TRACE);
controld_globals.logger_out->message(controld_globals.logger_out,
"xml-patchset", diff);
if (cib__config_changed_v1(NULL, NULL, &diff)) {
abort_transition(INFINITY, pcmk__graph_restart, "Non-status change",
diff);
goto bail; /* configuration changed */
}
/* Tickets Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" PCMK__XE_DIFF_ADDED
"//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, pcmk__graph_restart,
"Ticket attribute: update", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Tickets Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" PCMK__XE_DIFF_REMOVED
"//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, pcmk__graph_restart,
"Ticket attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Transient Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" PCMK__XE_DIFF_REMOVED
- "//" XML_TAG_TRANSIENT_NODEATTRS);
+ "//" PCMK__XE_TRANSIENT_ATTRIBUTES);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, pcmk__graph_restart,
"Transient attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
// Check for PCMK__XE_LRM_RESOURCE entries
xpathObj = xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" PCMK__XE_DIFF_ADDED
"//" PCMK__XE_LRM_RESOURCE);
max = numXpathResults(xpathObj);
/*
* Updates by, or in response to, graph actions will never affect more than
* one resource at a time, so such updates indicate an LRM refresh. In that
* case, start a new transition rather than check each result individually,
* which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
if ((controld_globals.transition_graph->pending == 0) && (max > 1)) {
crm_debug("Ignoring resource operation updates due to history refresh of %d resources",
max);
crm_log_xml_trace(diff, "lrm-refresh");
abort_transition(INFINITY, pcmk__graph_restart, "History refresh",
NULL);
goto bail;
}
if (max == 1) {
xmlNode *lrm_resource = getXpathResult(xpathObj, 0);
if (shutdown_lock_cleared(lrm_resource)) {
// @TODO would be more efficient to abort once after transition done
abort_transition(INFINITY, pcmk__graph_restart,
"Shutdown lock cleared", lrm_resource);
// Still process results, so we stop timers and update failcounts
}
}
freeXpathObject(xpathObj);
/* Process operation updates */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" PCMK__XE_DIFF_ADDED
"//" PCMK__XE_LRM_RSC_OP);
max = numXpathResults(xpathObj);
if (max > 0) {
int lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
}
freeXpathObject(xpathObj);
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff,
"//" PCMK__XE_DIFF_REMOVED
"//" PCMK__XE_LRM_RSC_OP);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
const char *op_id = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if(match == NULL) { continue; };
op_id = ID(match);
if (rsc_op_xpath == NULL) {
rsc_op_xpath = g_string_new(RSC_OP_PREFIX);
} else {
g_string_truncate(rsc_op_xpath, sizeof(RSC_OP_PREFIX) - 1);
}
pcmk__g_strcat(rsc_op_xpath, op_id, "']", NULL);
op_match = xpath_search(diff, (const char *) rsc_op_xpath->str);
if (numXpathResults(op_match) == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
pcmk__graph_action_t *cancelled = get_cancel_action(op_id, node);
if (cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)",
(const char *) rsc_op_xpath->str, op_id, node);
abort_transition(INFINITY, pcmk__graph_restart,
"Resource op removal", match);
freeXpathObject(op_match);
goto bail;
} else {
crm_debug("Deleted " PCMK__XE_LRM_RSC_OP " %s on %s was for "
"graph event %d",
op_id, node, cancelled->id);
}
}
freeXpathObject(op_match);
}
bail:
freeXpathObject(xpathObj);
if (rsc_op_xpath != NULL) {
g_string_free(rsc_op_xpath, TRUE);
}
}
static void
process_lrm_resource_diff(xmlNode *lrm_resource, const char *node)
{
for (xmlNode *rsc_op = pcmk__xml_first_child(lrm_resource); rsc_op != NULL;
rsc_op = pcmk__xml_next(rsc_op)) {
process_graph_event(rsc_op, node);
}
if (shutdown_lock_cleared(lrm_resource)) {
// @TODO would be more efficient to abort once after transition done
abort_transition(INFINITY, pcmk__graph_restart, "Shutdown lock cleared",
lrm_resource);
}
}
static void
process_resource_updates(const char *node, xmlNode *xml, xmlNode *change,
const char *op, const char *xpath)
{
xmlNode *rsc = NULL;
if (xml == NULL) {
return;
}
if (pcmk__xe_is(xml, PCMK__XE_LRM)) {
xml = first_named_child(xml, PCMK__XE_LRM_RESOURCES);
CRM_CHECK(xml != NULL, return);
}
CRM_CHECK(pcmk__xe_is(xml, PCMK__XE_LRM_RESOURCES), return);
/*
* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time, so such updates indicate an
* LRM refresh.
*
* In that case, start a new transition rather than check each result
* individually, which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
if ((controld_globals.transition_graph->pending == 0)
&& (xml->children != NULL) && (xml->children->next != NULL)) {
crm_log_xml_trace(change, "lrm-refresh");
abort_transition(INFINITY, pcmk__graph_restart, "History refresh",
NULL);
return;
}
for (rsc = pcmk__xml_first_child(xml); rsc != NULL;
rsc = pcmk__xml_next(rsc)) {
crm_trace("Processing %s", ID(rsc));
process_lrm_resource_diff(rsc, node);
}
}
static char *extract_node_uuid(const char *xpath)
{
char *mutable_path = strdup(xpath);
char *node_uuid = NULL;
char *search = NULL;
char *match = NULL;
match = strstr(mutable_path, PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "=\'");
if (match == NULL) {
free(mutable_path);
return NULL;
}
match += strlen(PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "=\'");
search = strchr(match, '\'');
if (search == NULL) {
free(mutable_path);
return NULL;
}
search[0] = 0;
node_uuid = strdup(match);
free(mutable_path);
return node_uuid;
}
static void
abort_unless_down(const char *xpath, const char *op, xmlNode *change,
const char *reason)
{
char *node_uuid = NULL;
pcmk__graph_action_t *down = NULL;
if(!pcmk__str_eq(op, "delete", pcmk__str_casei)) {
abort_transition(INFINITY, pcmk__graph_restart, reason, change);
return;
}
node_uuid = extract_node_uuid(xpath);
if(node_uuid == NULL) {
crm_err("Could not extract node ID from %s", xpath);
abort_transition(INFINITY, pcmk__graph_restart, reason, change);
return;
}
down = match_down_event(node_uuid);
if (down == NULL) {
crm_trace("Not expecting %s to be down (%s)", node_uuid, xpath);
abort_transition(INFINITY, pcmk__graph_restart, reason, change);
} else {
crm_trace("Expecting changes to %s (%s)", node_uuid, xpath);
}
free(node_uuid);
}
static void
process_op_deletion(const char *xpath, xmlNode *change)
{
char *mutable_key = strdup(xpath);
char *key;
char *node_uuid;
// Extract the part of xpath between last pair of single quotes
key = strrchr(mutable_key, '\'');
if (key != NULL) {
*key = '\0';
key = strrchr(mutable_key, '\'');
}
if (key == NULL) {
crm_warn("Ignoring malformed CIB update (resource deletion of %s)",
xpath);
free(mutable_key);
return;
}
++key;
node_uuid = extract_node_uuid(xpath);
if (confirm_cancel_action(key, node_uuid) == FALSE) {
abort_transition(INFINITY, pcmk__graph_restart,
"Resource operation removal", change);
}
free(mutable_key);
free(node_uuid);
}
static void
process_delete_diff(const char *xpath, const char *op, xmlNode *change)
{
if (strstr(xpath, "/" PCMK__XE_LRM_RSC_OP "[")) {
process_op_deletion(xpath, change);
} else if (strstr(xpath, "/" PCMK__XE_LRM "[")) {
abort_unless_down(xpath, op, change, "Resource state removal");
} else if (strstr(xpath, "/" PCMK__XE_NODE_STATE "[")) {
abort_unless_down(xpath, op, change, "Node state removal");
} else {
crm_trace("Ignoring delete of %s", xpath);
}
}
static void
process_node_state_diff(xmlNode *state, xmlNode *change, const char *op,
const char *xpath)
{
xmlNode *lrm = first_named_child(state, PCMK__XE_LRM);
process_resource_updates(ID(state), lrm, change, op, xpath);
}
static void
process_status_diff(xmlNode *status, xmlNode *change, const char *op,
const char *xpath)
{
for (xmlNode *state = pcmk__xml_first_child(status); state != NULL;
state = pcmk__xml_next(state)) {
process_node_state_diff(state, change, op, xpath);
}
}
static void
process_cib_diff(xmlNode *cib, xmlNode *change, const char *op,
const char *xpath)
{
xmlNode *status = first_named_child(cib, PCMK_XE_STATUS);
xmlNode *config = first_named_child(cib, PCMK_XE_CONFIGURATION);
if (status) {
process_status_diff(status, change, op, xpath);
}
if (config) {
abort_transition(INFINITY, pcmk__graph_restart,
"Non-status-only change", change);
}
}
static void
te_update_diff_v2(xmlNode *diff)
{
crm_log_xml_trace(diff, "Patch:Raw");
for (xmlNode *change = pcmk__xml_first_child(diff); change != NULL;
change = pcmk__xml_next(change)) {
xmlNode *match = NULL;
const char *name = NULL;
const char *xpath = crm_element_value(change, PCMK_XA_PATH);
// Possible ops: create, modify, delete, move
const char *op = crm_element_value(change, PCMK_XA_OPERATION);
// Ignore uninteresting updates
if (op == NULL) {
continue;
} else if (xpath == NULL) {
crm_trace("Ignoring %s change for version field", op);
continue;
} else if ((strcmp(op, "move") == 0)
&& (strstr(xpath,
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION
"/" PCMK_XE_RESOURCES) == NULL)) {
/* We still need to consider moves within the resources section,
* since they affect placement order.
*/
crm_trace("Ignoring move change at %s", xpath);
continue;
}
// Find the result of create/modify ops
if (strcmp(op, "create") == 0) {
match = change->children;
} else if (strcmp(op, "modify") == 0) {
match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
match = match->children;
}
} else if (!pcmk__str_any_of(op, "delete", "move", NULL)) {
crm_warn("Ignoring malformed CIB update (%s operation on %s is unrecognized)",
op, xpath);
continue;
}
if (match) {
if (match->type == XML_COMMENT_NODE) {
crm_trace("Ignoring %s operation for comment at %s", op, xpath);
continue;
}
name = (const char *)match->name;
}
crm_trace("Handling %s operation for %s%s%s",
op, (xpath? xpath : "CIB"),
(name? " matched by " : ""), (name? name : ""));
if (strstr(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION)) {
abort_transition(INFINITY, pcmk__graph_restart,
"Configuration change", change);
break; // Won't be packaged with operation results we may be waiting for
} else if (strstr(xpath, "/" XML_CIB_TAG_TICKETS)
|| pcmk__str_eq(name, XML_CIB_TAG_TICKETS, pcmk__str_none)) {
abort_transition(INFINITY, pcmk__graph_restart,
"Ticket attribute change", change);
break; // Won't be packaged with operation results we may be waiting for
- } else if (strstr(xpath, "/" XML_TAG_TRANSIENT_NODEATTRS "[")
- || pcmk__str_eq(name, XML_TAG_TRANSIENT_NODEATTRS,
+ } else if (strstr(xpath, "/" PCMK__XE_TRANSIENT_ATTRIBUTES "[")
+ || pcmk__str_eq(name, PCMK__XE_TRANSIENT_ATTRIBUTES,
pcmk__str_none)) {
abort_unless_down(xpath, op, change, "Transient attribute change");
break; // Won't be packaged with operation results we may be waiting for
} else if (strcmp(op, "delete") == 0) {
process_delete_diff(xpath, op, change);
} else if (name == NULL) {
crm_warn("Ignoring malformed CIB update (%s at %s has no result)",
op, xpath);
} else if (strcmp(name, PCMK_XE_CIB) == 0) {
process_cib_diff(match, change, op, xpath);
} else if (strcmp(name, PCMK_XE_STATUS) == 0) {
process_status_diff(match, change, op, xpath);
} else if (strcmp(name, PCMK__XE_NODE_STATE) == 0) {
process_node_state_diff(match, change, op, xpath);
} else if (strcmp(name, PCMK__XE_LRM) == 0) {
process_resource_updates(ID(match), match, change, op, xpath);
} else if (strcmp(name, PCMK__XE_LRM_RESOURCES) == 0) {
char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM);
process_resource_updates(local_node, match, change, op, xpath);
free(local_node);
} else if (strcmp(name, PCMK__XE_LRM_RESOURCE) == 0) {
char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM);
process_lrm_resource_diff(match, local_node);
free(local_node);
} else if (strcmp(name, PCMK__XE_LRM_RSC_OP) == 0) {
char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM);
process_graph_event(match, local_node);
free(local_node);
} else {
crm_warn("Ignoring malformed CIB update (%s at %s has unrecognized result %s)",
op, xpath, name);
}
}
}
void
te_update_diff(const char *event, xmlNode * msg)
{
xmlNode *diff = NULL;
const char *op = NULL;
int rc = -EINVAL;
int format = 1;
int p_add[] = { 0, 0, 0 };
int p_del[] = { 0, 0, 0 };
CRM_CHECK(msg != NULL, return);
crm_element_value_int(msg, F_CIB_RC, &rc);
if (controld_globals.transition_graph == NULL) {
crm_trace("No graph");
return;
} else if (rc < pcmk_ok) {
crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc));
return;
} else if (controld_globals.transition_graph->complete
&& (controld_globals.fsa_state != S_IDLE)
&& (controld_globals.fsa_state != S_TRANSITION_ENGINE)
&& (controld_globals.fsa_state != S_POLICY_ENGINE)) {
crm_trace("Filter state=%s (complete)",
fsa_state2string(controld_globals.fsa_state));
return;
}
op = crm_element_value(msg, F_CIB_OPERATION);
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
xml_patch_versions(diff, p_add, p_del);
crm_debug("Processing (%s) diff: %d.%d.%d -> %d.%d.%d (%s)", op,
p_del[0], p_del[1], p_del[2], p_add[0], p_add[1], p_add[2],
fsa_state2string(controld_globals.fsa_state));
crm_element_value_int(diff, PCMK_XA_FORMAT, &format);
switch (format) {
case 1:
te_update_diff_v1(event, diff);
break;
case 2:
te_update_diff_v2(diff);
break;
default:
crm_warn("Ignoring malformed CIB update (unknown patch format %d)",
format);
}
controld_remove_all_outside_events();
}
void
process_te_message(xmlNode * msg, xmlNode * xml_data)
{
const char *value = NULL;
xmlXPathObject *xpathObj = NULL;
int nmatches = 0;
CRM_CHECK(msg != NULL, return);
// Transition requests must specify transition engine as subsystem
value = crm_element_value(msg, PCMK__XA_CRM_SYS_TO);
if (pcmk__str_empty(value)
|| !pcmk__str_eq(value, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
crm_info("Received invalid transition request: subsystem '%s' not '"
CRM_SYSTEM_TENGINE "'", pcmk__s(value, ""));
return;
}
// Only the lrm_invoke command is supported as a transition request
value = crm_element_value(msg, PCMK__XA_CRM_TASK);
if (!pcmk__str_eq(value, CRM_OP_INVOKE_LRM, pcmk__str_none)) {
crm_info("Received invalid transition request: command '%s' not '"
CRM_OP_INVOKE_LRM "'", pcmk__s(value, ""));
return;
}
// Transition requests must be marked as coming from the executor
value = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM);
if (!pcmk__str_eq(value, CRM_SYSTEM_LRMD, pcmk__str_none)) {
crm_info("Received invalid transition request: from '%s' not '"
CRM_SYSTEM_LRMD "'", pcmk__s(value, ""));
return;
}
crm_debug("Processing transition request with ref='%s' origin='%s'",
pcmk__s(crm_element_value(msg, PCMK_XA_REFERENCE), ""),
pcmk__s(crm_element_value(msg, PCMK__XA_SRC), ""));
xpathObj = xpath_search(xml_data, "//" PCMK__XE_LRM_RSC_OP);
nmatches = numXpathResults(xpathObj);
if (nmatches == 0) {
crm_err("Received transition request with no results (bug?)");
} else {
for (int lpc = 0; lpc < nmatches; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
}
freeXpathObject(xpathObj);
}
void
cib_action_updated(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc < pcmk_ok) {
crm_err("Update %d FAILED: %s", call_id, pcmk_strerror(rc));
}
}
/*!
* \brief Handle a timeout in node-to-node communication
*
* \param[in,out] data Pointer to graph action
*
* \return FALSE (indicating that source should be not be re-added)
*/
gboolean
action_timer_callback(gpointer data)
{
pcmk__graph_action_t *action = (pcmk__graph_action_t *) data;
const char *task = NULL;
const char *on_node = NULL;
const char *via_node = NULL;
CRM_CHECK(data != NULL, return FALSE);
stop_te_timer(action);
task = crm_element_value(action->xml, PCMK_XA_OPERATION);
on_node = crm_element_value(action->xml, PCMK__META_ON_NODE);
via_node = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE);
if (controld_globals.transition_graph->complete) {
crm_notice("Node %s did not send %s result (via %s) within %dms "
"(ignoring because transition not in progress)",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"), action->timeout);
} else {
/* fail the action */
crm_err("Node %s did not send %s result (via %s) within %dms "
"(action timeout plus " PCMK_OPT_CLUSTER_DELAY ")",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"),
(action->timeout
+ controld_globals.transition_graph->network_delay));
pcmk__log_graph_action(LOG_ERR, action);
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
te_action_confirmed(action, controld_globals.transition_graph);
abort_transition(INFINITY, pcmk__graph_restart, "Action lost", NULL);
// Record timeout in the CIB if appropriate
if ((action->type == pcmk__rsc_graph_action)
&& controld_action_is_recordable(task)) {
controld_record_action_timeout(action);
}
}
return FALSE;
}
diff --git a/include/crm_internal.h b/include/crm_internal.h
index 31c0311504..e36ad13aca 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -1,223 +1,224 @@
/*
* 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
*/
#define PCMK__XE_ATTRIBUTES "attributes"
#define PCMK__XE_CRM_EVENT "crm_event"
#define PCMK__XE_DOWNED "downed"
#define PCMK__XE_GENERATION_TUPLE "generation_tuple"
#define PCMK__XE_LRM "lrm"
#define PCMK__XE_LRM_RESOURCE "lrm_resource"
#define PCMK__XE_LRM_RESOURCES "lrm_resources"
#define PCMK__XE_LRM_RSC_OP "lrm_rsc_op"
#define PCMK__XE_MAINTENANCE "maintenance"
#define PCMK__XE_NODE_STATE "node_state"
#define PCMK__XE_PARAM "param"
#define PCMK__XE_PING_RESPONSE "ping_response"
#define PCMK__XE_PSEUDO_EVENT "pseudo_event"
#define PCMK__XE_RSC_OP "rsc_op"
+#define PCMK__XE_TRANSIENT_ATTRIBUTES "transient_attributes"
#define PCMK__XE_TRANSITION_GRAPH "transition_graph"
// @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_CALL_ID "call-id"
#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_NODE_STATE "node_state"
#define PCMK__XA_OPERATION_KEY "operation_key"
#define PCMK__XA_OP_DIGEST "op-digest"
#define PCMK__XA_OP_FORCE_RESTART "op-force-restart"
#define PCMK__XA_OP_RESTART_DIGEST "op-restart-digest"
#define PCMK__XA_OP_SECURE_DIGEST "op-secure-digest"
#define PCMK__XA_OP_SECURE_PARAMS "op-secure-params"
#define PCMK__XA_OP_STATUS "op-status"
#define PCMK__XA_PACEMAKERD_STATE "pacemakerd_state"
#define PCMK__XA_PRIORITY "priority"
#define PCMK__XA_RC_CODE "rc-code"
#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 1.1.12
#define PCMK__XA_REF "ref"
// @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 1.1.12
#define PCMK__XA_TAG "tag"
// @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/cib/cib_attrs.c b/lib/cib/cib_attrs.c
index 95ed0f54ba..298b837713 100644
--- a/lib/cib/cib_attrs.c
+++ b/lib/cib/cib_attrs.c
@@ -1,745 +1,745 @@
/*
* 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 <sys/param.h>
#include <crm/crm.h>
#include <stdio.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <libgen.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/output_internal.h>
#include <crm/cib/internal.h>
static pcmk__output_t *
new_output_object(const char *ty)
{
int rc = pcmk_rc_ok;
pcmk__output_t *out = NULL;
const char* argv[] = { "", NULL };
pcmk__supported_format_t formats[] = {
PCMK__SUPPORTED_FORMAT_LOG,
PCMK__SUPPORTED_FORMAT_TEXT,
{ NULL, NULL, NULL }
};
pcmk__register_formats(NULL, formats);
rc = pcmk__output_new(&out, ty, NULL, (char**)argv);
if ((rc != pcmk_rc_ok) || (out == NULL)) {
crm_err("Can't out due to internal error: %s", pcmk_rc_str(rc));
return NULL;
}
return out;
}
static int
find_attr(cib_t *cib, const char *section, const char *node_uuid,
const char *attr_set_type, const char *set_name, const char *attr_id,
const char *attr_name, const char *user_name, xmlNode **result)
{
int rc = pcmk_rc_ok;
const char *xpath_base = NULL;
GString *xpath = NULL;
xmlNode *xml_search = NULL;
const char *set_type = NULL;
const char *node_type = NULL;
if (attr_set_type) {
set_type = attr_set_type;
} else {
set_type = PCMK_XE_INSTANCE_ATTRIBUTES;
}
if (pcmk__str_eq(section, PCMK_XE_CRM_CONFIG, pcmk__str_casei)) {
node_uuid = NULL;
set_type = PCMK_XE_CLUSTER_PROPERTY_SET;
} else if (pcmk__strcase_any_of(section,
PCMK_XE_OP_DEFAULTS, PCMK_XE_RSC_DEFAULTS,
NULL)) {
node_uuid = NULL;
set_type = PCMK_XE_META_ATTRIBUTES;
} else if (pcmk__str_eq(section, XML_CIB_TAG_TICKETS, pcmk__str_casei)) {
node_uuid = NULL;
section = PCMK_XE_STATUS;
node_type = XML_CIB_TAG_TICKETS;
} else if (node_uuid == NULL) {
return EINVAL;
}
xpath_base = pcmk_cib_xpath_for(section);
if (xpath_base == NULL) {
crm_warn("%s CIB section not known", section);
return ENOMSG;
}
xpath = g_string_sized_new(1024);
g_string_append(xpath, xpath_base);
if (pcmk__str_eq(node_type, XML_CIB_TAG_TICKETS, pcmk__str_casei)) {
pcmk__g_strcat(xpath, "//", node_type, NULL);
} else if (node_uuid) {
const char *node_type = PCMK_XE_NODE;
if (pcmk__str_eq(section, PCMK_XE_STATUS, pcmk__str_casei)) {
node_type = PCMK__XE_NODE_STATE;
- set_type = XML_TAG_TRANSIENT_NODEATTRS;
+ set_type = PCMK__XE_TRANSIENT_ATTRIBUTES;
}
pcmk__g_strcat(xpath,
"//", node_type, "[@" PCMK_XA_ID "='", node_uuid, "']",
NULL);
}
pcmk__g_strcat(xpath, "//", set_type, NULL);
if (set_name) {
pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "='", set_name, "']", NULL);
}
g_string_append(xpath, "//nvpair");
if (attr_id && attr_name) {
pcmk__g_strcat(xpath,
"[@" PCMK_XA_ID "='", attr_id, "' "
"and @" PCMK_XA_NAME "='", attr_name, "']", NULL);
} else if (attr_id) {
pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "='", attr_id, "']", NULL);
} else if (attr_name) {
pcmk__g_strcat(xpath, "[@" PCMK_XA_NAME "='", attr_name, "']", NULL);
}
rc = cib_internal_op(cib, PCMK__CIB_REQUEST_QUERY, NULL,
(const char *) xpath->str, NULL, &xml_search,
cib_sync_call|cib_scope_local|cib_xpath, user_name);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
crm_trace("Query failed for attribute %s (section=%s, node=%s, set=%s, xpath=%s): %s",
attr_name, section, pcmk__s(node_uuid, "<null>"),
pcmk__s(set_name, "<null>"), (const char *) xpath->str,
pcmk_rc_str(rc));
} else {
rc = pcmk_rc_ok;
crm_log_xml_debug(xml_search, "Match");
}
g_string_free(xpath, TRUE);
*result = xml_search;
return rc;
}
static int
handle_multiples(pcmk__output_t *out, xmlNode *search, const char *attr_name)
{
if ((search != NULL) && (search->children != NULL)) {
xmlNode *child = NULL;
out->info(out, "Multiple attributes match name=%s", attr_name);
for (child = pcmk__xml_first_child(search); child != NULL;
child = pcmk__xml_next(child)) {
out->info(out, " Value: %s \t(id=%s)",
crm_element_value(child, PCMK_XA_VALUE), ID(child));
}
return ENOTUNIQ;
} else {
return pcmk_rc_ok;
}
}
int
cib__update_node_attr(pcmk__output_t *out, cib_t *cib, int call_options, const char *section,
const char *node_uuid, const char *set_type, const char *set_name,
const char *attr_id, const char *attr_name, const char *attr_value,
const char *user_name, const char *node_type)
{
const char *tag = NULL;
int rc = pcmk_rc_ok;
xmlNode *xml_top = NULL;
xmlNode *xml_obj = NULL;
xmlNode *xml_search = NULL;
char *local_attr_id = NULL;
char *local_set_name = NULL;
CRM_CHECK((out != NULL) && (cib != NULL) && (section != NULL)
&& ((attr_id != NULL) || (attr_name != NULL))
&& (attr_value != NULL), return EINVAL);
rc = find_attr(cib, section, node_uuid, set_type, set_name, attr_id,
attr_name, user_name, &xml_search);
if (rc == pcmk_rc_ok) {
if (handle_multiples(out, xml_search, attr_name) == ENOTUNIQ) {
free_xml(xml_search);
return ENOTUNIQ;
} else {
pcmk__str_update(&local_attr_id,
crm_element_value(xml_search, PCMK_XA_ID));
attr_id = local_attr_id;
free_xml(xml_search);
goto do_modify;
}
} else if (rc != ENXIO) {
free_xml(xml_search);
return rc;
/* } else if(attr_id == NULL) { */
/* return EINVAL; */
} else {
free_xml(xml_search);
crm_trace("%s does not exist, create it", attr_name);
if (pcmk__str_eq(section, XML_CIB_TAG_TICKETS, pcmk__str_casei)) {
node_uuid = NULL;
section = PCMK_XE_STATUS;
node_type = XML_CIB_TAG_TICKETS;
xml_top = create_xml_node(xml_obj, PCMK_XE_STATUS);
xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS);
} else if (pcmk__str_eq(section, PCMK_XE_NODES, pcmk__str_casei)) {
if (node_uuid == NULL) {
return EINVAL;
}
if (pcmk__str_eq(node_type, "remote", pcmk__str_casei)) {
xml_top = create_xml_node(xml_obj, PCMK_XE_NODES);
xml_obj = create_xml_node(xml_top, PCMK_XE_NODE);
crm_xml_add(xml_obj, PCMK_XA_TYPE, "remote");
crm_xml_add(xml_obj, PCMK_XA_ID, node_uuid);
crm_xml_add(xml_obj, PCMK_XA_UNAME, node_uuid);
} else {
tag = PCMK_XE_NODE;
}
} else if (pcmk__str_eq(section, PCMK_XE_STATUS, pcmk__str_casei)) {
- tag = XML_TAG_TRANSIENT_NODEATTRS;
+ tag = PCMK__XE_TRANSIENT_ATTRIBUTES;
if (node_uuid == NULL) {
return EINVAL;
}
xml_top = create_xml_node(xml_obj, PCMK__XE_NODE_STATE);
crm_xml_add(xml_top, PCMK_XA_ID, node_uuid);
xml_obj = xml_top;
} else {
tag = section;
node_uuid = NULL;
}
if (set_name == NULL) {
if (pcmk__str_eq(section, PCMK_XE_CRM_CONFIG, pcmk__str_casei)) {
local_set_name = strdup(CIB_OPTIONS_FIRST);
} else if (pcmk__str_eq(node_type, XML_CIB_TAG_TICKETS, pcmk__str_casei)) {
local_set_name = crm_strdup_printf("%s-%s", section,
XML_CIB_TAG_TICKETS);
} else if (node_uuid) {
local_set_name = crm_strdup_printf("%s-%s", section, node_uuid);
if (set_type) {
char *tmp_set_name = local_set_name;
local_set_name = crm_strdup_printf("%s-%s", tmp_set_name,
set_type);
free(tmp_set_name);
}
} else {
local_set_name = crm_strdup_printf("%s-options", section);
}
set_name = local_set_name;
}
if (attr_id == NULL) {
local_attr_id = crm_strdup_printf("%s-%s", set_name, attr_name);
crm_xml_sanitize_id(local_attr_id);
attr_id = local_attr_id;
} else if (attr_name == NULL) {
attr_name = attr_id;
}
crm_trace("Creating %s/%s", section, tag);
if (tag != NULL) {
xml_obj = create_xml_node(xml_obj, tag);
crm_xml_add(xml_obj, PCMK_XA_ID, node_uuid);
if (xml_top == NULL) {
xml_top = xml_obj;
}
}
if (node_uuid == NULL && !pcmk__str_eq(node_type, XML_CIB_TAG_TICKETS, pcmk__str_casei)) {
if (pcmk__str_eq(section, PCMK_XE_CRM_CONFIG, pcmk__str_casei)) {
xml_obj = create_xml_node(xml_obj,
PCMK_XE_CLUSTER_PROPERTY_SET);
} else {
xml_obj = create_xml_node(xml_obj, PCMK_XE_META_ATTRIBUTES);
}
} else if (set_type) {
xml_obj = create_xml_node(xml_obj, set_type);
} else {
xml_obj = create_xml_node(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES);
}
crm_xml_add(xml_obj, PCMK_XA_ID, set_name);
if (xml_top == NULL) {
xml_top = xml_obj;
}
}
do_modify:
xml_obj = crm_create_nvpair_xml(xml_obj, attr_id, attr_name, attr_value);
if (xml_top == NULL) {
xml_top = xml_obj;
}
crm_log_xml_trace(xml_top, "update_attr");
rc = cib_internal_op(cib, PCMK__CIB_REQUEST_MODIFY, NULL, section, xml_top,
NULL, call_options, user_name);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
out->err(out, "Error setting %s=%s (section=%s, set=%s): %s",
attr_name, attr_value, section, pcmk__s(set_name, "<null>"),
pcmk_rc_str(rc));
crm_log_xml_info(xml_top, "Update");
} else {
rc = pcmk_rc_ok;
}
free(local_set_name);
free(local_attr_id);
free_xml(xml_top);
return rc;
}
int
cib__get_node_attrs(pcmk__output_t *out, cib_t *cib, const char *section,
const char *node_uuid, const char *set_type, const char *set_name,
const char *attr_id, const char *attr_name, const char *user_name,
xmlNode **result)
{
int rc = pcmk_rc_ok;
CRM_ASSERT(result != NULL);
CRM_CHECK(section != NULL, return EINVAL);
*result = NULL;
rc = find_attr(cib, section, node_uuid, set_type, set_name, attr_id, attr_name,
user_name, result);
if (rc != pcmk_rc_ok) {
crm_trace("Query failed for attribute %s (section=%s node=%s set=%s): %s",
pcmk__s(attr_name, "with unspecified name"),
section, pcmk__s(set_name, "<null>"),
pcmk__s(node_uuid, "<null>"), pcmk_rc_str(rc));
}
return rc;
}
int
cib__delete_node_attr(pcmk__output_t *out, cib_t *cib, int options, const char *section,
const char *node_uuid, const char *set_type, const char *set_name,
const char *attr_id, const char *attr_name, const char *attr_value,
const char *user_name)
{
int rc = pcmk_rc_ok;
xmlNode *xml_obj = NULL;
xmlNode *xml_search = NULL;
char *local_attr_id = NULL;
CRM_CHECK(section != NULL, return EINVAL);
CRM_CHECK(attr_name != NULL || attr_id != NULL, return EINVAL);
if (attr_id == NULL) {
rc = find_attr(cib, section, node_uuid, set_type, set_name, attr_id,
attr_name, user_name, &xml_search);
if (rc != pcmk_rc_ok || handle_multiples(out, xml_search, attr_name) == ENOTUNIQ) {
free_xml(xml_search);
return rc;
} else {
pcmk__str_update(&local_attr_id,
crm_element_value(xml_search, PCMK_XA_ID));
attr_id = local_attr_id;
free_xml(xml_search);
}
}
xml_obj = crm_create_nvpair_xml(NULL, attr_id, attr_name, attr_value);
rc = cib_internal_op(cib, PCMK__CIB_REQUEST_DELETE, NULL, section, xml_obj,
NULL, options, user_name);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
} else {
rc = pcmk_rc_ok;
out->info(out, "Deleted %s %s: id=%s%s%s%s%s",
section, node_uuid ? "attribute" : "option", local_attr_id,
set_name ? " set=" : "", set_name ? set_name : "",
attr_name ? " name=" : "", attr_name ? attr_name : "");
}
free(local_attr_id);
free_xml(xml_obj);
return rc;
}
int
find_nvpair_attr_delegate(cib_t *cib, const char *attr, const char *section,
const char *node_uuid, const char *attr_set_type, const char *set_name,
const char *attr_id, const char *attr_name, gboolean to_console,
char **value, const char *user_name)
{
pcmk__output_t *out = NULL;
xmlNode *xml_search = NULL;
int rc = pcmk_ok;
out = new_output_object(to_console ? "text" : "log");
if (out == NULL) {
return pcmk_err_generic;
}
rc = find_attr(cib, section, node_uuid, attr_set_type, set_name, attr_id,
attr_name, user_name, &xml_search);
if (rc == pcmk_rc_ok) {
rc = handle_multiples(out, xml_search, attr_name);
if (rc == pcmk_rc_ok) {
pcmk__str_update(value, crm_element_value(xml_search, attr));
}
}
out->finish(out, CRM_EX_OK, true, NULL);
free_xml(xml_search);
pcmk__output_free(out);
return pcmk_rc2legacy(rc);
}
int
update_attr_delegate(cib_t *cib, int call_options, const char *section,
const char *node_uuid, const char *set_type, const char *set_name,
const char *attr_id, const char *attr_name, const char *attr_value,
gboolean to_console, const char *user_name, const char *node_type)
{
pcmk__output_t *out = NULL;
int rc = pcmk_ok;
out = new_output_object(to_console ? "text" : "log");
if (out == NULL) {
return pcmk_err_generic;
}
rc = cib__update_node_attr(out, cib, call_options, section, node_uuid, set_type,
set_name, attr_id, attr_name, attr_value, user_name,
node_type);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
return pcmk_rc2legacy(rc);
}
int
read_attr_delegate(cib_t *cib, const char *section, const char *node_uuid,
const char *set_type, const char *set_name, const char *attr_id,
const char *attr_name, char **attr_value, gboolean to_console,
const char *user_name)
{
pcmk__output_t *out = NULL;
xmlNode *result = NULL;
int rc = pcmk_ok;
out = new_output_object(to_console ? "text" : "log");
if (out == NULL) {
return pcmk_err_generic;
}
rc = cib__get_node_attrs(out, cib, section, node_uuid, set_type, set_name,
attr_id, attr_name, user_name, &result);
if (rc == pcmk_rc_ok) {
if (result->children == NULL) {
pcmk__str_update(attr_value,
crm_element_value(result, PCMK_XA_VALUE));
} else {
rc = ENOTUNIQ;
}
}
out->finish(out, CRM_EX_OK, true, NULL);
free_xml(result);
pcmk__output_free(out);
return pcmk_rc2legacy(rc);
}
int
delete_attr_delegate(cib_t *cib, int options, const char *section, const char *node_uuid,
const char *set_type, const char *set_name, const char *attr_id,
const char *attr_name, const char *attr_value, gboolean to_console,
const char *user_name)
{
pcmk__output_t *out = NULL;
int rc = pcmk_ok;
out = new_output_object(to_console ? "text" : "log");
if (out == NULL) {
return pcmk_err_generic;
}
rc = cib__delete_node_attr(out, cib, options, section, node_uuid, set_type,
set_name, attr_id, attr_name, attr_value, user_name);
out->finish(out, CRM_EX_OK, true, NULL);
pcmk__output_free(out);
return pcmk_rc2legacy(rc);
}
/*!
* \internal
* \brief Parse node UUID from search result
*
* \param[in] result XML search result
* \param[out] uuid If non-NULL, where to store parsed UUID
* \param[out] is_remote If non-NULL, set TRUE if result is remote node
*
* \return pcmk_ok if UUID was successfully parsed, -ENXIO otherwise
*/
static int
get_uuid_from_result(const xmlNode *result, char **uuid, int *is_remote)
{
int rc = -ENXIO;
const char *tag;
const char *parsed_uuid = NULL;
int parsed_is_remote = FALSE;
if (result == NULL) {
return rc;
}
/* If there are multiple results, the first is sufficient */
tag = (const char *) (result->name);
if (pcmk__str_eq(tag, "xpath-query", pcmk__str_casei)) {
result = pcmk__xml_first_child(result);
CRM_CHECK(result != NULL, return rc);
tag = (const char *) (result->name);
}
if (pcmk__str_eq(tag, PCMK_XE_NODE, pcmk__str_casei)) {
/* Result is <node> tag from <nodes> section */
if (pcmk__str_eq(crm_element_value(result, PCMK_XA_TYPE), "remote",
pcmk__str_casei)) {
parsed_uuid = crm_element_value(result, PCMK_XA_UNAME);
parsed_is_remote = TRUE;
} else {
parsed_uuid = ID(result);
parsed_is_remote = FALSE;
}
} else if (pcmk__str_eq(tag, PCMK_XE_PRIMITIVE, pcmk__str_casei)) {
/* Result is <primitive> for ocf:pacemaker:remote resource */
parsed_uuid = ID(result);
parsed_is_remote = TRUE;
} else if (pcmk__str_eq(tag, PCMK_XE_NVPAIR, pcmk__str_casei)) {
/* Result is PCMK_META_REMOTE_NODE parameter of <primitive> for guest
* node
*/
parsed_uuid = crm_element_value(result, PCMK_XA_VALUE);
parsed_is_remote = TRUE;
} else if (pcmk__str_eq(tag, PCMK__XE_NODE_STATE, pcmk__str_casei)) {
// Result is PCMK__XE_NODE_STATE tag from PCMK_XE_STATUS section
parsed_uuid = crm_element_value(result, PCMK_XA_UNAME);
if (pcmk__xe_attr_is_true(result, XML_NODE_IS_REMOTE)) {
parsed_is_remote = TRUE;
}
}
if (parsed_uuid) {
if (uuid) {
*uuid = strdup(parsed_uuid);
}
if (is_remote) {
*is_remote = parsed_is_remote;
}
rc = pcmk_ok;
}
return rc;
}
/* Search string to find a node by name, as:
* - cluster or remote node in nodes section
* - remote node in resources section
* - guest node in resources section
* - orphaned remote node or bundle guest node in status section
*/
#define XPATH_UPPER_TRANS "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
#define XPATH_LOWER_TRANS "abcdefghijklmnopqrstuvwxyz"
#define XPATH_NODE \
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_NODES \
"/" PCMK_XE_NODE "[translate(@" PCMK_XA_UNAME ",'" XPATH_UPPER_TRANS "','" XPATH_LOWER_TRANS "') ='%s']" \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"/" PCMK_XE_PRIMITIVE \
"[@class='ocf'][@provider='pacemaker'][@type='remote'][translate(@id,'" XPATH_UPPER_TRANS "','" XPATH_LOWER_TRANS "') ='%s']" \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"/" PCMK_XE_PRIMITIVE "/" PCMK_XE_META_ATTRIBUTES "/" PCMK_XE_NVPAIR \
"[@name='" PCMK_META_REMOTE_NODE "'][translate(@value,'" XPATH_UPPER_TRANS "','" XPATH_LOWER_TRANS "') ='%s']" \
"|/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK__XE_NODE_STATE \
"[@" XML_NODE_IS_REMOTE "='true'][translate(@" PCMK_XA_ID ",'" XPATH_UPPER_TRANS "','" XPATH_LOWER_TRANS "') ='%s']"
int
query_node_uuid(cib_t * the_cib, const char *uname, char **uuid, int *is_remote_node)
{
int rc = pcmk_ok;
char *xpath_string;
xmlNode *xml_search = NULL;
char *host_lowercase = NULL;
CRM_ASSERT(uname != NULL);
host_lowercase = g_ascii_strdown(uname, -1);
if (uuid) {
*uuid = NULL;
}
if (is_remote_node) {
*is_remote_node = FALSE;
}
xpath_string = crm_strdup_printf(XPATH_NODE, host_lowercase, host_lowercase, host_lowercase, host_lowercase);
if (cib_internal_op(the_cib, PCMK__CIB_REQUEST_QUERY, NULL, xpath_string,
NULL, &xml_search,
cib_sync_call|cib_scope_local|cib_xpath,
NULL) == pcmk_ok) {
rc = get_uuid_from_result(xml_search, uuid, is_remote_node);
} else {
rc = -ENXIO;
}
free(xpath_string);
free_xml(xml_search);
g_free(host_lowercase);
if (rc != pcmk_ok) {
crm_debug("Could not map node name '%s' to a UUID: %s",
uname, pcmk_strerror(rc));
} else {
crm_info("Mapped node name '%s' to UUID %s", uname, (uuid? *uuid : ""));
}
return rc;
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/cib/util_compat.h>
int
query_node_uname(cib_t * the_cib, const char *uuid, char **uname)
{
int rc = pcmk_ok;
xmlNode *a_child = NULL;
xmlNode *xml_obj = NULL;
xmlNode *fragment = NULL;
const char *child_name = NULL;
CRM_ASSERT(uname != NULL);
CRM_ASSERT(uuid != NULL);
rc = the_cib->cmds->query(the_cib, PCMK_XE_NODES, &fragment,
cib_sync_call | cib_scope_local);
if (rc != pcmk_ok) {
return rc;
}
xml_obj = fragment;
CRM_CHECK(pcmk__xe_is(xml_obj, PCMK_XE_NODES), return -ENOMSG);
crm_log_xml_trace(xml_obj, "Result section");
rc = -ENXIO;
*uname = NULL;
for (a_child = pcmk__xml_first_child(xml_obj); a_child != NULL;
a_child = pcmk__xml_next(a_child)) {
if (pcmk__str_eq((const char *) a_child->name, PCMK_XE_NODE,
pcmk__str_none)) {
child_name = ID(a_child);
if (pcmk__str_eq(uuid, child_name, pcmk__str_casei)) {
child_name = crm_element_value(a_child, PCMK_XA_UNAME);
if (child_name != NULL) {
*uname = strdup(child_name);
rc = pcmk_ok;
}
break;
}
}
}
free_xml(fragment);
return rc;
}
int
set_standby(cib_t * the_cib, const char *uuid, const char *scope, const char *standby_value)
{
int rc = pcmk_ok;
char *attr_id = NULL;
CRM_CHECK(uuid != NULL, return -EINVAL);
CRM_CHECK(standby_value != NULL, return -EINVAL);
if (pcmk__strcase_any_of(scope, "reboot", PCMK_XE_STATUS, NULL)) {
scope = PCMK_XE_STATUS;
attr_id = crm_strdup_printf("transient-standby-%.256s", uuid);
} else {
scope = PCMK_XE_NODES;
attr_id = crm_strdup_printf("standby-%.256s", uuid);
}
rc = update_attr_delegate(the_cib, cib_sync_call, scope, uuid, NULL, NULL,
attr_id, "standby", standby_value, TRUE, NULL, NULL);
free(attr_id);
return rc;
}
// LCOV_EXCL_STOP
// End deprecated API
diff --git a/lib/pacemaker/pcmk_injections.c b/lib/pacemaker/pcmk_injections.c
index b01eaf00b7..1b79c98d0a 100644
--- a/lib/pacemaker/pcmk_injections.c
+++ b/lib/pacemaker/pcmk_injections.c
@@ -1,795 +1,795 @@
/*
* 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 <sys/stat.h>
#include <sys/param.h>
#include <sys/types.h>
#include <dirent.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/cib/internal.h>
#include <crm/common/util.h>
#include <crm/common/iso8601.h>
#include <crm/common/xml_internal.h>
#include <crm/lrmd_events.h> // lrmd_event_data_t, etc.
#include <crm/lrmd_internal.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
bool pcmk__simulate_node_config = false;
#define XPATH_NODE_CONFIG "//" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='%s']"
#define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
#define XPATH_NODE_STATE_BY_ID "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']"
#define XPATH_RSC_HISTORY XPATH_NODE_STATE \
"//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
/*!
* \internal
* \brief Inject a fictitious transient node attribute into scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_node \c PCMK__XE_NODE_STATE XML to inject attribute into
* \param[in] name Transient node attribute name to inject
* \param[in] value Transient node attribute value to inject
*/
static void
inject_transient_attr(pcmk__output_t *out, xmlNode *cib_node,
const char *name, const char *value)
{
xmlNode *attrs = NULL;
xmlNode *instance_attrs = NULL;
const char *node_uuid = ID(cib_node);
out->message(out, "inject-attr", name, value, cib_node);
- attrs = first_named_child(cib_node, XML_TAG_TRANSIENT_NODEATTRS);
+ attrs = first_named_child(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES);
if (attrs == NULL) {
- attrs = create_xml_node(cib_node, XML_TAG_TRANSIENT_NODEATTRS);
+ attrs = create_xml_node(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES);
crm_xml_add(attrs, PCMK_XA_ID, node_uuid);
}
instance_attrs = first_named_child(attrs, PCMK_XE_INSTANCE_ATTRIBUTES);
if (instance_attrs == NULL) {
instance_attrs = create_xml_node(attrs, PCMK_XE_INSTANCE_ATTRIBUTES);
crm_xml_add(instance_attrs, PCMK_XA_ID, node_uuid);
}
crm_create_nvpair_xml(instance_attrs, NULL, name, value);
}
/*!
* \internal
* \brief Inject a fictitious fail count into a scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_node Node state XML to inject into
* \param[in] resource ID of resource for fail count to inject
* \param[in] task Action name for fail count to inject
* \param[in] interval_ms Action interval (in milliseconds) for fail count
* \param[in] rc Action result for fail count to inject (if 0, or
* 7 when interval_ms is 0, inject nothing)
*/
void
pcmk__inject_failcount(pcmk__output_t *out, xmlNode *cib_node,
const char *resource, const char *task,
guint interval_ms, int rc)
{
if (rc == 0) {
return;
} else if ((rc == 7) && (interval_ms == 0)) {
return;
} else {
char *name = NULL;
char *now = pcmk__ttoa(time(NULL));
name = pcmk__failcount_name(resource, task, interval_ms);
inject_transient_attr(out, cib_node, name, "value++");
free(name);
name = pcmk__lastfailure_name(resource, task, interval_ms);
inject_transient_attr(out, cib_node, name, now);
free(name);
free(now);
}
}
/*!
* \internal
* \brief Create a CIB configuration entry for a fictitious node
*
* \param[in,out] cib_conn CIB object to use
* \param[in] node Node name to use
*/
static void
create_node_entry(cib_t *cib_conn, const char *node)
{
int rc = pcmk_ok;
char *xpath = crm_strdup_printf(XPATH_NODE_CONFIG, node);
rc = cib_conn->cmds->query(cib_conn, xpath, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
if (rc == -ENXIO) { // Only add if not already existing
xmlNode *cib_object = create_xml_node(NULL, PCMK_XE_NODE);
crm_xml_add(cib_object, PCMK_XA_ID, node); // Use node name as ID
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
cib_conn->cmds->create(cib_conn, PCMK_XE_NODES, cib_object,
cib_sync_call|cib_scope_local);
/* Not bothering with subsequent query to see if it exists,
we'll bomb out later in the call to query_node_uuid()... */
free_xml(cib_object);
}
free(xpath);
}
/*!
* \internal
* \brief Synthesize a fake executor event for an action
*
* \param[in] cib_resource XML for any existing resource action history
* \param[in] task Name of action to synthesize
* \param[in] interval_ms Interval of action to synthesize
* \param[in] outcome Result of action to synthesize
*
* \return Newly allocated executor event
* \note It is the caller's responsibility to free the result with
* lrmd_free_event().
*/
static lrmd_event_data_t *
create_op(const xmlNode *cib_resource, const char *task, guint interval_ms,
int outcome)
{
lrmd_event_data_t *op = NULL;
xmlNode *xop = NULL;
op = lrmd_new_event(ID(cib_resource), task, interval_ms);
lrmd__set_result(op, outcome, PCMK_EXEC_DONE, "Simulated action result");
op->params = NULL; // Not needed for simulation purposes
op->t_run = (unsigned int) time(NULL);
op->t_rcchange = op->t_run;
// Use a call ID higher than any existing history entries
op->call_id = 0;
for (xop = pcmk__xe_first_child(cib_resource); xop != NULL;
xop = pcmk__xe_next(xop)) {
int tmp = 0;
crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp);
if (tmp > op->call_id) {
op->call_id = tmp;
}
}
op->call_id++;
return op;
}
/*!
* \internal
* \brief Inject a fictitious resource history entry into a scheduler input
*
* \param[in,out] cib_resource Resource history XML to inject entry into
* \param[in,out] op Action result to inject
* \param[in] target_rc Expected result for action to inject
*
* \return XML of injected resource history entry
*/
xmlNode *
pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op,
int target_rc)
{
return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET,
target_rc, NULL, crm_system_name);
}
/*!
* \internal
* \brief Inject a fictitious node into a scheduler input
*
* \param[in,out] cib_conn Scheduler input CIB to inject node into
* \param[in] node Name of node to inject
* \param[in] uuid UUID of node to inject
*
* \return XML of \c PCMK__XE_NODE_STATE entry for new node
* \note If the global pcmk__simulate_node_config has been set to true, a
* node entry in the configuration section will be added, as well as a
* node state entry in the status section.
*/
xmlNode *
pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid)
{
int rc = pcmk_ok;
xmlNode *cib_object = NULL;
char *xpath = crm_strdup_printf(XPATH_NODE_STATE, node);
bool duplicate = false;
char *found_uuid = NULL;
if (pcmk__simulate_node_config) {
create_node_entry(cib_conn, node);
}
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath|cib_sync_call|cib_scope_local);
if ((cib_object != NULL) && (ID(cib_object) == NULL)) {
crm_err("Detected multiple " PCMK__XE_NODE_STATE " entries for "
"xpath=%s, bailing",
xpath);
duplicate = true;
goto done;
}
if (rc == -ENXIO) {
if (uuid == NULL) {
query_node_uuid(cib_conn, node, &found_uuid, NULL);
} else {
found_uuid = strdup(uuid);
}
if (found_uuid) {
char *xpath_by_uuid = crm_strdup_printf(XPATH_NODE_STATE_BY_ID,
found_uuid);
/* It's possible that a PCMK__XE_NODE_STATE entry doesn't have a
* PCMK_XA_UNAME yet
*/
rc = cib_conn->cmds->query(cib_conn, xpath_by_uuid, &cib_object,
cib_xpath|cib_sync_call|cib_scope_local);
if ((cib_object != NULL) && (ID(cib_object) == NULL)) {
crm_err("Can't inject node state for %s because multiple "
"state entries found for ID %s", node, found_uuid);
duplicate = true;
free(xpath_by_uuid);
goto done;
} else if (cib_object != NULL) {
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_STATUS,
cib_object,
cib_sync_call|cib_scope_local);
}
free(xpath_by_uuid);
}
}
if (rc == -ENXIO) {
cib_object = create_xml_node(NULL, PCMK__XE_NODE_STATE);
crm_xml_add(cib_object, PCMK_XA_ID, found_uuid);
crm_xml_add(cib_object, PCMK_XA_UNAME, node);
cib_conn->cmds->create(cib_conn, PCMK_XE_STATUS, cib_object,
cib_sync_call|cib_scope_local);
free_xml(cib_object);
rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object,
cib_xpath|cib_sync_call|cib_scope_local);
crm_trace("Injecting node state for %s (rc=%d)", node, rc);
}
done:
free(found_uuid);
free(xpath);
if (duplicate) {
crm_log_xml_warn(cib_object, "Duplicates");
crm_exit(CRM_EX_SOFTWARE);
return NULL; // not reached, but makes static analysis happy
}
CRM_ASSERT(rc == pcmk_ok);
return cib_object;
}
/*!
* \internal
* \brief Inject a fictitious node state change into a scheduler input
*
* \param[in,out] cib_conn Scheduler input CIB to inject into
* \param[in] node Name of node to inject change for
* \param[in] up If true, change state to online, otherwise offline
*
* \return XML of changed (or added) node state entry
*/
xmlNode *
pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up)
{
xmlNode *cib_node = pcmk__inject_node(cib_conn, node, NULL);
if (up) {
pcmk__xe_set_props(cib_node,
PCMK__XA_IN_CCM, PCMK_VALUE_TRUE,
PCMK__XA_CRMD, ONLINESTATUS,
PCMK__XA_JOIN, CRMD_JOINSTATE_MEMBER,
PCMK__XA_EXPECTED, CRMD_JOINSTATE_MEMBER,
NULL);
} else {
pcmk__xe_set_props(cib_node,
PCMK__XA_IN_CCM, PCMK_VALUE_FALSE,
PCMK__XA_CRMD, OFFLINESTATUS,
PCMK__XA_JOIN, CRMD_JOINSTATE_DOWN,
PCMK__XA_EXPECTED, CRMD_JOINSTATE_DOWN,
NULL);
}
crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, crm_system_name);
return cib_node;
}
/*!
* \internal
* \brief Check whether a node has history for a given resource
*
* \param[in,out] cib_node Node state XML to check
* \param[in] resource Resource name to check for
*
* \return Resource's \c PCMK__XE_LRM_RESOURCE XML entry beneath \p cib_node if
* found, otherwise \c NULL
*/
static xmlNode *
find_resource_xml(xmlNode *cib_node, const char *resource)
{
const char *node = crm_element_value(cib_node, PCMK_XA_UNAME);
char *xpath = crm_strdup_printf(XPATH_RSC_HISTORY, node, resource);
xmlNode *match = get_xpath_object(xpath, cib_node, LOG_TRACE);
free(xpath);
return match;
}
/*!
* \internal
* \brief Inject a resource history element into a scheduler input
*
* \param[in,out] out Output object for displaying error messages
* \param[in,out] cib_node Node state XML to inject resource history entry into
* \param[in] resource ID (in configuration) of resource to inject
* \param[in] lrm_name ID as used in history (could be clone instance)
* \param[in] rclass Resource agent class of resource to inject
* \param[in] rtype Resource agent type of resource to inject
* \param[in] rprovider Resource agent provider of resource to inject
*
* \return XML of injected resource history element
* \note If a history element already exists under either \p resource or
* \p lrm_name, this will return it rather than injecting a new one.
*/
xmlNode *
pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node,
const char *resource, const char *lrm_name,
const char *rclass, const char *rtype,
const char *rprovider)
{
xmlNode *lrm = NULL;
xmlNode *container = NULL;
xmlNode *cib_resource = NULL;
cib_resource = find_resource_xml(cib_node, resource);
if (cib_resource != NULL) {
/* If an existing LRM history entry uses the resource name,
* continue using it, even if lrm_name is different.
*/
return cib_resource;
}
// Check for history entry under preferred name
if (strcmp(resource, lrm_name) != 0) {
cib_resource = find_resource_xml(cib_node, lrm_name);
if (cib_resource != NULL) {
return cib_resource;
}
}
if ((rclass == NULL) || (rtype == NULL)) {
// @TODO query configuration for class, provider, type
out->err(out,
"Resource %s not found in the status section of %s "
"(supply class and type to continue)",
resource, ID(cib_node));
return NULL;
} else if (!pcmk__strcase_any_of(rclass,
PCMK_RESOURCE_CLASS_OCF,
PCMK_RESOURCE_CLASS_STONITH,
PCMK_RESOURCE_CLASS_SERVICE,
PCMK_RESOURCE_CLASS_UPSTART,
PCMK_RESOURCE_CLASS_SYSTEMD,
PCMK_RESOURCE_CLASS_LSB, NULL)) {
out->err(out, "Invalid class for %s: %s", resource, rclass);
return NULL;
} else if (pcmk_is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider)
&& (rprovider == NULL)) {
// @TODO query configuration for provider
out->err(out, "Please specify the provider for resource %s", resource);
return NULL;
}
crm_info("Injecting new resource %s into node state '%s'",
lrm_name, ID(cib_node));
lrm = first_named_child(cib_node, PCMK__XE_LRM);
if (lrm == NULL) {
const char *node_uuid = ID(cib_node);
lrm = create_xml_node(cib_node, PCMK__XE_LRM);
crm_xml_add(lrm, PCMK_XA_ID, node_uuid);
}
container = first_named_child(lrm, PCMK__XE_LRM_RESOURCES);
if (container == NULL) {
container = create_xml_node(lrm, PCMK__XE_LRM_RESOURCES);
}
cib_resource = create_xml_node(container, PCMK__XE_LRM_RESOURCE);
// If we're creating a new entry, use the preferred name
crm_xml_add(cib_resource, PCMK_XA_ID, lrm_name);
crm_xml_add(cib_resource, PCMK_XA_CLASS, rclass);
crm_xml_add(cib_resource, PCMK_XA_PROVIDER, rprovider);
crm_xml_add(cib_resource, PCMK_XA_TYPE, rtype);
return cib_resource;
}
static int
find_ticket_state(pcmk__output_t *out, cib_t *the_cib, const char *ticket_id,
xmlNode **ticket_state_xml)
{
int rc = pcmk_ok;
xmlNode *xml_search = NULL;
GString *xpath = g_string_sized_new(256);
CRM_ASSERT(ticket_state_xml != NULL);
*ticket_state_xml = NULL;
g_string_append(xpath,
"/" PCMK_XE_CIB "/" PCMK_XE_STATUS
"/" XML_CIB_TAG_TICKETS);
if (ticket_id) {
pcmk__g_strcat(xpath,
"/" XML_CIB_TAG_TICKET_STATE
"[@" PCMK_XA_ID "=\"", ticket_id, "\"]", NULL);
}
rc = the_cib->cmds->query(the_cib, (const char *) xpath->str, &xml_search,
cib_sync_call|cib_scope_local|cib_xpath);
g_string_free(xpath, TRUE);
if (rc != pcmk_ok) {
return rc;
}
crm_log_xml_debug(xml_search, "Match");
if ((xml_search->children != NULL) && (ticket_id != NULL)) {
out->err(out, "Multiple ticket_states match ticket_id=%s", ticket_id);
}
*ticket_state_xml = xml_search;
return rc;
}
/*!
* \internal
* \brief Inject a ticket attribute into ticket state
*
* \param[in,out] out Output object for displaying error messages
* \param[in] ticket_id Ticket whose state should be changed
* \param[in] attr_name Ticket attribute name to inject
* \param[in] attr_value Boolean value of ticket attribute to inject
* \param[in,out] cib CIB object to use
*
* \return Standard Pacemaker return code
*/
static int
set_ticket_state_attr(pcmk__output_t *out, const char *ticket_id,
const char *attr_name, bool attr_value, cib_t *cib)
{
int rc = pcmk_rc_ok;
xmlNode *xml_top = NULL;
xmlNode *ticket_state_xml = NULL;
// Check for an existing ticket state entry
rc = find_ticket_state(out, cib, ticket_id, &ticket_state_xml);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) { // Ticket state found, use it
crm_debug("Injecting attribute into existing ticket state %s",
ticket_id);
xml_top = ticket_state_xml;
} else if (rc == ENXIO) { // No ticket state, create it
xmlNode *xml_obj = NULL;
xml_top = create_xml_node(NULL, PCMK_XE_STATUS);
xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS);
ticket_state_xml = create_xml_node(xml_obj, XML_CIB_TAG_TICKET_STATE);
crm_xml_add(ticket_state_xml, PCMK_XA_ID, ticket_id);
} else { // Error
return rc;
}
// Add the attribute to the ticket state
pcmk__xe_set_bool_attr(ticket_state_xml, attr_name, attr_value);
crm_log_xml_debug(xml_top, "Update");
// Commit the change to the CIB
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, xml_top,
cib_sync_call|cib_scope_local);
rc = pcmk_legacy2rc(rc);
free_xml(xml_top);
return rc;
}
/*!
* \internal
* \brief Inject a fictitious action into the cluster
*
* \param[in,out] out Output object for displaying error messages
* \param[in] spec Action specification to inject
* \param[in,out] cib CIB object for scheduler input
* \param[in] scheduler Scheduler data
*/
static void
inject_action(pcmk__output_t *out, const char *spec, cib_t *cib,
const pcmk_scheduler_t *scheduler)
{
int rc;
int outcome = PCMK_OCF_OK;
guint interval_ms = 0;
char *key = NULL;
char *node = NULL;
char *task = NULL;
char *resource = NULL;
const char *rtype = NULL;
const char *rclass = NULL;
const char *rprovider = NULL;
xmlNode *cib_op = NULL;
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
const pcmk_resource_t *rsc = NULL;
lrmd_event_data_t *op = NULL;
out->message(out, "inject-spec", spec);
key = calloc(1, strlen(spec) + 1);
node = calloc(1, strlen(spec) + 1);
rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome);
if (rc != 3) {
out->err(out, "Invalid operation spec: %s. Only found %d fields",
spec, rc);
goto done;
}
parse_op_key(key, &resource, &task, &interval_ms);
rsc = pe_find_resource(scheduler->resources, resource);
if (rsc == NULL) {
out->err(out, "Invalid resource name: %s", resource);
goto done;
}
rclass = crm_element_value(rsc->xml, PCMK_XA_CLASS);
rtype = crm_element_value(rsc->xml, PCMK_XA_TYPE);
rprovider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER);
cib_node = pcmk__inject_node(cib, node, NULL);
CRM_ASSERT(cib_node != NULL);
pcmk__inject_failcount(out, cib_node, resource, task, interval_ms, outcome);
cib_resource = pcmk__inject_resource_history(out, cib_node,
resource, resource,
rclass, rtype, rprovider);
CRM_ASSERT(cib_resource != NULL);
op = create_op(cib_resource, task, interval_ms, outcome);
CRM_ASSERT(op != NULL);
cib_op = pcmk__inject_action_result(cib_resource, op, 0);
CRM_ASSERT(cib_op != NULL);
lrmd_free_event(op);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
done:
free(task);
free(node);
free(key);
}
/*!
* \internal
* \brief Inject fictitious scheduler inputs
*
* \param[in,out] scheduler Scheduler data
* \param[in,out] cib CIB object for scheduler input to modify
* \param[in] injections Injections to apply
*/
void
pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib,
const pcmk_injections_t *injections)
{
int rc = pcmk_ok;
const GList *iter = NULL;
xmlNode *cib_node = NULL;
pcmk__output_t *out = scheduler->priv;
out->message(out, "inject-modify-config", injections->quorum,
injections->watchdog);
if (injections->quorum != NULL) {
xmlNode *top = create_xml_node(NULL, PCMK_XE_CIB);
/* crm_xml_add(top, PCMK_XA_DC_UUID, dc_uuid); */
crm_xml_add(top, PCMK_XA_HAVE_QUORUM, injections->quorum);
rc = cib->cmds->modify(cib, NULL, top, cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
}
if (injections->watchdog != NULL) {
rc = cib__update_node_attr(out, cib, cib_sync_call|cib_scope_local,
PCMK_XE_CRM_CONFIG, NULL, NULL, NULL,
NULL, PCMK_OPT_HAVE_WATCHDOG,
injections->watchdog, NULL, NULL);
CRM_ASSERT(rc == pcmk_rc_ok);
}
for (iter = injections->node_up; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
out->message(out, "inject-modify-node", "Online", node);
cib_node = pcmk__inject_node_state_change(cib, node, true);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
}
for (iter = injections->node_down; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
char *xpath = NULL;
out->message(out, "inject-modify-node", "Offline", node);
cib_node = pcmk__inject_node_state_change(cib, node, false);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='%s']"
"/" PCMK__XE_LRM,
node);
cib->cmds->remove(cib, xpath, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
free(xpath);
xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='%s']"
- "/" XML_TAG_TRANSIENT_NODEATTRS,
+ "/" PCMK__XE_TRANSIENT_ATTRIBUTES,
node);
cib->cmds->remove(cib, xpath, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
free(xpath);
}
for (iter = injections->node_fail; iter != NULL; iter = iter->next) {
const char *node = (const char *) iter->data;
out->message(out, "inject-modify-node", "Failing", node);
cib_node = pcmk__inject_node_state_change(cib, node, true);
crm_xml_add(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE);
CRM_ASSERT(cib_node != NULL);
rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
free_xml(cib_node);
}
for (iter = injections->ticket_grant; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Granting", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, "granted", true, cib);
CRM_ASSERT(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_revoke; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Revoking", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, "granted", false, cib);
CRM_ASSERT(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_standby; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Standby", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, "standby", true, cib);
CRM_ASSERT(rc == pcmk_rc_ok);
}
for (iter = injections->ticket_activate; iter != NULL; iter = iter->next) {
const char *ticket_id = (const char *) iter->data;
out->message(out, "inject-modify-ticket", "Activating", ticket_id);
rc = set_ticket_state_attr(out, ticket_id, "standby", false, cib);
CRM_ASSERT(rc == pcmk_rc_ok);
}
for (iter = injections->op_inject; iter != NULL; iter = iter->next) {
inject_action(out, (const char *) iter->data, cib, scheduler);
}
if (!out->is_quiet(out)) {
out->end_list(out);
}
}
void
pcmk_free_injections(pcmk_injections_t *injections)
{
if (injections == NULL) {
return;
}
g_list_free_full(injections->node_up, g_free);
g_list_free_full(injections->node_down, g_free);
g_list_free_full(injections->node_fail, g_free);
g_list_free_full(injections->op_fail, g_free);
g_list_free_full(injections->op_inject, g_free);
g_list_free_full(injections->ticket_grant, g_free);
g_list_free_full(injections->ticket_revoke, g_free);
g_list_free_full(injections->ticket_standby, g_free);
g_list_free_full(injections->ticket_activate, g_free);
free(injections->quorum);
free(injections->watchdog);
free(injections);
}
diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c
index f1ed5c7987..83a4d4aa8d 100644
--- a/lib/pacemaker/pcmk_simulate.c
+++ b/lib/pacemaker/pcmk_simulate.c
@@ -1,1005 +1,1005 @@
/*
* Copyright 2021-2024 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 <crm/cib/internal.h>
#include <crm/common/output.h>
#include <crm/common/results.h>
#include <crm/common/scheduler.h>
#include <pacemaker-internal.h>
#include <pacemaker.h>
#include <stdint.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include "libpacemaker_private.h"
static pcmk__output_t *out = NULL;
static cib_t *fake_cib = NULL;
static GList *fake_resource_list = NULL;
static const GList *fake_op_fail_list = NULL;
static void set_effective_date(pcmk_scheduler_t *scheduler, bool print_original,
const char *use_date);
/*!
* \internal
* \brief Create an action name for use in a dot graph
*
* \param[in] action Action to create name for
* \param[in] verbose If true, add action ID to name
*
* \return Newly allocated string with action name
* \note It is the caller's responsibility to free the result.
*/
static char *
create_action_name(const pcmk_action_t *action, bool verbose)
{
char *action_name = NULL;
const char *prefix = "";
const char *action_host = NULL;
const char *clone_name = NULL;
const char *task = action->task;
if (action->node != NULL) {
action_host = action->node->details->uname;
} else if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) {
action_host = "<none>";
}
if (pcmk__str_eq(action->task, PCMK_ACTION_CANCEL, pcmk__str_none)) {
prefix = "Cancel ";
task = action->cancel_task;
}
if (action->rsc != NULL) {
clone_name = action->rsc->clone_name;
}
if (clone_name != NULL) {
char *key = NULL;
guint interval_ms = 0;
if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0,
&interval_ms) != pcmk_rc_ok) {
interval_ms = 0;
}
if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY,
PCMK_ACTION_NOTIFIED, NULL)) {
const char *n_type = g_hash_table_lookup(action->meta,
"notify_key_type");
const char *n_task = g_hash_table_lookup(action->meta,
"notify_key_operation");
CRM_ASSERT(n_type != NULL);
CRM_ASSERT(n_task != NULL);
key = pcmk__notify_key(clone_name, n_type, n_task);
} else {
key = pcmk__op_key(clone_name, task, interval_ms);
}
if (action_host != NULL) {
action_name = crm_strdup_printf("%s%s %s",
prefix, key, action_host);
} else {
action_name = crm_strdup_printf("%s%s", prefix, key);
}
free(key);
} else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH,
pcmk__str_none)) {
const char *op = g_hash_table_lookup(action->meta, "stonith_action");
action_name = crm_strdup_printf("%s%s '%s' %s",
prefix, action->task, op, action_host);
} else if (action->rsc && action_host) {
action_name = crm_strdup_printf("%s%s %s",
prefix, action->uuid, action_host);
} else if (action_host) {
action_name = crm_strdup_printf("%s%s %s",
prefix, action->task, action_host);
} else {
action_name = crm_strdup_printf("%s", action->uuid);
}
if (verbose) {
char *with_id = crm_strdup_printf("%s (%d)", action_name, action->id);
free(action_name);
action_name = with_id;
}
return action_name;
}
/*!
* \internal
* \brief Display the status of a cluster
*
* \param[in,out] scheduler Scheduler data
* \param[in] show_opts How to modify display (as pcmk_show_opt_e flags)
* \param[in] section_opts Sections to display (as pcmk_section_e flags)
* \param[in] title What to use as list title
* \param[in] print_spacer Whether to display a spacer first
*/
static void
print_cluster_status(pcmk_scheduler_t *scheduler, uint32_t show_opts,
uint32_t section_opts, const char *title,
bool print_spacer)
{
pcmk__output_t *out = scheduler->priv;
GList *all = NULL;
crm_exit_t stonith_rc = 0;
enum pcmk_pacemakerd_state state = pcmk_pacemakerd_state_invalid;
section_opts |= pcmk_section_nodes | pcmk_section_resources;
show_opts |= pcmk_show_inactive_rscs | pcmk_show_failed_detail;
all = g_list_prepend(all, (gpointer) "*");
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
out->begin_list(out, NULL, NULL, "%s", title);
out->message(out, "cluster-status",
scheduler, state, stonith_rc, NULL,
false, section_opts, show_opts, NULL, all, all);
out->end_list(out);
g_list_free(all);
}
/*!
* \internal
* \brief Display a summary of all actions scheduled in a transition
*
* \param[in,out] scheduler Scheduler data (fully scheduled)
* \param[in] print_spacer Whether to display a spacer first
*/
static void
print_transition_summary(pcmk_scheduler_t *scheduler, bool print_spacer)
{
pcmk__output_t *out = scheduler->priv;
PCMK__OUTPUT_SPACER_IF(out, print_spacer);
out->begin_list(out, NULL, NULL, "Transition Summary");
pcmk__output_actions(scheduler);
out->end_list(out);
}
/*!
* \internal
* \brief Reset scheduler input, output, date, and flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] input What to set as cluster input
* \param[in] out What to set as cluster output object
* \param[in] use_date What to set as cluster's current timestamp
* \param[in] flags Group of enum pcmk_scheduler_flags to set
*/
static void
reset(pcmk_scheduler_t *scheduler, xmlNodePtr input, pcmk__output_t *out,
const char *use_date, unsigned int flags)
{
scheduler->input = input;
scheduler->priv = out;
set_effective_date(scheduler, true, use_date);
if (pcmk_is_set(flags, pcmk_sim_sanitized)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_sanitized);
}
if (pcmk_is_set(flags, pcmk_sim_show_scores)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores);
}
if (pcmk_is_set(flags, pcmk_sim_show_utilization)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization);
}
}
/*!
* \brief Write out a file in dot(1) format describing the actions that will
* be taken by the scheduler in response to an input CIB file.
*
* \param[in,out] scheduler Scheduler data
* \param[in] dot_file The filename to write
* \param[in] all_actions Write all actions, even those that are optional
* or are on unmanaged resources
* \param[in] verbose Add extra information, such as action IDs, to the
* output
*
* \return Standard Pacemaker return code
*/
static int
write_sim_dotfile(pcmk_scheduler_t *scheduler, const char *dot_file,
bool all_actions, bool verbose)
{
GList *iter = NULL;
FILE *dot_strm = fopen(dot_file, "w");
if (dot_strm == NULL) {
return errno;
}
fprintf(dot_strm, " digraph \"g\" {\n");
for (iter = scheduler->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
const char *style = "dashed";
const char *font = "black";
const char *color = "black";
char *action_name = create_action_name(action, verbose);
if (pcmk_is_set(action->flags, pcmk_action_pseudo)) {
font = "orange";
}
if (pcmk_is_set(action->flags, pcmk_action_added_to_graph)) {
style = "bold";
color = "green";
} else if ((action->rsc != NULL)
&& !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)) {
color = "red";
font = "purple";
if (!all_actions) {
goto do_not_write;
}
} else if (pcmk_is_set(action->flags, pcmk_action_optional)) {
color = "blue";
if (!all_actions) {
goto do_not_write;
}
} else {
color = "red";
CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pcmk_action_runnable));
}
pcmk__set_action_flags(action, pcmk_action_added_to_graph);
fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n",
action_name, style, color, font);
do_not_write:
free(action_name);
}
for (iter = scheduler->actions; iter != NULL; iter = iter->next) {
pcmk_action_t *action = (pcmk_action_t *) iter->data;
for (GList *before_iter = action->actions_before;
before_iter != NULL; before_iter = before_iter->next) {
pcmk__related_action_t *before = before_iter->data;
char *before_name = NULL;
char *after_name = NULL;
const char *style = "dashed";
bool optional = true;
if (before->state == pe_link_dumped) {
optional = false;
style = "bold";
} else if ((uint32_t) before->type == pcmk__ar_none) {
continue;
} else if (pcmk_is_set(before->action->flags,
pcmk_action_added_to_graph)
&& pcmk_is_set(action->flags, pcmk_action_added_to_graph)
&& (uint32_t) before->type != pcmk__ar_if_on_same_node_or_target) {
optional = false;
}
if (all_actions || !optional) {
before_name = create_action_name(before->action, verbose);
after_name = create_action_name(action, verbose);
fprintf(dot_strm, "\"%s\" -> \"%s\" [ style = %s]\n",
before_name, after_name, style);
free(before_name);
free(after_name);
}
}
}
fprintf(dot_strm, "}\n");
fflush(dot_strm);
fclose(dot_strm);
return pcmk_rc_ok;
}
/*!
* \brief Profile the configuration updates and scheduler actions in a single
* CIB file, printing the profiling timings.
*
* \note \p scheduler->priv must have been set to a valid \p pcmk__output_t
* object before this function is called.
*
* \param[in] xml_file The CIB file to profile
* \param[in] repeat Number of times to run
* \param[in,out] scheduler Scheduler data
* \param[in] use_date The date to set the cluster's time to (may be NULL)
*/
static void
profile_file(const char *xml_file, long long repeat,
pcmk_scheduler_t *scheduler, const char *use_date)
{
pcmk__output_t *out = scheduler->priv;
xmlNode *cib_object = NULL;
clock_t start = 0;
clock_t end;
unsigned long long scheduler_flags = pcmk_sched_no_compat;
CRM_ASSERT(out != NULL);
cib_object = filename2xml(xml_file);
start = clock();
if (pcmk_find_cib_element(cib_object, PCMK_XE_STATUS) == NULL) {
create_xml_node(cib_object, PCMK_XE_STATUS);
}
if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) {
free_xml(cib_object);
return;
}
if (validate_xml(cib_object, NULL, FALSE) != TRUE) {
free_xml(cib_object);
return;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) {
scheduler_flags |= pcmk_sched_output_scores;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) {
scheduler_flags |= pcmk_sched_show_utilization;
}
for (int i = 0; i < repeat; ++i) {
xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object);
scheduler->input = input;
set_effective_date(scheduler, false, use_date);
pcmk__schedule_actions(input, scheduler_flags, scheduler);
pe_reset_working_set(scheduler);
}
end = clock();
out->message(out, "profile", xml_file, start, end);
}
void
pcmk__profile_dir(const char *dir, long long repeat,
pcmk_scheduler_t *scheduler, const char *use_date)
{
pcmk__output_t *out = scheduler->priv;
struct dirent **namelist;
int file_num = scandir(dir, &namelist, 0, alphasort);
CRM_ASSERT(out != NULL);
if (file_num > 0) {
struct stat prop;
char buffer[FILENAME_MAX];
out->begin_list(out, NULL, NULL, "Timings");
while (file_num--) {
if ('.' == namelist[file_num]->d_name[0]) {
free(namelist[file_num]);
continue;
} else if (!pcmk__ends_with_ext(namelist[file_num]->d_name,
".xml")) {
free(namelist[file_num]);
continue;
}
snprintf(buffer, sizeof(buffer), "%s/%s",
dir, namelist[file_num]->d_name);
if (stat(buffer, &prop) == 0 && S_ISREG(prop.st_mode)) {
profile_file(buffer, repeat, scheduler, use_date);
}
free(namelist[file_num]);
}
free(namelist);
out->end_list(out);
}
}
/*!
* \brief Set the date of the cluster, either to the value given by
* \p use_date, or to the "execution-date" value in the CIB.
*
* \note \p scheduler->priv must have been set to a valid \p pcmk__output_t
* object before this function is called.
*
* \param[in,out] scheduler Scheduler data
* \param[in] print_original If \p true, the "execution-date" should
* also be printed
* \param[in] use_date The date to set the cluster's time to
* (may be NULL)
*/
static void
set_effective_date(pcmk_scheduler_t *scheduler, bool print_original,
const char *use_date)
{
pcmk__output_t *out = scheduler->priv;
time_t original_date = 0;
CRM_ASSERT(out != NULL);
crm_element_value_epoch(scheduler->input, "execution-date", &original_date);
if (use_date) {
scheduler->now = crm_time_new(use_date);
out->info(out, "Setting effective cluster time: %s", use_date);
crm_time_log(LOG_NOTICE, "Pretending 'now' is", scheduler->now,
crm_time_log_date | crm_time_log_timeofday);
} else if (original_date != 0) {
scheduler->now = pcmk__copy_timet(original_date);
if (print_original) {
char *when = crm_time_as_string(scheduler->now,
crm_time_log_date|crm_time_log_timeofday);
out->info(out, "Using the original execution date of: %s", when);
free(when);
}
}
}
/*!
* \internal
* \brief Simulate successfully executing a pseudo-action in a graph
*
* \param[in,out] graph Graph to update with pseudo-action result
* \param[in,out] action Pseudo-action to simulate executing
*
* \return Standard Pacemaker return code
*/
static int
simulate_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *node = crm_element_value(action->xml, PCMK__META_ON_NODE);
const char *task = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
out->message(out, "inject-pseudo-action", node, task);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate executing a resource action in a graph
*
* \param[in,out] graph Graph to update with resource action result
* \param[in,out] action Resource action to simulate executing
*
* \return Standard Pacemaker return code
*/
static int
simulate_resource_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
int rc;
lrmd_event_data_t *op = NULL;
int target_outcome = PCMK_OCF_OK;
const char *rtype = NULL;
const char *rclass = NULL;
const char *resource = NULL;
const char *rprovider = NULL;
const char *resource_config_name = NULL;
const char *operation = crm_element_value(action->xml, PCMK_XA_OPERATION);
const char *target_rc_s = crm_meta_value(action->params,
PCMK__META_OP_TARGET_RC);
xmlNode *cib_node = NULL;
xmlNode *cib_resource = NULL;
xmlNode *action_rsc = first_named_child(action->xml, PCMK_XE_PRIMITIVE);
char *node = crm_element_value_copy(action->xml, PCMK__META_ON_NODE);
char *uuid = NULL;
const char *router_node = crm_element_value(action->xml,
PCMK__XA_ROUTER_NODE);
// Certain actions don't need to be displayed or history entries
if (pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) {
crm_debug("No history injection for %s op on %s", operation, node);
goto done; // Confirm action and update graph
}
if (action_rsc == NULL) { // Shouldn't be possible
crm_log_xml_err(action->xml, "Bad");
free(node);
return EPROTO;
}
/* A resource might be known by different names in the configuration and in
* the action (for example, a clone instance). Grab the configuration name
* (which is preferred when writing history), and if necessary, the instance
* name.
*/
resource_config_name = crm_element_value(action_rsc, PCMK_XA_ID);
if (resource_config_name == NULL) { // Shouldn't be possible
crm_log_xml_err(action->xml, "No ID");
free(node);
return EPROTO;
}
resource = resource_config_name;
if (pe_find_resource(fake_resource_list, resource) == NULL) {
const char *longname = crm_element_value(action_rsc, PCMK__XA_LONG_ID);
if ((longname != NULL)
&& (pe_find_resource(fake_resource_list, longname) != NULL)) {
resource = longname;
}
}
// Certain actions need to be displayed but don't need history entries
if (pcmk__strcase_any_of(operation, PCMK_ACTION_DELETE,
PCMK_ACTION_META_DATA, NULL)) {
out->message(out, "inject-rsc-action", resource, operation, node,
(guint) 0);
goto done; // Confirm action and update graph
}
rclass = crm_element_value(action_rsc, PCMK_XA_CLASS);
rtype = crm_element_value(action_rsc, PCMK_XA_TYPE);
rprovider = crm_element_value(action_rsc, PCMK_XA_PROVIDER);
pcmk__scan_min_int(target_rc_s, &target_outcome, 0);
CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL,
cib_sync_call|cib_scope_local) == pcmk_ok);
// Ensure the action node is in the CIB
uuid = crm_element_value_copy(action->xml, PCMK__META_ON_NODE_UUID);
cib_node = pcmk__inject_node(fake_cib, node,
((router_node == NULL)? uuid: node));
free(uuid);
CRM_ASSERT(cib_node != NULL);
// Add a history entry for the action
cib_resource = pcmk__inject_resource_history(out, cib_node, resource,
resource_config_name,
rclass, rtype, rprovider);
if (cib_resource == NULL) {
crm_err("Could not simulate action %d history for resource %s",
action->id, resource);
free(node);
free_xml(cib_node);
return EINVAL;
}
// Simulate and display an executor event for the action result
op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE,
target_outcome, "User-injected result");
out->message(out, "inject-rsc-action", resource, op->op_type, node,
op->interval_ms);
// Check whether action is in a list of desired simulated failures
for (const GList *iter = fake_op_fail_list;
iter != NULL; iter = iter->next) {
const char *spec = (const char *) iter->data;
char *key = NULL;
const char *match_name = NULL;
// Allow user to specify anonymous clone with or without instance number
key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type,
op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = resource;
}
free(key);
// If not found, try the resource's name in the configuration
if ((match_name == NULL)
&& (strcmp(resource, resource_config_name) != 0)) {
key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource_config_name,
op->op_type, op->interval_ms, node);
if (strncasecmp(key, spec, strlen(key)) == 0) {
match_name = resource_config_name;
}
free(key);
}
if (match_name == NULL) {
continue; // This failed action entry doesn't match
}
// ${match_name}_${task}_${interval_in_ms}@${node}=${rc}
rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc);
if (rc != 1) {
out->err(out, "Invalid failed operation '%s' "
"(result code must be integer)", spec);
continue; // Keep checking other list entries
}
out->info(out, "Pretending action %d failed with rc=%d",
action->id, op->rc);
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
graph->abort_priority = INFINITY;
pcmk__inject_failcount(out, cib_node, match_name, op->op_type,
op->interval_ms, op->rc);
break;
}
pcmk__inject_action_result(cib_resource, op, target_outcome);
lrmd_free_event(op);
rc = fake_cib->cmds->modify(fake_cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
done:
free(node);
free_xml(cib_node);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate successfully executing a cluster action
*
* \param[in,out] graph Graph to update with action result
* \param[in,out] action Cluster action to simulate
*
* \return Standard Pacemaker return code
*/
static int
simulate_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *node = crm_element_value(action->xml, PCMK__META_ON_NODE);
const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION);
xmlNode *rsc = first_named_child(action->xml, PCMK_XE_PRIMITIVE);
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
out->message(out, "inject-cluster-action", node, task, rsc);
pcmk__update_graph(graph, action);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Simulate successfully executing a fencing action
*
* \param[in,out] graph Graph to update with action result
* \param[in,out] action Fencing action to simulate
*
* \return Standard Pacemaker return code
*/
static int
simulate_fencing_action(pcmk__graph_t *graph, pcmk__graph_action_t *action)
{
const char *op = crm_meta_value(action->params, "stonith_action");
char *target = crm_element_value_copy(action->xml, PCMK__META_ON_NODE);
out->message(out, "inject-fencing-action", target, op);
if (!pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) {
int rc = pcmk_ok;
GString *xpath = g_string_sized_new(512);
// Set node state to offline
xmlNode *cib_node = pcmk__inject_node_state_change(fake_cib, target,
false);
CRM_ASSERT(cib_node != NULL);
crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, __func__);
rc = fake_cib->cmds->replace(fake_cib, PCMK_XE_STATUS, cib_node,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
// Simulate controller clearing node's resource history and attributes
pcmk__g_strcat(xpath,
"//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='", target, "']/" PCMK__XE_LRM,
NULL);
fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
g_string_truncate(xpath, 0);
pcmk__g_strcat(xpath,
"//" PCMK__XE_NODE_STATE
"[@" PCMK_XA_UNAME "='", target, "']"
- "/" XML_TAG_TRANSIENT_NODEATTRS, NULL);
+ "/" PCMK__XE_TRANSIENT_ATTRIBUTES, NULL);
fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL,
cib_xpath|cib_sync_call|cib_scope_local);
free_xml(cib_node);
g_string_free(xpath, TRUE);
}
pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed);
pcmk__update_graph(graph, action);
free(target);
return pcmk_rc_ok;
}
enum pcmk__graph_status
pcmk__simulate_transition(pcmk_scheduler_t *scheduler, cib_t *cib,
const GList *op_fail_list)
{
pcmk__graph_t *transition = NULL;
enum pcmk__graph_status graph_rc;
pcmk__graph_functions_t simulation_fns = {
simulate_pseudo_action,
simulate_resource_action,
simulate_cluster_action,
simulate_fencing_action,
};
out = scheduler->priv;
fake_cib = cib;
fake_op_fail_list = op_fail_list;
if (!out->is_quiet(out)) {
out->begin_list(out, NULL, NULL, "Executing Cluster Transition");
}
pcmk__set_graph_functions(&simulation_fns);
transition = pcmk__unpack_graph(scheduler->graph, crm_system_name);
pcmk__log_graph(LOG_DEBUG, transition);
fake_resource_list = scheduler->resources;
do {
graph_rc = pcmk__execute_graph(transition);
} while (graph_rc == pcmk__graph_active);
fake_resource_list = NULL;
if (graph_rc != pcmk__graph_complete) {
out->err(out, "Transition failed: %s",
pcmk__graph_status2text(graph_rc));
pcmk__log_graph(LOG_ERR, transition);
out->err(out, "An invalid transition was produced");
}
pcmk__free_graph(transition);
if (!out->is_quiet(out)) {
// If not quiet, we'll need the resulting CIB for later display
xmlNode *cib_object = NULL;
int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object,
cib_sync_call|cib_scope_local);
CRM_ASSERT(rc == pcmk_ok);
pe_reset_working_set(scheduler);
scheduler->input = cib_object;
out->end_list(out);
}
return graph_rc;
}
int
pcmk__simulate(pcmk_scheduler_t *scheduler, pcmk__output_t *out,
const pcmk_injections_t *injections, unsigned int flags,
uint32_t section_opts, const char *use_date,
const char *input_file, const char *graph_file,
const char *dot_file)
{
int printed = pcmk_rc_no_output;
int rc = pcmk_rc_ok;
xmlNodePtr input = NULL;
cib_t *cib = NULL;
rc = cib__signon_query(out, &cib, &input);
if (rc != pcmk_rc_ok) {
goto simulate_done;
}
reset(scheduler, input, out, use_date, flags);
cluster_status(scheduler);
if ((cib->variant == cib_native)
&& pcmk_is_set(section_opts, pcmk_section_times)) {
if (pcmk__our_nodename == NULL) {
// Currently used only in the times section
pcmk__query_node_name(out, 0, &pcmk__our_nodename, 0);
}
scheduler->localhost = pcmk__our_nodename;
}
if (!out->is_quiet(out)) {
const bool show_pending = pcmk_is_set(flags, pcmk_sim_show_pending);
if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) {
printed = out->message(out, "maint-mode", scheduler->flags);
}
if (scheduler->disabled_resources || scheduler->blocked_resources) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
printed = out->info(out,
"%d of %d resource instances DISABLED and "
"%d BLOCKED from further action due to failure",
scheduler->disabled_resources,
scheduler->ninstances,
scheduler->blocked_resources);
}
/* Most formatted output headers use caps for each word, but this one
* only has the first word capitalized for compatibility with pcs.
*/
print_cluster_status(scheduler, (show_pending? pcmk_show_pending : 0),
section_opts, "Current cluster status",
(printed == pcmk_rc_ok));
printed = pcmk_rc_ok;
}
// If the user requested any injections, handle them
if ((injections->node_down != NULL)
|| (injections->node_fail != NULL)
|| (injections->node_up != NULL)
|| (injections->op_inject != NULL)
|| (injections->ticket_activate != NULL)
|| (injections->ticket_grant != NULL)
|| (injections->ticket_revoke != NULL)
|| (injections->ticket_standby != NULL)
|| (injections->watchdog != NULL)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
pcmk__inject_scheduler_input(scheduler, cib, injections);
printed = pcmk_rc_ok;
rc = cib->cmds->query(cib, NULL, &input, cib_sync_call);
if (rc != pcmk_rc_ok) {
rc = pcmk_legacy2rc(rc);
goto simulate_done;
}
cleanup_calculations(scheduler);
reset(scheduler, input, out, use_date, flags);
cluster_status(scheduler);
}
if (input_file != NULL) {
rc = write_xml_file(input, input_file, FALSE);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
goto simulate_done;
}
}
if (pcmk_any_flags_set(flags, pcmk_sim_process | pcmk_sim_simulate)) {
pcmk__output_t *logger_out = NULL;
unsigned long long scheduler_flags = pcmk_sched_no_compat;
if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) {
scheduler_flags |= pcmk_sched_output_scores;
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) {
scheduler_flags |= pcmk_sched_show_utilization;
}
if (pcmk_all_flags_set(scheduler->flags,
pcmk_sched_output_scores
|pcmk_sched_show_utilization)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL,
"Assignment Scores and Utilization Information");
printed = pcmk_rc_ok;
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL, "Assignment Scores");
printed = pcmk_rc_ok;
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) {
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
out->begin_list(out, NULL, NULL, "Utilization Information");
printed = pcmk_rc_ok;
} else {
rc = pcmk__log_output_new(&logger_out);
if (rc != pcmk_rc_ok) {
goto simulate_done;
}
pe__register_messages(logger_out);
pcmk__register_lib_messages(logger_out);
scheduler->priv = logger_out;
}
pcmk__schedule_actions(input, scheduler_flags, scheduler);
if (logger_out == NULL) {
out->end_list(out);
} else {
logger_out->finish(logger_out, CRM_EX_OK, true, NULL);
pcmk__output_free(logger_out);
scheduler->priv = out;
}
input = NULL; /* Don't try and free it twice */
if (graph_file != NULL) {
rc = write_xml_file(scheduler->graph, graph_file, FALSE);
if (rc < 0) {
rc = pcmk_rc_graph_error;
goto simulate_done;
}
}
if (dot_file != NULL) {
rc = write_sim_dotfile(scheduler, dot_file,
pcmk_is_set(flags, pcmk_sim_all_actions),
pcmk_is_set(flags, pcmk_sim_verbose));
if (rc != pcmk_rc_ok) {
rc = pcmk_rc_dot_error;
goto simulate_done;
}
}
if (!out->is_quiet(out)) {
print_transition_summary(scheduler, printed == pcmk_rc_ok);
}
}
rc = pcmk_rc_ok;
if (!pcmk_is_set(flags, pcmk_sim_simulate)) {
goto simulate_done;
}
PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok);
if (pcmk__simulate_transition(scheduler, cib, injections->op_fail)
!= pcmk__graph_complete) {
rc = pcmk_rc_invalid_transition;
}
if (out->is_quiet(out)) {
goto simulate_done;
}
set_effective_date(scheduler, true, use_date);
if (pcmk_is_set(flags, pcmk_sim_show_scores)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores);
}
if (pcmk_is_set(flags, pcmk_sim_show_utilization)) {
pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization);
}
cluster_status(scheduler);
print_cluster_status(scheduler, 0, section_opts, "Revised Cluster Status",
true);
simulate_done:
cib__clean_up_connection(&cib);
return rc;
}
int
pcmk_simulate(xmlNodePtr *xml, pcmk_scheduler_t *scheduler,
const pcmk_injections_t *injections, unsigned int flags,
unsigned int section_opts, const char *use_date,
const char *input_file, const char *graph_file,
const char *dot_file)
{
pcmk__output_t *out = NULL;
int rc = pcmk_rc_ok;
rc = pcmk__xml_output_new(&out, xml);
if (rc != pcmk_rc_ok) {
return rc;
}
pe__register_messages(out);
pcmk__register_lib_messages(out);
rc = pcmk__simulate(scheduler, out, injections, flags, section_opts,
use_date, input_file, graph_file, dot_file);
pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml);
return rc;
}
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
index 4a5dc4610b..19ae050a98 100644
--- a/lib/pengine/unpack.c
+++ b/lib/pengine/unpack.c
@@ -1,5147 +1,5147 @@
/*
* 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",
pcmk__node_name(node), reason, rsc->id);
} else {
pcmk__sched_warn("Guest node %s will be fenced "
"(by recovering its guest resource %s): %s",
pcmk__node_name(node), rsc->id, reason);
/* We don't mark the node as unclean because that would prevent the
* node from running resources. We want to allow it to run resources
* in this transition if the recovery succeeds.
*/
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",
pcmk__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",
pcmk__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",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
}
node->details->unclean = TRUE;
// No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes
pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler);
} else if (node->details->unclean) {
crm_trace("Cluster node %s %s because %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean",
reason);
} else {
pcmk__sched_warn("Cluster node %s %s: %s",
pcmk__node_name(node),
pe_can_fence(scheduler, node)? "will be fenced" : "is unclean",
reason);
node->details->unclean = TRUE;
pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler);
}
}
// @TODO xpaths can't handle templates, rules, or id-refs
// nvpair with provides or requires set to unfencing
#define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \
"[(@" PCMK_XA_NAME "='" PCMK_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 \
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \
"//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \
"|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \
"/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR
static void
set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler)
{
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, PCMK_XE_CLUSTER_PROPERTY_SET, &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__flag_text(scheduler->flags, pcmk_sched_stop_all));
set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER,
pcmk_sched_symmetric_cluster);
if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) {
crm_debug("Cluster is symmetric" " - resources can run anywhere by default");
}
value = 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__flag_text(scheduler->flags, pcmk_sched_in_maintenance));
set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL,
pcmk_sched_start_failure_fatal);
if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) {
crm_trace("Start failures are always fatal");
} else {
crm_trace("Start failures are handled by failcount");
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING,
pcmk_sched_startup_fencing);
}
if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) {
crm_trace("Unseen nodes will be fenced");
} else {
pcmk__warn_once(pcmk__wo_blind,
"Blind faith: not fencing unseen nodes");
}
pe__unpack_node_health_scores(scheduler);
scheduler->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,
PCMK_XE_META_ATTRIBUTES, 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, PCMK_XE_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, PCMK_XA_SCORE);
crm_trace("Processing node %s/%s", uname, id);
if (id == NULL) {
pcmk__config_err("Ignoring <" PCMK_XE_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, PCMK_XE_PRIMITIVE,
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, PCMK_XE_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, pcmk__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, PCMK_XE_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, PCMK__XE_NODE_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",
pcmk__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);
+ attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES, FALSE);
add_node_attrs(attrs, this_node, TRUE, scheduler);
if (pe__shutdown_requested(this_node)) {
crm_info("%s is shutting down", pcmk__node_name(this_node));
this_node->details->shutdown = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) {
crm_info("%s is in standby mode", pcmk__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", pcmk__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",
pcmk__node_name(this_node));
} else {
/* This is either a remote node with fencing enabled, or a guest
* node. We don't care whether fencing is enabled when fencing guest
* nodes, because they are "fenced" by recovering their containing
* resource.
*/
crm_info("%s has resource discovery disabled",
pcmk__node_name(this_node));
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,
+ const xmlNode *attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES,
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", pcmk__node_name(node));
node->details->standby = TRUE;
}
if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) {
crm_info("%s is in maintenance mode", pcmk__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",
pcmk__node_name(node));
}
}
/*!
* \internal
* \brief Unpack a node state entry (first pass)
*
* Unpack one node state entry from status. This unpacks information from the
* \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not
* the resource history inside it. Multiple passes through the status are needed
* to fully unpack everything.
*
* \param[in] state CIB node state XML
* \param[in,out] scheduler Scheduler data
*/
static void
unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler)
{
const char *id = NULL;
const char *uname = NULL;
pcmk_node_t *this_node = NULL;
id = crm_element_value(state, PCMK_XA_ID);
if (id == NULL) {
pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without "
PCMK_XA_ID);
crm_log_xml_info(state, "missing-id");
return;
}
uname = 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 PCMK__XE_NODE_STATE entry doesn't have
* a PCMK_XA_UNAME yet. We should recognize the node as `pending` and
* wait for it to join CPG.
*/
crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" "
"without " PCMK_XA_UNAME,
id);
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
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)",
pcmk__node_name(this_node), id);
determine_online_status(state, this_node, scheduler);
if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)
&& this_node->details->online
&& (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) {
/* Everything else should flow from this automatically
* (at least until the scheduler becomes able to migrate off
* healthy resources)
*/
pe_fence_node(scheduler, this_node, "cluster does not have quorum",
FALSE);
}
}
/*!
* \internal
* \brief Unpack nodes' resource history as much as possible
*
* Unpack as many nodes' resource history as possible in one pass through the
* status. We need to process Pacemaker Remote nodes' connections/containers
* before unpacking their history; the connection/container history will be
* in another node's history, so it might take multiple passes to unpack
* everything.
*
* \param[in] status CIB XML status section
* \param[in] fence If true, treat any not-yet-unpacked nodes as unseen
* \param[in,out] scheduler Scheduler data
*
* \return Standard Pacemaker return code (specifically pcmk_rc_ok if done,
* or EAGAIN if more unpacking remains to be done)
*/
static int
unpack_node_history(const xmlNode *status, bool fence,
pcmk_scheduler_t *scheduler)
{
int rc = pcmk_rc_ok;
// Loop through all PCMK__XE_NODE_STATE entries in CIB status
for (const xmlNode *state = first_named_child(status, PCMK__XE_NODE_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 "
PCMK__XE_NODE_STATE " without id and/or uname");
continue;
}
this_node = pe_find_node_any(scheduler->nodes, id, uname);
if (this_node == NULL) {
// Warning already logged in first pass through status section
crm_trace("Not unpacking resource history for node %s because "
"no longer in configuration", id);
continue;
}
if (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, PCMK__XE_NODE_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 = pcmk__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 \c PCMK__XE_NODE_STATE entry
* \param[in,out] scheduler Scheduler data
*
* \return Epoch time when node became a cluster member
* (or scheduler effective time for legacy entries) if a member,
* 0 if not a member, or -1 if no valid information available
*/
static long long
unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler)
{
const char *member_time = 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 " PCMK__XE_NODE_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 \c PCMK__XE_NODE_STATE entry
*
* \return Epoch time when node became online in process group (or 0 if not
* online, or 1 for legacy online entries)
*/
static long long
unpack_node_online(const xmlNode *node_state)
{
const char *peer_time = 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 "
PCMK__XE_NODE_STATE " entry, assuming offline", peer_time);
return 0LL;
}
return when_online;
}
}
/*!
* \internal
* \brief Unpack node attribute for user-requested fencing
*
* \param[in] node Node to check
* \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status
*
* \return \c true if fencing has been requested for \p node, otherwise \c false
*/
static bool
unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state)
{
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, pcmk__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", pcmk__node_name(this_node),
((when_member < 0)? "presumed " : ""));
} else if (when_online > 0) {
if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) {
online = TRUE;
} else {
crm_debug("Node %s is not ready to run resources: %s",
pcmk__node_name(this_node), join);
}
} else if (this_node->details->expected_up == FALSE) {
crm_trace("Node %s controller is down: "
"member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
} else {
/* mark it unclean */
pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE);
crm_info("Node %s member@%lld online@%lld join=%s expected=%s",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"));
}
return online;
}
/*!
* \internal
* \brief Check whether a node has taken too long to join controller group
*
* \param[in,out] scheduler Scheduler data
* \param[in] node Node to check
* \param[in] when_member Epoch time when node became a cluster member
* \param[in] when_online Epoch time when node joined controller group
*
* \return true if node has been pending (on the way up) longer than
* \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false
* \note This will also update the cluster's recheck time if appropriate.
*/
static inline bool
pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node,
long long when_member, long long when_online)
{
if ((scheduler->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",
pcmk__node_name(this_node), when_member, when_online,
pcmk__s(join, "<null>"), pcmk__s(exp_state, "<null>"),
(termination_requested? " (termination requested)" : ""));
if (this_node->details->shutdown) {
crm_debug("%s is shutting down", pcmk__node_name(this_node));
/* Slightly different criteria since we can't shut down a dead peer */
return (when_online > 0);
}
if (when_member < 0) {
pe_fence_node(scheduler, this_node,
"peer has not been seen by the cluster", FALSE);
return false;
}
if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) {
pe_fence_node(scheduler, this_node,
"peer failed Pacemaker membership criteria", FALSE);
} else if (termination_requested) {
if ((when_member <= 0) && (when_online <= 0)
&& pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) {
crm_info("%s was fenced as requested", pcmk__node_name(this_node));
return false;
}
pe_fence_node(scheduler, this_node, "fencing was requested", false);
} else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN,
pcmk__str_null_matches)) {
if (pending_too_long(scheduler, this_node, when_member, when_online)) {
pe_fence_node(scheduler, this_node,
"peer pending timed out on joining the process group",
FALSE);
} else if ((when_member > 0) || (when_online > 0)) {
crm_info("- %s is not ready to run resources",
pcmk__node_name(this_node));
this_node->details->standby = TRUE;
this_node->details->pending = TRUE;
} else {
crm_trace("%s is down or still coming up",
pcmk__node_name(this_node));
}
} else if (when_member <= 0) {
// Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes
pe_fence_node(scheduler, this_node,
"peer is no longer part of the cluster", TRUE);
} else if (when_online <= 0) {
pe_fence_node(scheduler, this_node,
"peer process is no longer available", FALSE);
/* Everything is running at this point, now check join state */
} else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) {
crm_info("%s is active", pcmk__node_name(this_node));
} else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING,
CRMD_JOINSTATE_DOWN, NULL)) {
crm_info("%s is not ready to run resources",
pcmk__node_name(this_node));
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", pcmk__node_name(this_node));
} else if (this_node->details->unclean) {
pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node));
} else if (this_node->details->online) {
crm_info("%s is %s", pcmk__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", pcmk__node_name(this_node));
}
}
/*!
* \internal
* \brief Find the end of a resource's name, excluding any clone suffix
*
* \param[in] id Resource ID to check
*
* \return Pointer to last character of resource's base name
*/
const char *
pe_base_name_end(const char *id)
{
if (!pcmk__str_empty(id)) {
const char *end = id + strlen(id) - 1;
for (const char *s = end; s > id; --s) {
switch (*s) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case ':':
return (s == end)? s : (s - 1);
default:
return end;
}
}
return end;
}
return NULL;
}
/*!
* \internal
* \brief Get a resource name excluding any clone suffix
*
* \param[in] last_rsc_id Resource ID to check
*
* \return Pointer to newly allocated string with resource's base name
* \note It is the caller's responsibility to free() the result.
* This asserts on error, so callers can assume result is not NULL.
*/
char *
clone_strip(const char *last_rsc_id)
{
const char *end = pe_base_name_end(last_rsc_id);
char *basename = NULL;
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, PCMK_XE_PRIMITIVE);
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, pcmk__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, pcmk__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__same_node((pcmk_node_t *) locations->data, node)) {
/* This child instance is active on the requested node, so check
* for a corresponding configured resource. We use find_rsc()
* instead of child because child may be a cloned group, and we
* need the particular member corresponding to rsc_id.
*
* If the history entry 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, pcmk__node_name(node));
skip_inactive = TRUE;
rsc = NULL;
} else {
pcmk__rsc_trace(parent, "Resource %s, active", rsc->id);
}
}
}
g_list_free(locations);
} else {
pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id);
if (!skip_inactive && !inactive_instance
&& !pcmk_is_set(child->flags, pcmk_rsc_blocked)) {
// Remember one inactive instance in case we don't find active
inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL,
pcmk_rsc_match_clone_only);
/* ... but don't use it if it was already associated with a
* pending action on another node
*/
if ((inactive_instance != NULL) &&
(inactive_instance->pending_node != NULL) &&
!pcmk__same_node(inactive_instance->pending_node, node)) {
inactive_instance = NULL;
}
}
}
}
if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) {
pcmk__rsc_trace(parent, "Resource %s, empty slot",
inactive_instance->id);
rsc = inactive_instance;
}
/* If the resource has "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, pcmk__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, pcmk__node_name(node));
rsc = create_fake_resource(rsc_id, rsc_entry, scheduler);
if (rsc == NULL) {
return NULL;
}
if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed);
} else {
CRM_CHECK(rsc != NULL, return NULL);
pcmk__rsc_trace(rsc, "Added 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, pcmk_role_text(rsc->role), pcmk__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),
pcmk__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, pcmk__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, pcmk__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, pcmk__node_name(node));
break;
/* Need to check if there's a monitor for role="Stopped" */
} else if (start_index < stop_index && counter <= stop_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active",
id, pcmk__node_name(node));
continue;
} else if (counter < start_index) {
pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d",
id, pcmk__node_name(node), counter);
continue;
}
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, pcmk__node_name(node));
continue;
}
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, pcmk__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, pcmk__node_name(node));
custom_action(rsc, key, task, node, TRUE, scheduler);
}
}
void
calculate_active_ops(const GList *sorted_op_list, int *start_index,
int *stop_index)
{
int counter = -1;
int implied_monitor_start = -1;
int implied_clone_start = -1;
const char *task = NULL;
const char *status = NULL;
*stop_index = -1;
*start_index = -1;
for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) {
const xmlNode *rsc_op = (const xmlNode *) iter->data;
counter++;
task = crm_element_value(rsc_op, PCMK_XA_OPERATION);
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, PCMK__XA_RC_CODE);
if (pcmk__strcase_any_of(rc, "0", "8", NULL)) {
implied_monitor_start = counter;
}
} else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE,
PCMK_ACTION_DEMOTE, NULL)) {
implied_clone_start = counter;
}
}
if (*start_index == -1) {
if (implied_clone_start != -1) {
*start_index = implied_clone_start;
} else if (implied_monitor_start != -1) {
*start_index = implied_monitor_start;
}
}
}
// If resource history entry has shutdown lock, remember lock node and time
static void
unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc,
const pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
time_t lock_time = 0; // When lock started (i.e. node shutdown time)
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, pcmk__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 \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status
*
* \param[in,out] node Node whose status is being unpacked
* \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked
* \param[in,out] scheduler Scheduler data
*
* \return Resource corresponding to the entry, or NULL if no operation history
*/
static pcmk_resource_t *
unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource,
pcmk_scheduler_t *scheduler)
{
GList *gIter = NULL;
int stop_index = -1;
int start_index = -1;
enum rsc_role_e req_role = pcmk_role_unknown;
const char *rsc_id = 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 " PCMK__XE_LRM_RESOURCE
" entry: No " PCMK_XA_ID);
crm_log_xml_info(lrm_resource, "missing-id");
return NULL;
}
crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s",
rsc_id, pcmk__node_name(node));
/* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort
* them
*/
for (rsc_op = first_named_child(lrm_resource, PCMK__XE_LRM_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, pcmk_role_text(rsc->next_role),
pcmk_role_text(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, PCMK__XE_LRM_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 PCMK__XE_LRM_RESOURCES section
xml = find_xml_node(xml, PCMK__XE_LRM, FALSE);
if (xml == NULL) {
return;
}
xml = find_xml_node(xml, PCMK__XE_LRM_RESOURCES, FALSE);
if (xml == NULL) {
return;
}
// Unpack each PCMK__XE_LRM_RESOURCE entry
for (const xmlNode *rsc_entry = first_named_child(xml,
PCMK__XE_LRM_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 "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \
"/" PCMK__XE_NODE_STATE
#define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \
"/" PCMK__XE_LRM_RESOURCES \
"/" PCMK__XE_LRM_RESOURCE
#define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP
static xmlNode *
find_lrm_op(const char *resource, const char *op, const char *node, const char *source,
int target_rc, pcmk_scheduler_t *scheduler)
{
GString *xpath = NULL;
xmlNode *xml = NULL;
CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL),
return NULL);
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']"
SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'",
NULL);
/* Need to check against transition_magic too? */
if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_TARGET "='", source, "']",
NULL);
} else if ((source != NULL)
&& (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) {
pcmk__g_strcat(xpath,
" and @" PCMK__META_MIGRATE_SOURCE "='", source, "']",
NULL);
} else {
g_string_append_c(xpath, ']');
}
xml = 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, PCMK__XA_RC_CODE, &rc);
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;
char *xpath = NULL;
xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"
SUB_XPATH_LRM_RSC_OP
"[@" PCMK__XA_RC_CODE "!='%d']",
node_name, rsc->id, PCMK_OCF_UNKNOWN);
search = xpath_search(rsc->cluster->input, xpath);
result = (numXpathResults(search) == 0);
freeXpathObject(search);
free(xpath);
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, PCMK__XE_LRM_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, PCMK__META_MIGRATE_SOURCE);
target = crm_element_value(xml_op, PCMK__META_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, PCMK__META_MIGRATE_SOURCE);
*target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET);
if ((*source_name == NULL) || (*target_name == NULL)) {
pcmk__config_err("Ignoring resource history entry %s without "
PCMK__META_MIGRATE_SOURCE " and "
PCMK__META_MIGRATE_TARGET, 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 "
PCMK__META_MIGRATE_SOURCE "='%s' does not match %s",
ID(entry), *source_name, pcmk__node_name(source_node));
return pcmk_rc_unpack_error;
}
if ((target_node != NULL)
&& !pcmk__str_eq(*target_name, target_node->details->uname,
pcmk__str_casei|pcmk__str_null_matches)) {
pcmk__config_err("Ignoring resource history entry %s because "
PCMK__META_MIGRATE_TARGET "='%s' does not match %s",
ID(entry), *target_name, pcmk__node_name(target_node));
return pcmk_rc_unpack_error;
}
return pcmk_rc_ok;
}
/*
* \internal
* \brief Add a migration source to a resource's list of dangling migrations
*
* If the migrate_to and migrate_from actions in a live migration both
* succeeded, but there is no stop on the source, the migration is considered
* "dangling." Add the source to the resource's dangling migration list, which
* will be used to schedule a stop on the source without affecting the target.
*
* \param[in,out] rsc Resource involved in migration
* \param[in] node Migration source
*/
static void
add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node)
{
pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s",
rsc->id, pcmk__node_name(node));
rsc->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, PCMK__XA_RC_CODE, &from_rc);
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 = pcmk__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, pcmk__node_name(history->node));
return;
}
}
crm_trace("Adding entry for %s on %s to failed action list",
history->key, pcmk__node_name(history->node));
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, PCMK_XA_LAST_RC_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,
pcmk__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,
pcmk__node_name(history->node), last_change_s,
history->exit_status, history->id);
if (is_probe && (history->exit_status != PCMK_OCF_OK)
&& (history->exit_status != PCMK_OCF_NOT_RUNNING)
&& (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) {
/* A failed (not just unexpected) probe result could mean the user
* didn't know resources will be probed even where they can't run.
*/
crm_notice("If it is not possible for %s to run on %s, see "
"the " PCMK_XA_RESOURCE_DISCOVERY " option for location "
"constraints",
history->rsc->id, pcmk__node_name(history->node));
}
record_failed_op(history);
}
free(last_change_s);
if (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, pcmk_role_text(history->rsc->role),
pcmk__btoa(history->node->details->unclean),
fail2text(config_on_fail), pcmk_role_text(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,
pcmk__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, pcmk__node_name(history->node));
} else if (history->exit_status == history->expected_exit_status) {
remap_because(history, &why, PCMK_EXEC_DONE, "expected result");
} else {
remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result");
pcmk__rsc_debug(history->rsc,
"%s on %s: expected %d (%s), got %d (%s%s%s)",
history->key, pcmk__node_name(history->node),
history->expected_exit_status,
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, pcmk__node_name(history->node),
last_change_s);
free(last_change_s);
}
break;
case PCMK_OCF_NOT_RUNNING:
if (is_probe
|| (history->expected_exit_status == history->exit_status)
|| !pcmk_is_set(history->rsc->flags, pcmk_rsc_managed)) {
/* For probes, recurring monitors for the Stopped role, and
* unmanaged resources, "not running" is not considered a
* failure.
*/
remap_because(history, &why, PCMK_EXEC_DONE, "exit status");
history->rsc->role = pcmk_role_stopped;
*on_fail = pcmk_on_fail_ignore;
pe__set_next_role(history->rsc, pcmk_role_unknown,
"not running");
}
break;
case PCMK_OCF_RUNNING_PROMOTED:
if (is_probe
&& (history->exit_status != history->expected_exit_status)) {
char *last_change_s = last_change_str(history->xml);
remap_because(history, &why, PCMK_EXEC_DONE, "probe");
pcmk__rsc_info(history->rsc,
"Probe found %s active and promoted on %s at %s",
history->rsc->id,
pcmk__node_name(history->node), last_change_s);
free(last_change_s);
}
if (!expired
|| (history->exit_status == history->expected_exit_status)) {
history->rsc->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,
pcmk__node_name(history->node), last_change_s);
remap_because(history, &why, PCMK_EXEC_ERROR,
"unknown exit status");
free(last_change_s);
}
break;
}
remap_done:
if (why != NULL) {
pcmk__rsc_trace(history->rsc,
"Remapped %s result from [%s: %s] to [%s: %s] "
"because of %s",
history->key, pcmk_exec_status_str(orig_exec_status),
crm_exit_str(orig_exit_status),
pcmk_exec_status_str(history->execution_status),
crm_exit_str(history->exit_status), why);
}
}
// return TRUE if start or monitor last failure but parameters changed
static bool
should_clear_for_param_change(const xmlNode *xml_op, const char *task,
pcmk_resource_t *rsc, pcmk_node_t *node)
{
if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) {
if (pe__bundle_needs_remote_name(rsc)) {
/* We haven't allocated resources yet, so we can't reliably
* substitute addr parameters for the REMOTE_CONTAINER_HACK.
* When that's needed, defer the check until later.
*/
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, pcmk__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, pcmk__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, PCMK_XA_LAST_RC_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, pcmk__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, pcmk__node_name(history->node));
expired = false;
break;
}
}
return expired;
}
int
pe__target_rc_from_xml(const xmlNode *xml_op)
{
int target_rc = 0;
const char *key = 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, pcmk__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, pcmk__node_name(history->node));
set_active(history->rsc);
}
if (!clear_past_failure) {
return;
}
switch (*on_fail) {
case pcmk_on_fail_stop:
case pcmk_on_fail_ban:
case pcmk_on_fail_standby_node:
case pcmk_on_fail_fence_node:
pcmk__rsc_trace(history->rsc,
"%s (%s) is not cleared by a completed %s",
history->rsc->id, 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, 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 " PCMK__XA_OP_STATUS " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_OP_STATUS),
""));
return pcmk_rc_unpack_error;
}
if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE,
&(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 " PCMK__XA_RC_CODE " '%s'",
history->id, history->rsc->id,
pcmk__node_name(history->node),
pcmk__s(crm_element_value(history->xml,
PCMK__XA_RC_CODE),
""));
return pcmk_rc_unpack_error;
#else
history->exit_status = CRM_EX_ERROR;
#endif
}
history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Process an action history entry whose result expired
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
*
* \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the
* entry needs no further processing)
*/
static int
process_expired_result(struct action_history *history, int orig_exit_status)
{
if (!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,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->exit_status == history->expected_exit_status) {
return pcmk_rc_undetermined; // Only failures expire
}
if (history->interval_ms == 0) {
crm_notice("Ignoring resource history entry %s for %s of %s on %s: "
"Expired failure",
history->id, history->task, history->rsc->id,
pcmk__node_name(history->node));
return pcmk_rc_ok;
}
if (history->node->details->online && !history->node->details->unclean) {
/* Reschedule the recurring action. schedule_cancel() won't work at
* this stage, so as a hacky workaround, forcibly change the restart
* digest so pcmk__check_action_config() does what we want later.
*
* @TODO We should skip this if there is a newer successful monitor.
* Also, this causes rescheduling only if the history entry
* has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure
* scheduler regression test doesn't, but that may not be a
* realistic scenario in production).
*/
crm_notice("Rescheduling %s-interval %s of %s on %s "
"after failure expired",
pcmk__readable_interval(history->interval_ms), history->task,
history->rsc->id, pcmk__node_name(history->node));
crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST,
"calculated-failure-timeout");
return pcmk_rc_ok;
}
return pcmk_rc_undetermined;
}
/*!
* \internal
* \brief Process a masked probe failure
*
* \param[in,out] history Parsed action history entry
* \param[in] orig_exit_status Action exit status before remapping
* \param[in] last_failure Resource's last_failure entry, if known
* \param[in,out] on_fail Resource's current failure handling
*/
static void
mask_probe_failure(struct action_history *history, int orig_exit_status,
const xmlNode *last_failure,
enum 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,
pcmk__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
* \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be
* the same.
*/
static bool
failure_is_newer(const struct action_history *history,
const xmlNode *last_failure)
{
guint failure_interval_ms = 0U;
long long failure_change = 0LL;
long long this_change = 0LL;
if (last_failure == NULL) {
return false; // Resource has no last_failure entry
}
if (!pcmk__str_eq(history->task,
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, PCMK_XA_LAST_RC_CHANGE),
&this_change, 0LL) != pcmk_rc_ok)
|| (pcmk__scan_ll(crm_element_value(last_failure,
PCMK_XA_LAST_RC_CHANGE),
&failure_change, 0LL) != pcmk_rc_ok)
|| (failure_change < this_change)) {
return false; // Failure is not known to be newer
}
return true;
}
/*!
* \internal
* \brief Update a resource's role etc. for a pending action
*
* \param[in,out] history Parsed history entry for pending action
* \param[in] last_failure Resource's last_failure entry, if known
*/
static void
process_pending_action(struct action_history *history,
const xmlNode *last_failure)
{
/* For recurring monitors, a failure is recorded only in RSC_last_failure_0,
* and there might be a RSC_monitor_INTERVAL entry with the last successful
* or pending result.
*
* If last_failure contains the failure of the pending recurring monitor
* we're processing here, and is newer, the action is no longer pending.
* (Pending results have call ID -1, which sorts last, so the last failure
* if any should be known.)
*/
if (failure_is_newer(history, last_failure)) {
return;
}
if (strcmp(history->task, PCMK_ACTION_START) == 0) {
pcmk__set_rsc_flags(history->rsc, pcmk_rsc_start_pending);
set_active(history->rsc);
} else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) {
history->rsc->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,
PCMK__META_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, pcmk__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, pcmk__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, pcmk__node_name(node),
history.task);
return;
}
if (unpack_action_result(&history) != pcmk_rc_ok) {
return; // Error already logged
}
history.expected_exit_status = pe__target_rc_from_xml(xml_op);
history.key = pcmk__xe_history_key(xml_op);
crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id));
pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)",
history.id, history.task, history.call_id,
pcmk__node_name(node),
pcmk_exec_status_str(history.execution_status),
crm_exit_str(history.exit_status));
if (node->details->unclean) {
pcmk__rsc_trace(rsc,
"%s is running on %s, which is unclean (further action "
"depends on value of stop's on-fail attribute)",
rsc->id, pcmk__node_name(node));
}
expired = check_operation_expiry(&history);
old_rc = history.exit_status;
remap_operation(&history, on_fail, expired);
if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) {
goto done;
}
if (!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, pcmk__node_name(node),
history.execution_status, history.exit_status,
history.id);
/* Also for printing it as "FAILED" by marking it as
* pcmk_rsc_failed later
*/
*on_fail = pcmk_on_fail_ban;
}
resource_location(parent, node, -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,
pcmk__node_name(node), last_change_s, history.id);
free(last_change_s);
update_resource_state(&history, history.expected_exit_status,
*last_failure, on_fail);
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, pcmk__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, pcmk__node_name(node), history.id,
pcmk_role_text(rsc->role),
pcmk_role_text(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, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data,
node->details->attrs, NULL, overwrite,
scheduler);
pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_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, PCMK__XE_LRM_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, PCMK_XE_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, PCMK__XE_NODE_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, PCMK__XE_LRM, FALSE);
tmp = find_xml_node(tmp, PCMK__XE_LRM_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,
PCMK__XE_LRM_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;
}