static enum crmd_fsa_input
handle_failcount_op(xmlNode * stored_msg)
{
const char *rsc = NULL;
xmlNode *xml_rsc = get_xpath_object("//" XML_CIB_TAG_RESOURCE, stored_msg, LOG_ERR);
if (xml_rsc) {
rsc = ID(xml_rsc);
}
if (rsc) {
char *attr = NULL;
crm_info("Removing failcount for %s", rsc);
attr = crm_concat("fail-count", rsc, '-');
update_attrd(NULL, attr, NULL, NULL);
free(attr);
attr = crm_concat("last-failure", rsc, '-');
update_attrd(NULL, attr, NULL, NULL);
free(attr);
lrm_clear_last_failure(rsc);
} else {
crm_log_xml_warn(stored_msg, "invalid failcount op");
}
return I_NULL;
}
enum crmd_fsa_input
handle_shutdown_request(xmlNode * stored_msg)
{
/* handle here to avoid potential version issues
* where the shutdown message/proceedure may have
* been changed in later versions.
*
* This way the DC is always in control of the shutdown
*/
char *now_s = NULL;
time_t now = time(NULL);
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
if (host_from == NULL) {
/* we're shutting down and the DC */
host_from = fsa_our_uname;
}
crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state));
crm_log_xml_trace(stored_msg, "message");
now_s = crm_itoa(now);
update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, FALSE);
free(now_s);
/* will be picked up by the TE as long as its running */
return I_NULL;
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node joining
*
* \param[in] node_name Name of newly integrated pacemaker_remote node
*/
static void
remote_node_up(const char *node_name)
{
int call_opt, call_id = 0;
xmlNode *update, *state;
crm_node_t *node;
CRM_CHECK(node_name != NULL, return);
crm_info("Announcing pacemaker_remote node %s", node_name);
/* Clear node's operation history. The node's transient attributes should
* and normally will be cleared when the node leaves, but since remote node
* state has a number of corner cases, clear them here as well, to be sure.
*/
call_opt = crmd_cib_smart_opt();
erase_status_tag(node_name, XML_CIB_TAG_LRM, call_opt);
erase_status_tag(node_name, XML_TAG_TRANSIENT_NODEATTRS, call_opt);
/* Clear node's probed attribute */
update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE);
/* Ensure node is in the remote peer cache with member status */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0);
/* pacemaker_remote nodes don't participate in the membership layer,
* so cluster nodes don't automatically get notified when they come and go.
* We send a cluster message to the DC, and update the CIB node state entry,
* so the DC will get it sooner (via message) or later (via CIB refresh),
* and any other interested parties can query the CIB.
*/
send_remote_state_message(node_name, TRUE);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_cluster, update,
__FUNCTION__);
/* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever
* needs to be fenced, this flag will allow various actions to determine
* whether the fencing has happened yet.
*/
crm_xml_add(state, XML_NODE_IS_FENCED, "0");
/* TODO: If the remote connection drops, and this (async) CIB update either
* failed or has not yet completed, later actions could mistakenly think the
* node has already been fenced (if the XML_NODE_IS_FENCED attribute was
* previously set, because it won't have been cleared). This could prevent
* actual fencing or allow recurring monitor failures to be cleared too
* soon. Ideally, we wouldn't rely on the CIB for the fenced status.
*/
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_WARNING, "%s CIB node state setup", node_name);
}
free_xml(update);
}
void
do_dc_join_final(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)
{
crm_debug("Ensuring DC, quorum and node attributes are up-to-date");
update_attrd(NULL, NULL, NULL, NULL, FALSE);
crm_update_quorum(crm_have_quorum, TRUE);
}
static enum crmd_fsa_input
handle_failcount_op(xmlNode * stored_msg)
{
const char *rsc = NULL;
const char *uname = NULL;
gboolean is_remote_node = FALSE;
xmlNode *xml_rsc = get_xpath_object("//" XML_CIB_TAG_RESOURCE, stored_msg, LOG_ERR);
if (xml_rsc) {
rsc = ID(xml_rsc);
}
uname = crm_element_value(stored_msg, XML_LRM_ATTR_TARGET);
if (crm_element_value(stored_msg, XML_LRM_ATTR_ROUTER_NODE)) {
is_remote_node = TRUE;
}
if (rsc) {
char *attr = NULL;
crm_info("Removing failcount for %s", rsc);
attr = crm_concat("fail-count", rsc, '-');
update_attrd(uname, attr, NULL, NULL, is_remote_node);
free(attr);
attr = crm_concat("last-failure", rsc, '-');
update_attrd(uname, attr, NULL, NULL, is_remote_node);
free(attr);
lrm_clear_last_failure(rsc, uname);
} else {
crm_log_xml_warn(stored_msg, "invalid failcount op");
}
return I_NULL;
}
void
do_dc_join_final(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)
{
crm_debug("Ensuring DC, quorum and node attributes are up-to-date");
#if !HAVE_ATOMIC_ATTRD
/* Ask attrd to write all attributes to disk. This is not needed for
* atomic attrd because atomic attrd does a peer sync and write-out
* when winning an election.
*/
update_attrd(NULL, NULL, NULL, NULL, FALSE);
#endif
crm_update_quorum(crm_have_quorum, TRUE);
}
static void
remote_proxy_cb(lrmd_t *lrmd, void *userdata, xmlNode *msg)
{
lrm_state_t *lrm_state = userdata;
const char *op = crm_element_value(msg, F_LRMD_IPC_OP);
const char *session = crm_element_value(msg, F_LRMD_IPC_SESSION);
int msg_id = 0;
/* sessions are raw ipc connections to IPC,
* all we do is proxy requests/responses exactly
* like they are given to us at the ipc level. */
CRM_CHECK(op != NULL, return);
CRM_CHECK(session != NULL, return);
crm_element_value_int(msg, F_LRMD_IPC_MSG_ID, &msg_id);
/* This is msg from remote ipc client going to real ipc server */
if (safe_str_eq(op, LRMD_IPC_OP_SHUTDOWN_REQ)) {
char *now_s = NULL;
time_t now = time(NULL);
crm_notice("Graceful proxy shutdown of %s", lrm_state->node_name);
now_s = crm_itoa(now);
update_attrd(lrm_state->node_name, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, TRUE);
free(now_s);
remote_proxy_ack_shutdown(lrmd);
return;
} else if (safe_str_eq(op, LRMD_IPC_OP_NEW)) {
int rc;
const char *channel = crm_element_value(msg, F_LRMD_IPC_IPC_SERVER);
CRM_CHECK(channel != NULL, return);
if (remote_proxy_new(lrm_state->node_name, session, channel) == NULL) {
remote_proxy_notify_destroy(lrmd, session);
}
crm_trace("new remote proxy client established to %s, session id %s", channel, session);
/* Look up stonith-watchdog-timeout and send to the remote peer for validation */
rc = fsa_cib_conn->cmds->query(fsa_cib_conn, XML_CIB_TAG_CRMCONFIG, NULL, cib_scope_local);
fsa_cib_conn->cmds->register_callback_full(fsa_cib_conn, rc, 10, FALSE, lrmd, "remote_config_check", remote_config_check, NULL);
} else if (safe_str_eq(op, LRMD_IPC_OP_DESTROY)) {
static void
remote_proxy_cb(lrmd_t *lrmd, void *userdata, xmlNode *msg)
{
lrm_state_t *lrm_state = userdata;
const char *op = crm_element_value(msg, F_LRMD_IPC_OP);
const char *session = crm_element_value(msg, F_LRMD_IPC_SESSION);
int msg_id = 0;
/* sessions are raw ipc connections to IPC,
* all we do is proxy requests/responses exactly
* like they are given to us at the ipc level. */
CRM_CHECK(op != NULL, return);
CRM_CHECK(session != NULL, return);
crm_element_value_int(msg, F_LRMD_IPC_MSG_ID, &msg_id);
/* This is msg from remote ipc client going to real ipc server */
if (safe_str_eq(op, LRMD_IPC_OP_SHUTDOWN_REQ)) {
char *now_s = NULL;
time_t now = time(NULL);
crm_notice("Graceful proxy shutdown of %s", lrm_state->node_name);
now_s = crm_itoa(now);
update_attrd(lrm_state->node_name, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, TRUE);
free(now_s);
remote_proxy_ack_shutdown(lrmd);
return;
} else if (safe_str_eq(op, LRMD_IPC_OP_NEW)) {
const char *channel = crm_element_value(msg, F_LRMD_IPC_IPC_SERVER);
CRM_CHECK(channel != NULL, return);
if (remote_proxy_new(lrm_state->node_name, session, channel) == NULL) {
remote_proxy_notify_destroy(lrmd, session);
}
crm_trace("new remote proxy client established to %s, session id %s", channel, session);
} else if (safe_str_eq(op, LRMD_IPC_OP_DESTROY)) {
enum crmd_fsa_input
handle_shutdown_request(xmlNode * stored_msg)
{
/* handle here to avoid potential version issues
* where the shutdown message/proceedure may have
* been changed in later versions.
*
* This way the DC is always in control of the shutdown
*/
char *now_s = NULL;
time_t now = time(NULL);
xmlNode *node_state = NULL;
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
if (host_from == NULL) {
/* we're shutting down and the DC */
host_from = fsa_our_uname;
}
crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state));
crm_log_xml(LOG_MSG, "message", stored_msg);
node_state = create_node_state(host_from, NULL, NULL, NULL, NULL,
CRMD_STATE_INACTIVE, FALSE, __FUNCTION__);
fsa_cib_anon_update(XML_CIB_TAG_STATUS, node_state, cib_quorum_override);
crm_log_xml_debug_2(node_state, "Shutdown update");
free_xml(node_state);
now_s = crm_itoa(now);
update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL);
crm_free(now_s);
/* will be picked up by the TE as long as its running */
return I_NULL;
}
static gboolean
update_failcount(xmlNode * event, const char *event_node_uuid, int rc, int target_rc, gboolean do_update)
{
int interval = 0;
char *task = NULL;
char *rsc_id = NULL;
char *attr_name = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
const char *on_uname = get_uname_from_event(event);
const char *origin = crm_element_value(event, XML_ATTR_ORIGIN);
if (rc == 99) {
/* this is an internal code for "we're busy, try again" */
return FALSE;
} else if (rc == target_rc) {
return FALSE;
}
if (safe_str_eq(origin, "build_active_RAs")) {
crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh",
id, rc, on_uname);
return FALSE;
}
CRM_LOG_ASSERT(on_uname != NULL);
if (on_uname == NULL) {
return TRUE;
}
if (failed_stop_offset == NULL) {
failed_stop_offset = strdup(INFINITY_S);
}
if (failed_start_offset == NULL) {
failed_start_offset = strdup(INFINITY_S);
}
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval), crm_err("Couldn't parse: %s", ID(event));
goto bail);
CRM_CHECK(task != NULL, goto bail);
CRM_CHECK(rsc_id != NULL, goto bail);
if (do_update || interval > 0) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
do_update = TRUE;
value = failed_start_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
value = failed_stop_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
value = failed_stop_offset;
} else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_DEMOTE)) {
do_update = TRUE;
}
if (value == NULL || safe_str_neq(value, INFINITY_S)) {
value = XML_NVPAIR_ATTR_VALUE "++";
}
if (do_update) {
char *now = crm_itoa(time(NULL));
gboolean is_remote_node = get_is_remote_from_event(event);
crm_warn("Updating failcount for %s on %s after failed %s:"
" rc=%d (update=%s, time=%s)", rsc_id, on_uname, task, rc, value, now);
attr_name = crm_concat("fail-count", rsc_id, '-');
update_attrd(on_uname, attr_name, value, NULL, is_remote_node);
free(attr_name);
attr_name = crm_concat("last-failure", rsc_id, '-');
update_attrd(on_uname, attr_name, now, NULL, is_remote_node);
free(attr_name);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
/*!
* \internal
* \brief Update failure-related node attributes if warranted
*
* \param[in] event XML describing operation that (maybe) failed
* \param[in] event_node_uuid Node that event occurred on
* \param[in] rc Actual operation return code
* \param[in] target_rc Expected operation return code
* \param[in] do_update If TRUE, do update regardless of operation type
* \param[in] ignore_failures If TRUE, update last failure but not fail count
*
* \return TRUE if this was not a direct nack, success or lrm status refresh
*/
static gboolean
update_failcount(xmlNode * event, const char *event_node_uuid, int rc,
int target_rc, gboolean do_update, gboolean ignore_failures)
{
int interval = 0;
char *task = NULL;
char *rsc_id = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
const char *on_uname = crm_peer_uname(event_node_uuid);
const char *origin = crm_element_value(event, XML_ATTR_ORIGIN);
/* Nothing needs to be done for success, lrm status refresh,
* or direct nack (internal code for "busy, try again")
*/
if ((rc == CRM_DIRECT_NACK_RC) || (rc == target_rc)) {
return FALSE;
} else if (safe_str_eq(origin, "build_active_RAs")) {
crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh",
id, rc, on_uname);
return FALSE;
}
/* Sanity check */
CRM_CHECK(on_uname != NULL, return TRUE);
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval),
crm_err("Couldn't parse: %s", ID(event)); goto bail);
CRM_CHECK(task != NULL, goto bail);
CRM_CHECK(rsc_id != NULL, goto bail);
/* Decide whether update is necessary and what value to use */
if ((interval > 0) || safe_str_eq(task, CRMD_ACTION_PROMOTE)
|| safe_str_eq(task, CRMD_ACTION_DEMOTE)) {
do_update = TRUE;
} else if (safe_str_eq(task, CRMD_ACTION_START)) {
do_update = TRUE;
if (failed_start_offset == NULL) {
failed_start_offset = strdup(INFINITY_S);
}
value = failed_start_offset;
} else if (safe_str_eq(task, CRMD_ACTION_STOP)) {
do_update = TRUE;
if (failed_stop_offset == NULL) {
failed_stop_offset = strdup(INFINITY_S);
}
value = failed_stop_offset;
}
/* Fail count will be either incremented or set to infinity */
if (value == NULL || safe_str_neq(value, INFINITY_S)) {
value = XML_NVPAIR_ATTR_VALUE "++";
}
if (do_update) {
char *now = crm_itoa(time(NULL));
char *attr_name = NULL;
gboolean is_remote_node = FALSE;
if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) {
is_remote_node = TRUE;
}
crm_info("Updating %s for %s on %s after failed %s: rc=%d (update=%s, time=%s)",
(ignore_failures? "last failure" : "failcount"),
rsc_id, on_uname, task, rc, value, now);
/* Update the fail count, if we're not ignoring failures */
if (!ignore_failures) {
attr_name = crm_concat("fail-count", rsc_id, '-');
update_attrd(on_uname, attr_name, value, NULL, is_remote_node);
free(attr_name);
}
/* Update the last failure time (even if we're ignoring failures,
* so that failure can still be detected and shown, e.g. by crm_mon)
*/
attr_name = crm_concat("last-failure", rsc_id, '-');
update_attrd(on_uname, attr_name, now, NULL, is_remote_node);
free(attr_name);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
/* aka. this is notification that we have (or have not) been accepted */
void
do_cl_join_finalize_respond(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)
{
xmlNode *tmp1 = NULL;
gboolean was_nack = TRUE;
static gboolean first_join = TRUE;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
const char *start_state = daemon_option("node_start_state");
int join_id = -1;
const char *op = crm_element_value(input->msg, F_CRM_TASK);
const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK);
const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM);
if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) {
crm_trace("Ignoring op=%s message", op);
return;
}
/* calculate if it was an ack or a nack */
if (crm_is_true(ack_nack)) {
was_nack = FALSE;
}
crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id);
if (was_nack) {
crm_err("Shutting down because cluster join with leader %s failed "
CRM_XS" join-%d NACK'd", welcome_from, join_id);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) {
crm_warn("Discarding our own welcome - we're no longer the DC");
return;
}
if (update_dc(input->msg) == FALSE) {
crm_warn("Discarding %s from node %s (expected from %s)",
op, welcome_from, fsa_our_dc);
return;
}
update_dc_expected(input->msg);
/* send our status section to the DC */
tmp1 = do_lrm_query(TRUE, fsa_our_uname);
if (tmp1 != NULL) {
xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id);
crm_debug("Confirming join-%d: sending local operation history to %s",
join_id, fsa_our_dc);
/*
* If this is the node's first join since the crmd started on it, clear
* any previous transient node attributes, to handle the case where
* the node restarted so quickly that the cluster layer didn't notice.
*
* Do not remove the resources though, they'll be cleaned up in
* do_dc_join_ack(). Removing them here creates a race condition if the
* crmd is being recovered. Instead of a list of active resources from
* the lrmd, we may end up with a blank status section. If we are _NOT_
* lucky, we will probe for the "wrong" instance of anonymous clones and
* end up with multiple active instances on the machine.
*/
if (first_join && is_not_set(fsa_input_register, R_SHUTDOWN)) {
first_join = FALSE;
erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0);
update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE);
update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE);
if (start_state) {
set_join_state(start_state);
}
}
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
if (AM_I_DC == FALSE) {
register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__);
update_attrd(NULL, NULL, NULL, NULL, FALSE);
}
free_xml(tmp1);
} else {
crm_err("Could not confirm join-%d with %s: Local operation history failed",
join_id, fsa_our_dc);
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
}
void
tengine_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data)
{
char *uuid = NULL;
int stonith_id = -1;
int transition_id = -1;
crm_action_t *action = NULL;
int call_id = data->call_id;
int rc = data->rc;
char *userdata = data->userdata;
CRM_CHECK(userdata != NULL, return);
crm_notice("Stonith operation %d/%s: %s (%d)", call_id, (char *)userdata,
pcmk_strerror(rc), rc);
if (AM_I_DC == FALSE) {
return;
}
/* crm_info("call=%d, optype=%d, node_name=%s, result=%d, node_list=%s, action=%s", */
/* op->call_id, op->optype, op->node_name, op->op_result, */
/* (char *)op->node_list, op->private_data); */
/* filter out old STONITH actions */
CRM_CHECK(decode_transition_key(userdata, &uuid, &transition_id, &stonith_id, NULL),
goto bail);
if (transition_graph->complete || stonith_id < 0 || safe_str_neq(uuid, te_uuid)
|| transition_graph->id != transition_id) {
crm_info("Ignoring STONITH action initiated outside of the current transition");
goto bail;
}
action = get_action(stonith_id, FALSE);
if (action == NULL) {
crm_err("Stonith action not matched");
goto bail;
}
stop_te_timer(action->timer);
if (rc == pcmk_ok) {
const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
const char *uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
const char *op = crm_meta_value(action->params, "stonith_action");
crm_info("Stonith operation %d for %s passed", call_id, target);
if (action->confirmed == FALSE) {
te_action_confirmed(action);
if (safe_str_eq("on", op)) {
const char *value = NULL;
char *now = crm_itoa(time(NULL));
update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, FALSE);
free(now);
value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_ALL);
update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, FALSE);
value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_SECURE);
update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, FALSE);
} else if (action->sent_update == FALSE) {
send_stonith_update(action, target, uuid);
action->sent_update = TRUE;
}
}
st_fail_count_reset(target);
} else {
const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
enum transition_action abort_action = tg_restart;
action->failed = TRUE;
crm_notice("Stonith operation %d for %s failed (%s): aborting transition.",
call_id, target, pcmk_strerror(rc));
/* If no fence devices were available, there's no use in immediately
* checking again, so don't start a new transition in that case.
*/
if (rc == -ENODEV) {
crm_warn("No devices found in cluster to fence %s, giving up",
target);
abort_action = tg_stop;
}
/* Increment the fail count now, so abort_for_stonith_failure() can
* check it. Non-DC nodes will increment it in tengine_stonith_notify().
*/
st_fail_count_increment(target);
abort_for_stonith_failure(abort_action, target, NULL);
}
update_graph(transition_graph, action);
trigger_graph();
bail:
free(userdata);
free(uuid);
return;
}
/* aka. this is notification that we have (or have not) been accepted */
void
do_cl_join_finalize_respond(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)
{
xmlNode *tmp1 = NULL;
gboolean was_nack = TRUE;
static gboolean first_join = TRUE;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
int join_id = -1;
const char *op = crm_element_value(input->msg, F_CRM_TASK);
const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK);
const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM);
if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) {
crm_trace("Ignoring op=%s message", op);
return;
}
/* calculate if it was an ack or a nack */
if (crm_is_true(ack_nack)) {
was_nack = FALSE;
}
crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id);
if (was_nack) {
crm_err("Join (join-%d) with leader %s failed (NACK'd): Shutting down",
join_id, welcome_from);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) {
crm_warn("Discarding our own welcome - we're no longer the DC");
return;
}
if (update_dc(input->msg) == FALSE) {
crm_warn("Discarding %s from %s (expected %s)", op, welcome_from, fsa_our_dc);
return;
}
/* send our status section to the DC */
crm_debug("Confirming join join-%d: %s", join_id, crm_element_value(input->msg, F_CRM_TASK));
tmp1 = do_lrm_query(TRUE, fsa_our_uname);
if (tmp1 != NULL) {
xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id);
crm_debug("join-%d: Join complete."
" Sending local LRM status to %s", join_id, fsa_our_dc);
if (first_join) {
first_join = FALSE;
/*
* Clear any previous transient node attribute and lrm operations
*
* Corosync has a nasty habit of not being able to tell if a
* node is returning or didn't leave in the first place.
* This confuses Pacemaker because it never gets a "node up"
* event which is normally used to clean up the status section.
*
* Do not remove the resources though, they'll be cleaned up in
* do_dc_join_ack(). Removing them here creates a race
* condition if the crmd is being recovered.
* Instead of a list of active resources from the lrmd
* we may end up with a blank status section.
* If we are _NOT_ lucky, we will probe for the "wrong" instance
* of anonymous clones and end up with multiple active
* instances on the machine.
*/
erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0);
/* Just in case attrd was still around too */
if (is_not_set(fsa_input_register, R_SHUTDOWN)) {
update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE);
update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE);
}
}
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
if (AM_I_DC == FALSE) {
register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__);
update_attrd(NULL, NULL, NULL, NULL, FALSE);
}
free_xml(tmp1);
} else {
crm_err("Could not send our LRM state to the DC");
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
}
