static void
join_make_offer(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *offer = NULL;
crm_node_t *member = (crm_node_t *)value;
CRM_ASSERT(member != NULL);
if (crm_is_peer_active(member) == FALSE) {
crm_info("Not making an offer to %s: not active (%s)", member->uname, member->state);
if(member->expected == NULL && safe_str_eq(member->state, CRM_NODE_LOST)) {
/* You would think this unsafe, but in fact this plus an
* active resource is what causes it to be fenced.
*
* Yes, this does mean that any node that dies at the same
* time as the old DC and is not running resource (still)
* won't be fenced.
*
* I'm not happy about this either.
*/
crm_update_peer_expected(__FUNCTION__, member, CRMD_JOINSTATE_DOWN);
}
return;
}
if (member->uname == NULL) {
crm_err("No recipient for welcome message");
return;
}
if (saved_ccm_membership_id != crm_peer_seq) {
saved_ccm_membership_id = crm_peer_seq;
crm_info("Making join offers based on membership %llu", crm_peer_seq);
}
if(user_data && member->join > crm_join_none) {
crm_info("Skipping %s: already known %d", member->uname, member->join);
return;
}
crm_update_peer_join(__FUNCTION__, (crm_node_t*)member, crm_join_none);
offer = create_request(CRM_OP_JOIN_OFFER, NULL, member->uname,
CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
crm_xml_add_int(offer, F_CRM_JOIN_ID, current_join_id);
/* send the welcome */
crm_info("join-%d: Sending offer to %s", current_join_id, member->uname);
send_cluster_message(member, crm_msg_crmd, offer, TRUE);
free_xml(offer);
crm_update_peer_join(__FUNCTION__, member, crm_join_welcomed);
/* crm_update_peer_expected(__FUNCTION__, member, CRMD_JOINSTATE_PENDING); */
}
/* A_DC_JOIN_PROCESS_ACK */
void
do_dc_join_ack(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)
{
int join_id = -1;
int call_id = 0;
ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
const char *op = crm_element_value(join_ack->msg, F_CRM_TASK);
const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM);
crm_node_t *peer = crm_get_peer(0, join_from);
if (safe_str_neq(op, CRM_OP_JOIN_CONFIRM) || peer == NULL) {
crm_debug("Ignoring op=%s message from %s", op, join_from);
return;
}
crm_trace("Processing ack from %s", join_from);
crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id);
if (peer->join != crm_join_finalized) {
crm_info("Join not in progress: ignoring join-%d from %s (phase = %d)",
join_id, join_from, peer->join);
return;
} else if (join_id != current_join_id) {
crm_err("Invalid response from %s: join-%d vs. join-%d",
join_from, join_id, current_join_id);
crm_update_peer_join(__FUNCTION__, peer, crm_join_nack);
return;
}
crm_update_peer_join(__FUNCTION__, peer, crm_join_confirmed);
crm_info("join-%d: Updating node state to %s for %s",
join_id, CRMD_JOINSTATE_MEMBER, join_from);
/* update CIB with the current LRM status from the node
* We dont need to notify the TE of these updates, a transition will
* be started in due time
*/
erase_status_tag(join_from, XML_CIB_TAG_LRM, cib_scope_local);
fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml,
cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL);
fsa_register_cib_callback(call_id, FALSE, NULL, join_update_complete_callback);
crm_debug("join-%d: Registered callback for LRM update %d", join_id, call_id);
}
void
initialize_join(gboolean before)
{
GHashTableIter iter;
crm_node_t *peer = NULL;
/* clear out/reset a bunch of stuff */
crm_debug("join-%d: Initializing join data (flag=%s)",
current_join_id, before ? "true" : "false");
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
crm_update_peer_join(__FUNCTION__, peer, crm_join_none);
}
if (before) {
if (max_generation_from != NULL) {
free(max_generation_from);
max_generation_from = NULL;
}
if (max_generation_xml != NULL) {
free_xml(max_generation_xml);
max_generation_xml = NULL;
}
clear_bit(fsa_input_register, R_HAVE_CIB);
clear_bit(fsa_input_register, R_CIB_ASKED);
}
}
static void
reap_dead_nodes(gpointer key, gpointer value, gpointer user_data)
{
crm_node_t *node = value;
if (crm_is_peer_active(node) == FALSE) {
crm_update_peer_join(__FUNCTION__, node, crm_join_none);
if(node && node->uname) {
election_remove(fsa_election, node->uname);
if (safe_str_eq(fsa_our_uname, node->uname)) {
crm_err("We're not part of the cluster anymore");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
} else if (AM_I_DC == FALSE && safe_str_eq(node->uname, fsa_our_dc)) {
crm_warn("Our DC node (%s) left the cluster", node->uname);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
}
}
if (fsa_state == S_INTEGRATION || fsa_state == S_FINALIZE_JOIN) {
check_join_state(fsa_state, __FUNCTION__);
}
fail_incompletable_actions(transition_graph, node->uuid);
}
}
static void
join_make_offer(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *offer = NULL;
crm_node_t *member = (crm_node_t *)value;
CRM_ASSERT(member != NULL);
if (crm_is_peer_active(member) == FALSE) {
crm_info("Not making an offer to %s: not active (%s)", member->uname, member->state);
if(member->expected == NULL && safe_str_eq(member->state, CRM_NODE_LOST)) {
crm_update_peer_expected(__FUNCTION__, member, CRMD_JOINSTATE_DOWN);
}
return;
}
if (member->uname == NULL) {
crm_err("No recipient for welcome message");
return;
}
if (saved_ccm_membership_id != crm_peer_seq) {
saved_ccm_membership_id = crm_peer_seq;
crm_info("Making join offers based on membership %llu", crm_peer_seq);
}
if(user_data && member->join > crm_join_none) {
crm_info("Skipping %s: already known %d", member->uname, member->join);
return;
}
crm_update_peer_join(__FUNCTION__, (crm_node_t*)member, crm_join_none);
offer = create_request(CRM_OP_JOIN_OFFER, NULL, member->uname,
CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
crm_xml_add_int(offer, F_CRM_JOIN_ID, current_join_id);
/* send the welcome */
crm_info("join-%d: Sending offer to %s", current_join_id, member->uname);
send_cluster_message(member, crm_msg_crmd, offer, TRUE);
free_xml(offer);
crm_update_peer_join(__FUNCTION__, member, crm_join_welcomed);
/* crm_update_peer_expected(__FUNCTION__, member, CRMD_JOINSTATE_PENDING); */
}
void
finalize_join_for(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *acknak = NULL;
xmlNode *tmp1 = NULL;
crm_node_t *join_node = value;
const char *join_to = join_node->uname;
if(join_node->join != crm_join_integrated) {
crm_trace("Skipping %s in state %d", join_to, join_node->join);
return;
}
/* make sure a node entry exists for the new node */
crm_trace("Creating node entry for %s", join_to);
tmp1 = create_xml_node(NULL, XML_CIB_TAG_NODE);
set_uuid(tmp1, XML_ATTR_UUID, join_node);
crm_xml_add(tmp1, XML_ATTR_UNAME, join_to);
fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1,
cib_scope_local | cib_quorum_override | cib_can_create);
free_xml(tmp1);
join_node = crm_get_peer(0, join_to);
if (crm_is_peer_active(join_node) == FALSE) {
/*
* NACK'ing nodes that the membership layer doesn't know about yet
* simply creates more churn
*
* Better to leave them waiting and let the join restart when
* the new membership event comes in
*
* All other NACKs (due to versions etc) should still be processed
*/
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_PENDING);
return;
}
/* send the ack/nack to the node */
acknak = create_request(CRM_OP_JOIN_ACKNAK, NULL, join_to,
CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
crm_xml_add_int(acknak, F_CRM_JOIN_ID, current_join_id);
crm_debug("join-%d: ACK'ing join request from %s",
current_join_id, join_to);
crm_xml_add(acknak, CRM_OP_JOIN_ACKNAK, XML_BOOLEAN_TRUE);
crm_update_peer_join(__FUNCTION__, join_node, crm_join_finalized);
crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_MEMBER);
send_cluster_message(crm_get_peer(0, join_to), crm_msg_crmd, acknak, TRUE);
free_xml(acknak);
return;
}
void
send_stonith_update(crm_action_t * action, const char *target, const char *uuid)
{
int rc = pcmk_ok;
crm_node_t *peer = NULL;
/* zero out the node-status & remove all LRM status info */
xmlNode *node_state = NULL;
CRM_CHECK(target != NULL, return);
CRM_CHECK(uuid != NULL, return);
/* Make sure the membership and join caches are accurate */
peer = crm_get_peer_full(0, target, CRM_GET_PEER_CLUSTER | CRM_GET_PEER_REMOTE);
CRM_CHECK(peer != NULL, return);
if (peer->uuid == NULL) {
crm_info("Recording uuid '%s' for node '%s'", uuid, target);
peer->uuid = strdup(uuid);
}
crm_update_peer_proc(__FUNCTION__, peer, crm_proc_none, NULL);
crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_LOST, 0);
crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN);
crm_update_peer_join(__FUNCTION__, peer, crm_join_none);
node_state =
do_update_node_cib(peer,
node_update_cluster | node_update_peer | node_update_join |
node_update_expected, NULL, __FUNCTION__);
/* Force our known ID */
crm_xml_add(node_state, XML_ATTR_UUID, uuid);
rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, node_state,
cib_quorum_override | cib_scope_local | cib_can_create);
/* Delay processing the trigger until the update completes */
crm_debug("Sending fencing update %d for %s", rc, target);
fsa_register_cib_callback(rc, FALSE, strdup(target), cib_fencing_updated);
/* Make sure it sticks */
/* fsa_cib_conn->cmds->bump_epoch(fsa_cib_conn, cib_quorum_override|cib_scope_local); */
erase_status_tag(target, XML_CIB_TAG_LRM, cib_scope_local);
erase_status_tag(target, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
free_xml(node_state);
return;
}
/* A_DC_JOIN_PROCESS_REQ */
void
do_dc_join_filter_offer(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 *generation = NULL;
int cmp = 0;
int join_id = -1;
gboolean ack_nack_bool = TRUE;
const char *ack_nack = CRMD_JOINSTATE_MEMBER;
ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM);
const char *ref = crm_element_value(join_ack->msg, F_CRM_REFERENCE);
crm_node_t *join_node = crm_get_peer(0, join_from);
crm_debug("Processing req from %s", join_from);
generation = join_ack->xml;
crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id);
if (max_generation_xml != NULL && generation != NULL) {
int lpc = 0;
const char *attributes[] = {
XML_ATTR_GENERATION_ADMIN,
XML_ATTR_GENERATION,
XML_ATTR_NUMUPDATES,
};
for (lpc = 0; cmp == 0 && lpc < DIMOF(attributes); lpc++) {
cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]);
}
}
if (join_id != current_join_id) {
crm_debug("Invalid response from %s: join-%d vs. join-%d",
join_from, join_id, current_join_id);
check_join_state(cur_state, __FUNCTION__);
return;
} else if (join_node == NULL || crm_is_peer_active(join_node) == FALSE) {
crm_err("Node %s is not a member", join_from);
ack_nack_bool = FALSE;
} else if (generation == NULL) {
crm_err("Generation was NULL");
ack_nack_bool = FALSE;
} else if (max_generation_xml == NULL) {
max_generation_xml = copy_xml(generation);
max_generation_from = strdup(join_from);
} else if (cmp < 0 || (cmp == 0 && safe_str_eq(join_from, fsa_our_uname))) {
crm_debug("%s has a better generation number than"
" the current max %s", join_from, max_generation_from);
if (max_generation_xml) {
crm_log_xml_debug(max_generation_xml, "Max generation");
}
crm_log_xml_debug(generation, "Their generation");
free(max_generation_from);
free_xml(max_generation_xml);
max_generation_from = strdup(join_from);
max_generation_xml = copy_xml(join_ack->xml);
}
if (ack_nack_bool == FALSE) {
/* NACK this client */
ack_nack = CRMD_JOINSTATE_NACK;
crm_update_peer_join(__FUNCTION__, join_node, crm_join_nack);
crm_err("Rejecting cluster join request from %s " CRM_XS
" NACK join-%d ref=%s", join_from, join_id, ref);
} else {
crm_debug("join-%d: Welcoming node %s (ref %s)", join_id, join_from, ref);
crm_update_peer_join(__FUNCTION__, join_node, crm_join_integrated);
}
crm_update_peer_expected(__FUNCTION__, join_node, ack_nack);
crm_debug("%u nodes have been integrated into join-%d",
crmd_join_phase_count(crm_join_integrated), join_id);
if (check_join_state(cur_state, __FUNCTION__) == FALSE) {
/* don't waste time by invoking the PE yet; */
crm_debug("join-%d: Still waiting on %d outstanding offers",
join_id, crmd_join_phase_count(crm_join_welcomed));
}
}
/* A_DC_JOIN_OFFER_ONE */
void
do_dc_join_offer_one(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_node_t *member;
ha_msg_input_t *welcome = NULL;
const char *op = NULL;
const char *join_to = NULL;
if (msg_data->data) {
welcome = fsa_typed_data(fsa_dt_ha_msg);
} else {
crm_info("An unknown node joined - (re-)offer to any unconfirmed nodes");
g_hash_table_foreach(crm_peer_cache, join_make_offer, &member);
check_join_state(cur_state, __FUNCTION__);
return;
}
if (welcome == NULL) {
crm_err("Attempt to send welcome message without a message to reply to!");
return;
}
join_to = crm_element_value(welcome->msg, F_CRM_HOST_FROM);
if (join_to == NULL) {
crm_err("Attempt to send welcome message without a host to reply to!");
return;
}
member = crm_get_peer(0, join_to);
op = crm_element_value(welcome->msg, F_CRM_TASK);
if (join_to != NULL && (cur_state == S_INTEGRATION || cur_state == S_FINALIZE_JOIN)) {
/* note: it _is_ possible that a node will have been
* sick or starting up when the original offer was made.
* however, it will either re-announce itself in due course
* _or_ we can re-store the original offer on the client.
*/
crm_trace("(Re-)offering membership to %s...", join_to);
}
crm_info("join-%d: Processing %s request from %s in state %s",
current_join_id, op, join_to, fsa_state2string(cur_state));
crm_update_peer_join(__FUNCTION__, member, crm_join_none);
join_make_offer(NULL, member, NULL);
/* always offer to the DC (ourselves)
* this ensures the correct value for max_generation_from
*/
member = crm_get_peer(0, fsa_our_uname);
join_make_offer(NULL, member, NULL);
/* this was a genuine join request, cancel any existing
* transition and invoke the PE
*/
abort_transition(INFINITY, tg_restart, "Node join", NULL);
/* don't waste time by invoking the PE yet; */
crm_debug("Waiting on %d outstanding join acks for join-%d",
crmd_join_phase_count(crm_join_welcomed), current_join_id);
}
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
uint32_t old = 0;
uint32_t changed = 0;
bool appeared = FALSE;
bool is_remote = is_set(node->flags, crm_remote_node);
const char *status = NULL;
/* Crmd waits to receive some information from the membership layer before
* declaring itself operational. If this is being called for a cluster node,
* indicate that we have it.
*/
if (!is_remote) {
set_bit(fsa_input_register, R_PEER_DATA);
}
if (node->uname == NULL) {
return;
}
switch (type) {
case crm_status_uname:
/* If we've never seen the node, then it also won't be in the status section */
crm_info("%s node %s is now %s",
(is_remote? "Remote" : "Cluster"),
node->uname, state_text(node->state));
return;
case crm_status_rstate:
case crm_status_nstate:
/* This callback should not be called unless the state actually
* changed, but here's a failsafe just in case.
*/
CRM_CHECK(safe_str_neq(data, node->state), return);
crm_info("%s node %s is now %s (was %s)",
(is_remote? "Remote" : "Cluster"),
node->uname, state_text(node->state), state_text(data));
if (safe_str_eq(CRM_NODE_MEMBER, node->state)) {
appeared = TRUE;
if (!is_remote) {
remove_stonith_cleanup(node->uname);
}
}
crmd_alert_node_event(node);
break;
case crm_status_processes:
if (data) {
old = *(const uint32_t *)data;
changed = node->processes ^ old;
}
status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
node->uname, peer2text(proc_flags), status,
AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);
if ((changed & proc_flags) == 0) {
/* Peer process did not change */
crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
return;
} else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
crm_trace("Not connected");
return;
} else if (fsa_state == S_STOPPING) {
crm_trace("Stopping");
return;
}
appeared = (node->processes & proc_flags) != 0;
if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
/* Did we get evicted? */
crm_notice("Our peer connection failed");
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);
} else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
/* Did the DC leave us? */
crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);
/* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't
* want to fence it. Newer DCs will send their shutdown request
* to all peers, who will update the DC's expected state to
* down, thus avoiding fencing. We can safely erase the DC's
* transient attributes when it leaves in that case. However,
* the only way to avoid fencing older DCs is to leave the
* transient attributes intact until it rejoins.
*/
if (compare_version(fsa_our_dc_version, "3.0.9") > 0) {
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
}
} else if(AM_I_DC && appeared == FALSE) {
crm_info("Peer %s left us", node->uname);
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
}
break;
}
if (AM_I_DC) {
xmlNode *update = NULL;
int flags = node_update_peer;
gboolean alive = is_remote? appeared : crm_is_peer_active(node);
crm_action_t *down = match_down_event(node->uuid, appeared);
crm_trace("Alive=%d, appeared=%d, down=%d",
alive, appeared, (down? down->id : -1));
if (alive && type == crm_status_processes) {
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
if (down) {
const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);
if (safe_str_eq(task, CRM_OP_FENCE)) {
/* tengine_stonith_callback() confirms fence actions */
crm_trace("Updating CIB %s stonithd reported fencing of %s complete",
(down->confirmed? "after" : "before"), node->uname);
} else if ((alive == FALSE) && safe_str_eq(task, CRM_OP_SHUTDOWN)) {
crm_notice("%s of peer %s is complete "CRM_XS" op=%d",
task, node->uname, down->id);
/* down->confirmed = TRUE; */
stop_te_timer(down->timer);
if (!is_remote) {
flags |= node_update_join | node_update_expected;
crmd_peer_down(node, FALSE);
check_join_state(fsa_state, __FUNCTION__);
}
update_graph(transition_graph, down);
trigger_graph();
} else {
crm_trace("Node %s is %salive, was expected to %s (op %d)",
node->uname, (alive? "" : "not "), task, down->id);
}
} else if (appeared == FALSE) {
crm_notice("Stonith/shutdown of %s not matched", node->uname);
if (!is_remote) {
crm_update_peer_join(__FUNCTION__, node, crm_join_none);
check_join_state(fsa_state, __FUNCTION__);
}
abort_transition(INFINITY, tg_restart, "Node failure", NULL);
fail_incompletable_actions(transition_graph, node->uuid);
} else {
crm_trace("Node %s came up, was not expected to be down",
node->uname);
}
if (is_remote) {
/* A pacemaker_remote node won't have its cluster status updated
* in the CIB by membership-layer callbacks, so do it here.
*/
flags |= node_update_cluster;
/* Trigger resource placement on newly integrated nodes */
if (appeared) {
abort_transition(INFINITY, tg_restart,
"pacemaker_remote node integrated", NULL);
}
}
/* Update the CIB node state */
update = create_node_state_update(node, flags, NULL, __FUNCTION__);
fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
cib_scope_local | cib_quorum_override | cib_can_create);
free_xml(update);
}
trigger_fsa(fsa_source);
}
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
uint32_t old = 0;
uint32_t changed = 0;
bool appeared = FALSE;
const char *status = NULL;
set_bit(fsa_input_register, R_PEER_DATA);
if (node->uname == NULL) {
return;
}
switch (type) {
case crm_status_uname:
/* If we've never seen the node, then it also wont be in the status section */
crm_info("%s is now %s", node->uname, node->state);
return;
case crm_status_rstate:
crm_info("Remote node %s is now %s (was %s)", node->uname, node->state, (const char *)data);
/* Keep going */
case crm_status_nstate:
crm_info("%s is now %s (was %s)", node->uname, node->state, (const char *)data);
if (safe_str_eq(data, node->state)) {
/* State did not change */
return;
} else if(safe_str_eq(CRM_NODE_MEMBER, node->state)) {
appeared = TRUE;
}
break;
case crm_status_processes:
if (data) {
old = *(const uint32_t *)data;
changed = node->processes ^ old;
}
/* crmd_proc_update(node, proc_flags); */
status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
node->uname, peer2text(proc_flags), status,
AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);
if ((changed & proc_flags) == 0) {
/* Peer process did not change */
crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
return;
} else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
crm_trace("Not connected");
return;
} else if (fsa_state == S_STOPPING) {
crm_trace("Stopping");
return;
}
appeared = (node->processes & proc_flags) != 0;
if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
/* Did we get evicted? */
crm_notice("Our peer connection failed");
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);
} else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
/* Did the DC leave us? */
crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);
} else if(AM_I_DC && appeared == FALSE) {
crm_info("Peer %s left us", node->uname);
/* crm_update_peer_join(__FUNCTION__, node, crm_join_none); */
}
break;
}
if (AM_I_DC) {
xmlNode *update = NULL;
int flags = node_update_peer;
gboolean alive = crm_is_peer_active(node);
crm_action_t *down = match_down_event(0, node->uuid, NULL, appeared);
crm_trace("Alive=%d, appear=%d, down=%p", alive, appeared, down);
if (alive && type == crm_status_processes) {
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
if (down) {
const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);
if (alive && safe_str_eq(task, CRM_OP_FENCE)) {
crm_info("Node return implies stonith of %s (action %d) completed", node->uname,
down->id);
erase_status_tag(node->uname, XML_CIB_TAG_LRM, cib_scope_local);
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
/* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
down->sent_update = TRUE; /* Prevent tengine_stonith_callback() from calling send_stonith_update() */
} else if (safe_str_eq(task, CRM_OP_FENCE)) {
crm_trace("Waiting for stonithd to report the fencing of %s is complete", node->uname); /* via tengine_stonith_callback() */
} else if (alive == FALSE) {
crm_notice("%s of %s (op %d) is complete", task, node->uname, down->id);
/* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
stop_te_timer(down->timer);
flags |= node_update_join | node_update_expected;
crmd_peer_down(node, FALSE);
check_join_state(fsa_state, __FUNCTION__);
update_graph(transition_graph, down);
trigger_graph();
} else {
crm_trace("Other %p", down);
}
} else if (appeared == FALSE) {
crm_notice("Stonith/shutdown of %s not matched", node->uname);
crm_update_peer_join(__FUNCTION__, node, crm_join_none);
check_join_state(fsa_state, __FUNCTION__);
abort_transition(INFINITY, tg_restart, "Node failure", NULL);
fail_incompletable_actions(transition_graph, node->uuid);
} else {
crm_trace("Other %p", down);
}
update = do_update_node_cib(node, flags, NULL, __FUNCTION__);
fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
cib_scope_local | cib_quorum_override | cib_can_create);
free_xml(update);
}
trigger_fsa(fsa_source);
}
static void
tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event)
{
if(te_client_id == NULL) {
te_client_id = g_strdup_printf("%s.%d", crm_system_name, getpid());
}
if (st_event == NULL) {
crm_err("Notify data not found");
return;
}
if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) {
crm_crit("We were alegedly just fenced by %s for %s with %s!", st_event->executioner,
st_event->origin, st_event->device); /* Dumps blackbox if enabled */
qb_log_fini(); /* Try to get the above log message to disk - somehow */
/* Get out ASAP and do not come back up.
*
* Triggering a reboot is also not the worst idea either since
* the rest of the cluster thinks we're safely down
*/
#ifdef RB_HALT_SYSTEM
reboot(RB_HALT_SYSTEM);
#endif
/*
* If reboot() fails or is not supported, coming back up will
* probably lead to a situation where the other nodes set our
* status to 'lost' because of the fencing callback and will
* discard subsequent election votes with:
*
* Election 87 (current: 5171, owner: 103): Processed vote from east-03 (Peer is not part of our cluster)
*
* So just stay dead, something is seriously messed up anyway.
*
*/
exit(100); /* None of our wrappers since we already called qb_log_fini() */
return;
}
if (st_event->result == pcmk_ok &&
safe_str_eq(st_event->operation, T_STONITH_NOTIFY_FENCE)) {
st_fail_count_reset(st_event->target);
}
crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s) by client %s",
st_event->target, st_event->result == pcmk_ok ? "" : " not",
st_event->action,
st_event->executioner ? st_event->executioner : "<anyone>",
st_event->origin, pcmk_strerror(st_event->result), st_event->id,
st_event->client_origin ? st_event->client_origin : "<unknown>");
#if SUPPORT_CMAN
if (st_event->result == pcmk_ok && is_cman_cluster()) {
int local_rc = 0;
char *target_copy = strdup(st_event->target);
/* In case fenced hasn't noticed yet
*
* Any fencing that has been inititated will be completed by way of the fence_pcmk redirect
*/
local_rc = fenced_external(target_copy);
if (local_rc != 0) {
crm_err("Could not notify CMAN that '%s' is now fenced: %d", st_event->target,
local_rc);
} else {
crm_notice("Notified CMAN that '%s' is now fenced", st_event->target);
}
free(target_copy);
}
#endif
if (st_event->result == pcmk_ok) {
crm_node_t *peer = crm_get_peer_full(0, st_event->target, CRM_GET_PEER_REMOTE | CRM_GET_PEER_CLUSTER);
const char *uuid = crm_peer_uuid(peer);
gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname);
crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc);
if(AM_I_DC) {
/* The DC always sends updates */
send_stonith_update(NULL, st_event->target, uuid);
if (st_event->client_origin && safe_str_neq(st_event->client_origin, te_client_id)) {
/* Abort the current transition graph if it wasn't us
* that invoked stonith to fence someone
*/
crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target);
abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL);
}
/* Assume it was our leader if we dont currently have one */
} else if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, st_event->target)) {
crm_notice("Target %s our leader %s (recorded: %s)",
fsa_our_dc ? "was" : "may have been", st_event->target,
fsa_our_dc ? fsa_our_dc : "<unset>");
/* Given the CIB resyncing that occurs around elections,
* have one node update the CIB now and, if the new DC is different,
* have them do so too after the election
*/
if (we_are_executioner) {
send_stonith_update(NULL, st_event->target, uuid);
}
stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target));
}
crm_update_peer_proc(__FUNCTION__, peer, crm_proc_none, NULL);
crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_LOST, 0);
crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN);
crm_update_peer_join(__FUNCTION__, peer, crm_join_none);
}
}
