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;
}
static void
attrd_cpg_dispatch(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
uint32_t kind = 0;
xmlNode *xml = NULL;
const char *from = NULL;
char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from);
if(data == NULL) {
return;
}
if (kind == crm_class_cluster) {
xml = string2xml(data);
}
if (xml == NULL) {
crm_err("Bad message of class %d received from %s[%u]: '%.120s'", kind, from, nodeid, data);
} else {
crm_node_t *peer = crm_get_peer(nodeid, from);
attrd_peer_message(peer, xml);
}
free_xml(xml);
free(data);
}
/*!
* \internal
* \brief Update a node's process list, notifying clients and peers if needed
*
* \param[in] id Node ID of affected node
* \param[in] uname Uname of affected node
* \param[in] procs Affected node's process list mask
*
* \return TRUE if the process list changed, FALSE otherwise
*/
gboolean
update_node_processes(uint32_t id, const char *uname, uint32_t procs)
{
gboolean changed = FALSE;
crm_node_t *node = crm_get_peer(id, uname);
if (procs != 0) {
if (procs != node->processes) {
crm_debug("Node %s now has process list: %.32x (was %.32x)",
node->uname, procs, node->processes);
node->processes = procs;
changed = TRUE;
/* If local node's processes have changed, notify clients/peers */
if (id == local_nodeid) {
update_process_clients(NULL);
update_process_peers();
}
} else {
crm_trace("Node %s still has process list: %.32x", node->uname, procs);
}
}
return changed;
}
/*!
* \brief Check to see if this operation is a duplicate of another in flight
* operation. If so merge this operation into the inflight operation, and mark
* it as a duplicate.
*/
static void
merge_duplicates(remote_fencing_op_t * op)
{
GHashTableIter iter;
remote_fencing_op_t *other = NULL;
time_t now = time(NULL);
g_hash_table_iter_init(&iter, remote_op_list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&other)) {
crm_node_t *peer = NULL;
if (other->state > st_exec) {
/* Must be in-progress */
continue;
} else if (safe_str_neq(op->target, other->target)) {
/* Must be for the same node */
continue;
} else if (safe_str_neq(op->action, other->action)) {
crm_trace("Must be for the same action: %s vs. ", op->action, other->action);
continue;
} else if (safe_str_eq(op->client_name, other->client_name)) {
crm_trace("Must be for different clients: %s", op->client_name);
continue;
} else if (safe_str_eq(other->target, other->originator)) {
crm_trace("Can't be a suicide operation: %s", other->target);
continue;
}
peer = crm_get_peer(0, other->originator);
if(fencing_peer_active(peer) == FALSE) {
crm_notice("Failing stonith action %s for node %s originating from %s@%s.%.8s: Originator is dead",
other->action, other->target, other->client_name, other->originator, other->id);
other->state = st_failed;
continue;
} else if(other->total_timeout > 0 && now > (other->total_timeout + other->created)) {
crm_info("Stonith action %s for node %s originating from %s@%s.%.8s is too old: %d vs. %d + %d",
other->action, other->target, other->client_name, other->originator, other->id,
now, other->created, other->total_timeout);
continue;
}
/* There is another in-flight request to fence the same host
* Piggyback on that instead. If it fails, so do we.
*/
other->duplicates = g_list_append(other->duplicates, op);
if (other->total_timeout == 0) {
crm_trace("Making a best-guess as to the timeout used");
other->total_timeout = op->total_timeout =
TIMEOUT_MULTIPLY_FACTOR * get_op_total_timeout(op, NULL, op->base_timeout);
}
crm_notice
("Merging stonith action %s for node %s originating from client %s.%.8s with identical request from %s@%s.%.8s (%ds)",
op->action, op->target, op->client_name, op->id, other->client_name, other->originator,
other->id, other->total_timeout);
report_timeout_period(op, other->total_timeout);
op->state = st_duplicate;
}
}
/* A_DC_TAKEOVER */
void
do_dc_takeover(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 rc = pcmk_ok;
xmlNode *cib = NULL;
GListPtr gIter = NULL;
const char *cluster_type = name_for_cluster_type(get_cluster_type());
crm_info("Taking over DC status for this partition");
set_bit(fsa_input_register, R_THE_DC);
for (gIter = stonith_cleanup_list; gIter != NULL; gIter = gIter->next) {
char *target = gIter->data;
crm_node_t *target_node = crm_get_peer(0, target);
const char *uuid = crm_peer_uuid(target_node);
crm_notice("Marking %s, target of a previous stonith action, as clean", target);
send_stonith_update(NULL, target, uuid);
free(target);
}
g_list_free(stonith_cleanup_list);
stonith_cleanup_list = NULL;
#if SUPPORT_COROSYNC
if (is_classic_ais_cluster()) {
send_ais_text(crm_class_quorum, NULL, TRUE, NULL, crm_msg_ais);
}
#endif
if (voted != NULL) {
crm_trace("Destroying voted hash");
g_hash_table_destroy(voted);
voted = NULL;
}
set_bit(fsa_input_register, R_JOIN_OK);
set_bit(fsa_input_register, R_INVOKE_PE);
fsa_cib_conn->cmds->set_master(fsa_cib_conn, cib_scope_local);
cib = create_xml_node(NULL, XML_TAG_CIB);
crm_xml_add(cib, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
fsa_cib_update(XML_TAG_CIB, cib, cib_quorum_override, rc, NULL);
fsa_register_cib_callback(rc, FALSE, NULL, feature_update_callback);
update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL,
"dc-version", VERSION "-" BUILD_VERSION, FALSE, NULL);
update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL,
"cluster-infrastructure", cluster_type, FALSE, NULL);
mainloop_set_trigger(config_read);
free_xml(cib);
}
/*!
* \internal
* \brief Remember if DC is shutting down as we join
*
* If we're joining while the current DC is shutting down, update its expected
* state, so we don't fence it if we become the new DC. (We weren't a peer
* when it broadcast its shutdown request.)
*
* \param[in] msg A join message from the DC
*/
static void
update_dc_expected(xmlNode *msg)
{
if (fsa_our_dc && crm_is_true(crm_element_value(msg, F_CRM_DC_LEAVING))) {
crm_node_t *dc_node = crm_get_peer(0, fsa_our_dc);
crm_update_peer_expected(__FUNCTION__, dc_node, CRMD_JOINSTATE_DOWN);
}
}
gboolean
finalize_join_for(gpointer key, gpointer value, gpointer user_data)
{
const char *join_to = NULL;
const char *join_state = NULL;
xmlNode *acknak = NULL;
crm_node_t *join_node = NULL;
if (key == NULL || value == NULL) {
return TRUE;
}
join_to = (const char *)key;
join_state = (const char *)value;
/* make sure the node exists in the config section */
create_node_entry(join_to, join_to, NORMALNODE);
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
*/
return TRUE;
}
/* 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);
/* set the ack/nack */
if (safe_str_eq(join_state, CRMD_JOINSTATE_MEMBER)) {
crm_debug("join-%d: ACK'ing join request from %s, state %s",
current_join_id, join_to, join_state);
crm_xml_add(acknak, CRM_OP_JOIN_ACKNAK, XML_BOOLEAN_TRUE);
g_hash_table_insert(finalized_nodes,
strdup(join_to), strdup(CRMD_JOINSTATE_MEMBER));
} else {
crm_warn("join-%d: NACK'ing join request from %s, state %s",
current_join_id, join_to, join_state);
crm_xml_add(acknak, CRM_OP_JOIN_ACKNAK, XML_BOOLEAN_FALSE);
}
send_cluster_message(join_to, crm_msg_crmd, acknak, TRUE);
free_xml(acknak);
return TRUE;
}
int
stonith_fence_history(xmlNode * msg, xmlNode ** output)
{
int rc = 0;
const char *target = NULL;
xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_TRACE);
if (dev) {
int options = 0;
target = crm_element_value(dev, F_STONITH_TARGET);
crm_element_value_int(msg, F_STONITH_CALLOPTS, &options);
if (target && (options & st_opt_cs_nodeid)) {
int nodeid = crm_atoi(target, NULL);
crm_node_t *node = crm_get_peer(nodeid, NULL);
if (node) {
target = node->uname;
}
}
}
crm_trace("Looking for operations on %s in %p", target, remote_op_list);
*output = create_xml_node(NULL, F_STONITH_HISTORY_LIST);
if (remote_op_list) {
GHashTableIter iter;
remote_fencing_op_t *op = NULL;
g_hash_table_iter_init(&iter, remote_op_list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&op)) {
xmlNode *entry = NULL;
if (target && strcmp(op->target, target) != 0) {
continue;
}
rc = 0;
crm_trace("Attaching op %s", op->id);
entry = create_xml_node(*output, STONITH_OP_EXEC);
crm_xml_add(entry, F_STONITH_TARGET, op->target);
crm_xml_add(entry, F_STONITH_ACTION, op->action);
crm_xml_add(entry, F_STONITH_ORIGIN, op->originator);
crm_xml_add(entry, F_STONITH_DELEGATE, op->delegate);
crm_xml_add(entry, F_STONITH_CLIENTNAME, op->client_name);
crm_xml_add_int(entry, F_STONITH_DATE, op->completed);
crm_xml_add_int(entry, F_STONITH_STATE, op->state);
}
}
return rc;
}
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);
if (get_node_uuid(0, target) == NULL) {
set_node_uuid(target, uuid);
}
/* Make sure the membership and join caches are accurate */
peer = crm_get_peer(0, target);
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);
erase_node_from_join(target);
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;
}
static void
report_timeout_period(remote_fencing_op_t * op, int op_timeout)
{
GListPtr iter = NULL;
xmlNode *update = NULL;
const char *client_node = NULL;
const char *client_id = NULL;
const char *call_id = NULL;
if (op->call_options & st_opt_sync_call) {
/* There is no reason to report the timeout for a syncronous call. It
* is impossible to use the reported timeout to do anything when the client
* is blocking for the response. This update is only important for
* async calls that require a callback to report the results in. */
return;
} else if (!op->request) {
return;
}
crm_trace("Reporting timeout for %s.%.8s", op->client_name, op->id);
client_node = crm_element_value(op->request, F_STONITH_CLIENTNODE);
call_id = crm_element_value(op->request, F_STONITH_CALLID);
client_id = crm_element_value(op->request, F_STONITH_CLIENTID);
if (!client_node || !call_id || !client_id) {
return;
}
if (safe_str_eq(client_node, stonith_our_uname)) {
/* The client is connected to this node, send the update direclty to them */
do_stonith_async_timeout_update(client_id, call_id, op_timeout);
return;
}
/* The client is connected to another node, relay this update to them */
update = stonith_create_op(op->client_callid, op->id, STONITH_OP_TIMEOUT_UPDATE, NULL, 0);
crm_xml_add(update, F_STONITH_REMOTE_OP_ID, op->id);
crm_xml_add(update, F_STONITH_CLIENTID, client_id);
crm_xml_add(update, F_STONITH_CALLID, call_id);
crm_xml_add_int(update, F_STONITH_TIMEOUT, op_timeout);
send_cluster_message(crm_get_peer(0, client_node), crm_msg_stonith_ng, update, FALSE);
free_xml(update);
for (iter = op->duplicates; iter != NULL; iter = iter->next) {
remote_fencing_op_t *dup = iter->data;
crm_trace("Reporting timeout for duplicate %s.%.8s", dup->client_name, dup->id);
report_timeout_period(iter->data, op_timeout);
}
}
/* 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);
}
static gboolean
send_peer_reply(xmlNode * msg, xmlNode * result_diff, const char *originator, gboolean broadcast)
{
CRM_ASSERT(msg != NULL);
if (broadcast) {
/* this (successful) call modified the CIB _and_ the
* change needs to be broadcast...
* send via HA to other nodes
*/
int diff_add_updates = 0;
int diff_add_epoch = 0;
int diff_add_admin_epoch = 0;
int diff_del_updates = 0;
int diff_del_epoch = 0;
int diff_del_admin_epoch = 0;
const char *digest = NULL;
CRM_LOG_ASSERT(result_diff != NULL);
digest = crm_element_value(result_diff, XML_ATTR_DIGEST);
cib_diff_version_details(result_diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
crm_trace("Sending update diff %d.%d.%d -> %d.%d.%d %s",
diff_del_admin_epoch, diff_del_epoch, diff_del_updates,
diff_add_admin_epoch, diff_add_epoch, diff_add_updates, digest);
crm_xml_add(msg, F_CIB_ISREPLY, originator);
crm_xml_add(msg, F_CIB_GLOBAL_UPDATE, XML_BOOLEAN_TRUE);
crm_xml_add(msg, F_CIB_OPERATION, CIB_OP_APPLY_DIFF);
CRM_ASSERT(digest != NULL);
add_message_xml(msg, F_CIB_UPDATE_DIFF, result_diff);
crm_log_xml_explicit(msg, "copy");
return send_cluster_message(NULL, crm_msg_cib, msg, TRUE);
} else if (originator != NULL) {
/* send reply via HA to originating node */
crm_trace("Sending request result to %s only", originator);
crm_xml_add(msg, F_CIB_ISREPLY, originator);
return send_cluster_message(crm_get_peer(0, originator), crm_msg_cib, msg, FALSE);
}
return FALSE;
}
void
send_sync_request(const char *host)
{
xmlNode *sync_me = create_xml_node(NULL, "sync-me");
crm_info("Requesting re-sync from peer");
sync_in_progress++;
crm_xml_add(sync_me, F_TYPE, "cib");
crm_xml_add(sync_me, F_CIB_OPERATION, CIB_OP_SYNC_ONE);
crm_xml_add(sync_me, F_CIB_DELEGATED, cib_our_uname);
send_cluster_message(host ? crm_get_peer(0, host) : NULL, crm_msg_cib, sync_me, FALSE);
free_xml(sync_me);
}
/*!
* \internal
* \brief Send stonith updates for all entries in cleanup list, then purge it
*/
void
execute_stonith_cleanup()
{
GListPtr iter;
for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) {
char *target = iter->data;
crm_node_t *target_node = crm_get_peer(0, target);
const char *uuid = crm_peer_uuid(target_node);
crm_notice("Marking %s, target of a previous stonith action, as clean", target);
send_stonith_update(NULL, target, uuid);
free(target);
}
g_list_free(stonith_cleanup_list);
stonith_cleanup_list = NULL;
}
crm_node_t *
crm_get_peer_full(unsigned int id, const char *uname, int flags)
{
crm_node_t *node = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
crm_peer_init();
if (flags & CRM_GET_PEER_REMOTE) {
node = g_hash_table_lookup(crm_remote_peer_cache, uname);
}
if (node == NULL && (flags & CRM_GET_PEER_CLUSTER)) {
node = crm_get_peer(id, uname);
}
return node;
}
void
join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
xmlNode *local_cib = NULL;
char *join_id = user_data;
xmlNode *generation = create_xml_node(NULL, XML_CIB_TAG_GENERATION_TUPPLE);
CRM_LOG_ASSERT(join_id != NULL);
query_call_id = 0;
if (rc == pcmk_ok) {
local_cib = output;
CRM_LOG_ASSERT(safe_str_eq(crm_element_name(local_cib), XML_TAG_CIB));
}
if (local_cib != NULL) {
xmlNode *reply = NULL;
crm_debug("Respond to join offer join-%s", join_id);
crm_debug("Acknowledging %s as our DC", fsa_our_dc);
copy_in_properties(generation, local_cib);
reply = create_request(CRM_OP_JOIN_REQUEST, generation, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add(reply, F_CRM_JOIN_ID, join_id);
if (fsa_our_dc) {
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
} else {
crm_warn("No DC for join-%s", join_id);
send_cluster_message(NULL, crm_msg_crmd, reply, TRUE);
}
free_xml(reply);
} else {
crm_err("Could not retrieve Generation to attach to our"
" join acknowledgement: %s", pcmk_strerror(rc));
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
}
free(join_id);
free_xml(generation);
}
/* ELECTION投票処理 */
void
election_vote(election_t *e)
{
struct timeval age;
xmlNode *vote = NULL;
crm_node_t *our_node;
if(e == NULL) {
crm_trace("Not voting in election: not initialized");
return;
}
our_node = crm_get_peer(0, e->uname);
if (our_node == NULL || crm_is_peer_active(our_node) == FALSE) {
crm_trace("Cannot vote yet: %p", our_node);
return;
}
/* 自ノードの情報を生成する */
e->state = election_in_progress;
/* CRM_OP_VOTE(F_ATTRD_TASKフィールド)メッセージを生成する */
vote = create_request(CRM_OP_VOTE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
e->count++;
crm_xml_add(vote, F_CRM_ELECTION_OWNER, our_node->uuid);
crm_xml_add_int(vote, F_CRM_ELECTION_ID, e->count);
crm_uptime(&age);
crm_xml_add_int(vote, F_CRM_ELECTION_AGE_S, age.tv_sec);
crm_xml_add_int(vote, F_CRM_ELECTION_AGE_US, age.tv_usec);
/* クラスタに投票メッセージを送信する */
send_cluster_message(NULL, crm_msg_crmd, vote, TRUE);
free_xml(vote);
crm_debug("Started election %d", e->count);
if (e->voted) {
g_hash_table_destroy(e->voted);
e->voted = NULL;
}
/* ELECTIONのタイマーを開始する */
election_timeout_start(e);
return;
}
gboolean
init_cs_connection_once(crm_cluster_t * cluster)
{
const char *uuid = NULL;
crm_node_t *peer = NULL;
enum cluster_type_e stack = get_cluster_type();
crm_peer_init();
/* Here we just initialize comms */
if (stack != pcmk_cluster_corosync) {
crm_err("Invalid cluster type: %s (%d)", name_for_cluster_type(stack), stack);
return FALSE;
}
if (cluster_connect_cpg(cluster) == FALSE) {
return FALSE;
}
crm_info("Connection to '%s': established", name_for_cluster_type(stack));
cluster->nodeid = get_local_nodeid(0);
if(cluster->nodeid == 0) {
crm_err("Could not establish local nodeid");
return FALSE;
}
cluster->uname = get_node_name(0);
if(cluster->uname == NULL) {
crm_err("Could not establish local node name");
return FALSE;
}
/* Ensure the local node always exists */
peer = crm_get_peer(cluster->nodeid, cluster->uname);
uuid = get_corosync_uuid(peer);
if(uuid) {
cluster->uuid = strdup(uuid);
}
return TRUE;
}
static int crm_terminate_member_common(int nodeid, const char *uname, IPC_Channel *cluster, int *connection)
{
crm_node_t *node = NULL;
gboolean success = FALSE;
const char *reason = "Cluster connection failed";
node = crm_get_peer(nodeid, uname);
if(cluster == NULL) {
reason = "No connection to the cluster";
} else if(node == NULL) {
if(uname) {
crm_err("Nothing known about node uname=%s", uname);
} else if(nodeid > 0) {
crm_err("Nothing known about node id=%d", nodeid);
} else {
crm_err("A node id or uname is required, got %d/%p", nodeid, uname);
}
return -1;
} else {
time_t now = time(NULL);
char *now_s = crm_itoa(now);
if(cluster) {
success = attrd_update(cluster, 'U', node->uname, "terminate", now_s, XML_CIB_TAG_STATUS, NULL, NULL);
} else {
success = attrd_update_no_mainloop(connection, 'U', node->uname, "terminate", now_s, XML_CIB_TAG_STATUS, NULL, NULL);
}
crm_free(now_s);
}
if(success) {
crm_info("Requested that node %d/%s be terminated", nodeid, node->uname);
return 1;
}
crm_err("Could not terminate node %d/%s: %s", nodeid, node->uname, reason);
return 0;
}
void
join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
char *join_id = user_data;
xmlNode *generation = create_xml_node(NULL, XML_CIB_TAG_GENERATION_TUPPLE);
CRM_LOG_ASSERT(join_id != NULL);
if (query_call_id != call_id) {
crm_trace("Query %d superceeded", call_id);
goto done;
}
query_call_id = 0;
if(rc != pcmk_ok || output == NULL) {
crm_err("Could not retrieve version details for join-%s: %s (%d)",
join_id, pcmk_strerror(rc), rc);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
} else if (fsa_our_dc == NULL) {
crm_debug("Membership is in flux, not continuing join-%s", join_id);
} else {
xmlNode *reply = NULL;
crm_debug("Respond to join offer join-%s from %s", join_id, fsa_our_dc);
copy_in_properties(generation, output);
reply = create_request(CRM_OP_JOIN_REQUEST, generation, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add(reply, F_CRM_JOIN_ID, join_id);
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
}
done:
free_xml(generation);
free(join_id);
}
/*!
* \brief Start a new election by offering local node's candidacy
*
* Broadcast a "vote" election message containing the local node's ID,
* (incremented) election counter, and uptime, and start the election timer.
*
* \param[in] e Election object
* \note Any nodes agreeing to the candidacy will send a "no-vote" reply, and if
* all active peers do so, or if the election times out, the local node
* wins the election. (If we lose to any peer vote, we will stop the
* timer, so a timeout means we did not lose -- either some peer did not
* vote, or we did not call election_check() in time.)
*/
void
election_vote(election_t *e)
{
struct timeval age;
xmlNode *vote = NULL;
crm_node_t *our_node;
if (e == NULL) {
crm_trace("Election vote requested, but no election available");
return;
}
our_node = crm_get_peer(0, e->uname);
if ((our_node == NULL) || (crm_is_peer_active(our_node) == FALSE)) {
crm_trace("Cannot vote in %s yet: local node not connected to cluster",
e->name);
return;
}
election_reset(e);
e->state = election_in_progress;
vote = create_request(CRM_OP_VOTE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
e->count++;
crm_xml_add(vote, F_CRM_ELECTION_OWNER, our_node->uuid);
crm_xml_add_int(vote, F_CRM_ELECTION_ID, e->count);
crm_uptime(&age);
crm_xml_add_int(vote, F_CRM_ELECTION_AGE_S, age.tv_sec);
crm_xml_add_int(vote, F_CRM_ELECTION_AGE_US, age.tv_usec);
send_cluster_message(NULL, crm_msg_crmd, vote, TRUE);
free_xml(vote);
crm_debug("Started %s round %d", e->name, e->count);
election_timeout_start(e);
return;
}
/*!
* \internal
* \brief Notify the DC of a remote node state change
*
* \param[in] node_name Node's name
* \param[in] node_up TRUE if node is up, FALSE if down
*/
void
send_remote_state_message(const char *node_name, gboolean node_up)
{
/* If we don't have a DC, or the message fails, we have a failsafe:
* the DC will eventually pick up the change via the CIB node state.
* The message allows it to happen sooner if possible.
*/
if (fsa_our_dc) {
xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_info("Notifying DC %s of pacemaker_remote node %s %s",
fsa_our_dc, node_name, (node_up? "coming up" : "going down"));
crm_xml_add(msg, XML_ATTR_ID, node_name);
crm_xml_add_boolean(msg, XML_NODE_IN_CLUSTER, node_up);
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, msg,
TRUE);
free_xml(msg);
} else {
crm_debug("No DC to notify of pacemaker_remote node %s %s",
node_name, (node_up? "coming up" : "going down"));
}
}
static void
forward_request(xmlNode * request, crm_client_t * cib_client, int call_options)
{
const char *op = crm_element_value(request, F_CIB_OPERATION);
const char *host = crm_element_value(request, F_CIB_HOST);
crm_xml_add(request, F_CIB_DELEGATED, cib_our_uname);
if (host != NULL) {
crm_trace("Forwarding %s op to %s", op, host);
send_cluster_message(crm_get_peer(0, host), crm_msg_cib, request, FALSE);
} else {
crm_trace("Forwarding %s op to master instance", op);
send_cluster_message(NULL, crm_msg_cib, request, FALSE);
}
/* Return the request to its original state */
xml_remove_prop(request, F_CIB_DELEGATED);
if (call_options & cib_discard_reply) {
crm_trace("Client not interested in reply");
}
}
gboolean
cluster_connect_cpg(crm_cluster_t *cluster)
{
int rc = -1;
int fd = 0;
int retries = 0;
uint32_t id = 0;
crm_node_t *peer = NULL;
cpg_handle_t handle = 0;
struct mainloop_fd_callbacks cpg_fd_callbacks = {
.dispatch = pcmk_cpg_dispatch,
.destroy = cluster->destroy,
};
cpg_callbacks_t cpg_callbacks = {
.cpg_deliver_fn = cluster->cpg.cpg_deliver_fn,
.cpg_confchg_fn = cluster->cpg.cpg_confchg_fn,
/* .cpg_deliver_fn = pcmk_cpg_deliver, */
/* .cpg_confchg_fn = pcmk_cpg_membership, */
};
cpg_evicted = FALSE;
cluster->group.length = 0;
cluster->group.value[0] = 0;
/* group.value is char[128] */
strncpy(cluster->group.value, crm_system_name?crm_system_name:"unknown", 127);
cluster->group.value[127] = 0;
cluster->group.length = 1 + QB_MIN(127, strlen(cluster->group.value));
cs_repeat(retries, 30, rc = cpg_initialize(&handle, &cpg_callbacks));
if (rc != CS_OK) {
crm_err("Could not connect to the Cluster Process Group API: %d", rc);
goto bail;
}
id = get_local_nodeid(handle);
if (id == 0) {
crm_err("Could not get local node id from the CPG API");
goto bail;
}
cluster->nodeid = id;
retries = 0;
cs_repeat(retries, 30, rc = cpg_join(handle, &cluster->group));
if (rc != CS_OK) {
crm_err("Could not join the CPG group '%s': %d", crm_system_name, rc);
goto bail;
}
rc = cpg_fd_get(handle, &fd);
if (rc != CS_OK) {
crm_err("Could not obtain the CPG API connection: %d", rc);
goto bail;
}
pcmk_cpg_handle = handle;
cluster->cpg_handle = handle;
mainloop_add_fd("corosync-cpg", G_PRIORITY_MEDIUM, fd, cluster, &cpg_fd_callbacks);
bail:
if (rc != CS_OK) {
cpg_finalize(handle);
return FALSE;
}
peer = crm_get_peer(id, NULL);
crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, ONLINESTATUS);
return TRUE;
}
gboolean
send_cluster_message_cs(xmlNode * msg, gboolean local, crm_node_t * node, enum crm_ais_msg_types dest)
{
gboolean rc = TRUE;
char *data = NULL;
data = dump_xml_unformatted(msg);
rc = send_cluster_text(crm_class_cluster, data, local, node, dest);
free(data);
return rc;
}
gboolean
send_cluster_text(int class, const char *data,
gboolean local, crm_node_t * node, enum crm_ais_msg_types dest)
{
static int msg_id = 0;
static int local_pid = 0;
static int local_name_len = 0;
static const char *local_name = NULL;
char *target = NULL;
struct iovec *iov;
AIS_Message *msg = NULL;
enum crm_ais_msg_types sender = text2msg_type(crm_system_name);
/* There are only 6 handlers registered to crm_lib_service in plugin.c */
CRM_CHECK(class < 6, crm_err("Invalid message class: %d", class);
return FALSE);
#if !SUPPORT_PLUGIN
CRM_CHECK(dest != crm_msg_ais, return FALSE);
#endif
if(local_name == NULL) {
local_name = get_local_node_name();
}
if(local_name_len == 0 && local_name) {
local_name_len = strlen(local_name);
}
if (data == NULL) {
data = "";
}
if (local_pid == 0) {
local_pid = getpid();
}
if (sender == crm_msg_none) {
sender = local_pid;
}
msg = calloc(1, sizeof(AIS_Message));
msg_id++;
msg->id = msg_id;
msg->header.id = class;
msg->header.error = CS_OK;
msg->host.type = dest;
msg->host.local = local;
if (node) {
if (node->uname) {
target = strdup(node->uname);
msg->host.size = strlen(node->uname);
memset(msg->host.uname, 0, MAX_NAME);
memcpy(msg->host.uname, node->uname, msg->host.size);
} else {
target = crm_strdup_printf("%u", node->id);
}
msg->host.id = node->id;
} else {
target = strdup("all");
}
msg->sender.id = 0;
msg->sender.type = sender;
msg->sender.pid = local_pid;
msg->sender.size = local_name_len;
memset(msg->sender.uname, 0, MAX_NAME);
if(local_name && msg->sender.size) {
memcpy(msg->sender.uname, local_name, msg->sender.size);
}
msg->size = 1 + strlen(data);
msg->header.size = sizeof(AIS_Message) + msg->size;
if (msg->size < CRM_BZ2_THRESHOLD) {
msg = realloc_safe(msg, msg->header.size);
memcpy(msg->data, data, msg->size);
} else {
char *compressed = NULL;
unsigned int new_size = 0;
char *uncompressed = strdup(data);
if (crm_compress_string(uncompressed, msg->size, 0, &compressed, &new_size)) {
msg->header.size = sizeof(AIS_Message) + new_size;
msg = realloc_safe(msg, msg->header.size);
memcpy(msg->data, compressed, new_size);
msg->is_compressed = TRUE;
msg->compressed_size = new_size;
} else {
msg = realloc_safe(msg, msg->header.size);
memcpy(msg->data, data, msg->size);
}
free(uncompressed);
free(compressed);
}
iov = calloc(1, sizeof(struct iovec));
iov->iov_base = msg;
iov->iov_len = msg->header.size;
if (msg->compressed_size) {
crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes compressed payload): %.200s",
msg->id, target, (unsigned long long) iov->iov_len,
msg->compressed_size, data);
} else {
crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes payload): %.200s",
msg->id, target, (unsigned long long) iov->iov_len,
msg->size, data);
}
free(target);
#if SUPPORT_PLUGIN
/* The plugin is the only time we dont use CPG messaging */
if(get_cluster_type() == pcmk_cluster_classic_ais) {
return send_plugin_text(class, iov);
}
#endif
send_cpg_iov(iov);
return TRUE;
}
enum crm_ais_msg_types
text2msg_type(const char *text)
{
int type = crm_msg_none;
CRM_CHECK(text != NULL, return type);
if (safe_str_eq(text, "ais")) {
type = crm_msg_ais;
} else if (safe_str_eq(text, "crm_plugin")) {
type = crm_msg_ais;
} else if (safe_str_eq(text, CRM_SYSTEM_CIB)) {
type = crm_msg_cib;
} else if (safe_str_eq(text, CRM_SYSTEM_CRMD)) {
type = crm_msg_crmd;
} else if (safe_str_eq(text, CRM_SYSTEM_DC)) {
type = crm_msg_crmd;
} else if (safe_str_eq(text, CRM_SYSTEM_TENGINE)) {
type = crm_msg_te;
} else if (safe_str_eq(text, CRM_SYSTEM_PENGINE)) {
type = crm_msg_pe;
} else if (safe_str_eq(text, CRM_SYSTEM_LRMD)) {
type = crm_msg_lrmd;
} else if (safe_str_eq(text, CRM_SYSTEM_STONITHD)) {
type = crm_msg_stonithd;
} else if (safe_str_eq(text, "stonith-ng")) {
type = crm_msg_stonith_ng;
} else if (safe_str_eq(text, "attrd")) {
type = crm_msg_attrd;
} else {
/* This will normally be a transient client rather than
* a cluster daemon. Set the type to the pid of the client
*/
int scan_rc = sscanf(text, "%d", &type);
if (scan_rc != 1 || type <= crm_msg_stonith_ng) {
/* Ensure it's sane */
type = crm_msg_none;
}
}
return type;
}
void
pcmk_cpg_membership(cpg_handle_t handle,
const struct cpg_name *groupName,
const struct cpg_address *member_list, size_t member_list_entries,
const struct cpg_address *left_list, size_t left_list_entries,
const struct cpg_address *joined_list, size_t joined_list_entries)
{
int i;
gboolean found = FALSE;
static int counter = 0;
uint32_t local_nodeid = get_local_nodeid(handle);
for (i = 0; i < left_list_entries; i++) {
crm_node_t *peer = crm_find_peer(left_list[i].nodeid, NULL);
crm_info("Node %u left group %s (peer=%s, counter=%d.%d)",
left_list[i].nodeid, groupName->value,
(peer? peer->uname : "<none>"), counter, i);
if (peer) {
crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, OFFLINESTATUS);
}
}
for (i = 0; i < joined_list_entries; i++) {
crm_info("Node %u joined group %s (counter=%d.%d)",
joined_list[i].nodeid, groupName->value, counter, i);
}
for (i = 0; i < member_list_entries; i++) {
crm_node_t *peer = crm_get_peer(member_list[i].nodeid, NULL);
crm_info("Node %u still member of group %s (peer=%s, counter=%d.%d)",
member_list[i].nodeid, groupName->value,
(peer? peer->uname : "<none>"), counter, i);
/* Anyone that is sending us CPG messages must also be a _CPG_ member.
* But it's _not_ safe to assume it's in the quorum membership.
* We may have just found out it's dead and are processing the last couple of messages it sent
*/
peer = crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, ONLINESTATUS);
if(peer && peer->state && crm_is_peer_active(peer) == FALSE) {
time_t now = time(NULL);
/* Co-opt the otherwise unused votes field */
if(peer->votes == 0) {
peer->votes = now;
} else if(now > (60 + peer->votes)) {
/* On the otherhand, if we're still getting messages, at a certain point
* we need to acknowledge our internal cache is probably wrong
*
* Set the threshold to 1 minute
*/
crm_err("Node %s[%u] appears to be online even though we think it is dead", peer->uname, peer->id);
if (crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_MEMBER, 0)) {
peer->votes = 0;
}
}
}
if (local_nodeid == member_list[i].nodeid) {
found = TRUE;
}
}
if (!found) {
crm_err("We're not part of CPG group '%s' anymore!", groupName->value);
cpg_evicted = TRUE;
}
counter++;
}
char *
pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *content,
uint32_t *kind, const char **from)
{
char *data = NULL;
AIS_Message *msg = (AIS_Message *) content;
if(handle) {
/* 'msg' came from CPG not the plugin
* Do filtering and field massaging
*/
uint32_t local_nodeid = get_local_nodeid(handle);
const char *local_name = get_local_node_name();
if (msg->sender.id > 0 && msg->sender.id != nodeid) {
crm_err("Nodeid mismatch from %d.%d: claimed nodeid=%u", nodeid, pid, msg->sender.id);
return NULL;
} else if (msg->host.id != 0 && (local_nodeid != msg->host.id)) {
/* Not for us */
crm_trace("Not for us: %u != %u", msg->host.id, local_nodeid);
return NULL;
} else if (msg->host.size != 0 && safe_str_neq(msg->host.uname, local_name)) {
/* Not for us */
crm_trace("Not for us: %s != %s", msg->host.uname, local_name);
return NULL;
}
msg->sender.id = nodeid;
if (msg->sender.size == 0) {
crm_node_t *peer = crm_get_peer(nodeid, NULL);
if (peer == NULL) {
crm_err("Peer with nodeid=%u is unknown", nodeid);
} else if (peer->uname == NULL) {
crm_err("No uname for peer with nodeid=%u", nodeid);
} else {
crm_notice("Fixing uname for peer with nodeid=%u", nodeid);
msg->sender.size = strlen(peer->uname);
memset(msg->sender.uname, 0, MAX_NAME);
memcpy(msg->sender.uname, peer->uname, msg->sender.size);
}
}
}
crm_trace("Got new%s message (size=%d, %d, %d)",
msg->is_compressed ? " compressed" : "",
ais_data_len(msg), msg->size, msg->compressed_size);
if (kind != NULL) {
*kind = msg->header.id;
}
if (from != NULL) {
*from = msg->sender.uname;
}
if (msg->is_compressed && msg->size > 0) {
int rc = BZ_OK;
char *uncompressed = NULL;
unsigned int new_size = msg->size + 1;
if (check_message_sanity(msg, NULL) == FALSE) {
goto badmsg;
}
crm_trace("Decompressing message data");
uncompressed = calloc(1, new_size);
rc = BZ2_bzBuffToBuffDecompress(uncompressed, &new_size, msg->data, msg->compressed_size, 1, 0);
if (rc != BZ_OK) {
crm_err("Decompression failed: %d", rc);
free(uncompressed);
goto badmsg;
}
CRM_ASSERT(rc == BZ_OK);
CRM_ASSERT(new_size == msg->size);
data = uncompressed;
} else if (check_message_sanity(msg, data) == FALSE) {
goto badmsg;
} else if (safe_str_eq("identify", data)) {
int pid = getpid();
char *pid_s = crm_itoa(pid);
send_cluster_text(crm_class_cluster, pid_s, TRUE, NULL, crm_msg_ais);
free(pid_s);
return NULL;
} else {
data = strdup(msg->data);
}
if (msg->header.id != crm_class_members) {
/* Is this even needed anymore? */
crm_get_peer(msg->sender.id, msg->sender.uname);
}
if (msg->header.id == crm_class_rmpeer) {
uint32_t id = crm_int_helper(data, NULL);
crm_info("Removing peer %s/%u", data, id);
reap_crm_member(id, NULL);
free(data);
return NULL;
#if SUPPORT_PLUGIN
} else if (is_classic_ais_cluster()) {
plugin_handle_membership(msg);
#endif
}
crm_trace("Payload: %.200s", data);
return data;
badmsg:
crm_err("Invalid message (id=%d, dest=%s:%s, from=%s:%s.%d):"
" min=%d, total=%d, size=%d, bz2_size=%d",
msg->id, ais_dest(&(msg->host)), msg_type2text(msg->host.type),
ais_dest(&(msg->sender)), msg_type2text(msg->sender.type),
msg->sender.pid, (int)sizeof(AIS_Message),
msg->header.size, msg->size, msg->compressed_size);
free(data);
return NULL;
}
void
attrd_peer_update(crm_node_t *peer, xmlNode *xml, bool filter)
{
bool changed = FALSE;
attribute_value_t *v = NULL;
const char *host = crm_element_value(xml, F_ATTRD_HOST);
const char *attr = crm_element_value(xml, F_ATTRD_ATTRIBUTE);
const char *value = crm_element_value(xml, F_ATTRD_VALUE);
attribute_t *a = g_hash_table_lookup(attributes, attr);
if(a == NULL) {
a = create_attribute(xml);
}
v = g_hash_table_lookup(a->values, host);
if(v == NULL) {
crm_trace("Setting %s[%s] to %s from %s", attr, host, value, peer->uname);
v = calloc(1, sizeof(attribute_value_t));
if(value) {
v->current = strdup(value);
}
v->nodename = strdup(host);
crm_element_value_int(xml, F_ATTRD_IS_REMOTE, &v->is_remote);
g_hash_table_replace(a->values, v->nodename, v);
if (v->is_remote == TRUE) {
crm_remote_peer_cache_add(host);
}
changed = TRUE;
} else if(filter
&& safe_str_neq(v->current, value)
&& safe_str_eq(host, attrd_cluster->uname)) {
xmlNode *sync = create_xml_node(NULL, __FUNCTION__);
crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s",
a->id, host, v->current, value, peer->uname);
crm_xml_add(sync, F_ATTRD_TASK, "sync-response");
v = g_hash_table_lookup(a->values, host);
build_attribute_xml(sync, a->id, a->set, a->uuid, a->timeout_ms, a->user, v->nodename, v->nodeid, v->current);
crm_xml_add_int(sync, F_ATTRD_WRITER, election_state(writer));
send_cluster_message(peer, crm_msg_attrd, sync, TRUE);
free_xml(sync);
} else if(safe_str_neq(v->current, value)) {
crm_info("Setting %s[%s]: %s -> %s from %s", attr, host, v->current, value, peer->uname);
free(v->current);
if(value) {
v->current = strdup(value);
} else {
v->current = NULL;
}
changed = TRUE;
} else {
crm_trace("Unchanged %s[%s] from %s is %s", attr, host, peer->uname, value);
}
a->changed |= changed;
/* this only involves cluster nodes. */
if(v->nodeid == 0 && (v->is_remote == FALSE)) {
if(crm_element_value_int(xml, F_ATTRD_HOST_ID, (int*)&v->nodeid) == 0) {
/* Create the name/id association */
crm_node_t *peer = crm_get_peer(v->nodeid, host);
crm_trace("We know %s's node id now: %s", peer->uname, peer->uuid);
if(election_state(writer) == election_won) {
write_attributes(FALSE, TRUE);
return;
}
}
}
if(changed) {
if(a->timer) {
crm_trace("Delayed write out (%dms) for %s", a->timeout_ms, a->id);
mainloop_timer_start(a->timer);
} else {
write_or_elect_attribute(a);
}
}
}
gboolean
fail_incompletable_actions(crm_graph_t * graph, const char *down_node)
{
const char *target_uuid = NULL;
const char *router = NULL;
const char *router_uuid = NULL;
xmlNode *last_action = NULL;
GListPtr gIter = NULL;
GListPtr gIter2 = NULL;
if (graph == NULL || graph->complete) {
return FALSE;
}
gIter = graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
synapse_t *synapse = (synapse_t *) gIter->data;
if (synapse->confirmed) {
continue;
}
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
crm_action_t *action = (crm_action_t *) gIter2->data;
if (action->type == action_type_pseudo || action->confirmed) {
continue;
} else if (action->type == action_type_crm) {
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
if (safe_str_eq(task, CRM_OP_FENCE)) {
continue;
}
}
target_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
router = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE);
if (router) {
crm_node_t *node = crm_get_peer(0, router);
if (node) {
router_uuid = node->uuid;
}
}
if (safe_str_eq(target_uuid, down_node) || safe_str_eq(router_uuid, down_node)) {
action->failed = TRUE;
synapse->failed = TRUE;
last_action = action->xml;
stop_te_timer(action->timer);
update_graph(graph, action);
if (synapse->executed) {
crm_notice("Action %d (%s) was pending on %s (offline)",
action->id, ID(action->xml), down_node);
} else {
crm_notice("Action %d (%s) is scheduled for %s (offline)",
action->id, ID(action->xml), down_node);
}
}
}
}
if (last_action != NULL) {
crm_warn("Node %s shutdown resulted in un-runnable actions", down_node);
abort_transition(INFINITY, tg_restart, "Node failure", last_action);
return TRUE;
}
return FALSE;
}
/*
A_DC_JOIN_PROCESS_REQアクション処理(CRM_OP_JOIN_REQUESTをDCノードが受信した時の処理)
*/
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, XML_ATTR_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);
/* JOIN状態チェックを行う */
check_join_state(cur_state, __FUNCTION__);
return;
} else if(join_node == NULL || crm_is_member_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 = crm_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");
crm_free(max_generation_from);
free_xml(max_generation_xml);
max_generation_from = crm_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_err("join-%d: NACK'ing node %s (ref %s)",
join_id, join_from, ref);
} else {
crm_debug("join-%d: Welcoming node %s (ref %s)",
join_id, join_from, ref);
}
/* add them to our list of CRMD_STATE_ACTIVE nodes */
/* integrated_nodesハッシュテーブルにCRM_OP_JOIN_REQUESTを送信して来たノードを追加する */
g_hash_table_insert(
integrated_nodes, crm_strdup(join_from), crm_strdup(ack_nack));
/* integrated_nodesハッシュテーブルサイズと、CRM_OP_JOIN_REQUESTを送信して来たノードjoin_idをログ出力する */
crm_debug("%u nodes have been integrated into join-%d",
g_hash_table_size(integrated_nodes), join_id);
/* welcomed_nodesハッシュテーブルからCRM_OP_JOIN_REQUESTを送信して来たノードを削除する */
g_hash_table_remove(welcomed_nodes, join_from);
/* JOIN状態チェックを行う */
if(check_join_state(cur_state, __FUNCTION__) == FALSE) {
/* dont waste time by invoking the PE yet; */
crm_debug("join-%d: Still waiting on %d outstanding offers",
join_id, g_hash_table_size(welcomed_nodes));
}
}
/* 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("A new node joined - wait until it contacts us");
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);
if(member == NULL || crm_is_member_active(member) == FALSE) {
crm_err("Attempt to send welcome message "
"to a node not part of our partition!");
return;
}
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_debug("(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));
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
*/
start_transition(fsa_state);
/* dont waste time by invoking the pe yet; */
crm_debug("Waiting on %d outstanding join acks for join-%d",
g_hash_table_size(welcomed_nodes), current_join_id);
}
