void
notify_crmd(crm_graph_t * graph)
{
const char *type = "unknown";
enum crmd_fsa_input event = I_NULL;
crm_debug("Processing transition completion in state %s", fsa_state2string(fsa_state));
CRM_CHECK(graph->complete, graph->complete = TRUE);
switch (graph->completion_action) {
case tg_stop:
type = "stop";
/* fall through */
case tg_done:
type = "done";
if (fsa_state == S_TRANSITION_ENGINE) {
event = I_TE_SUCCESS;
}
break;
case tg_restart:
type = "restart";
if (fsa_state == S_TRANSITION_ENGINE) {
if (transition_timer->period_ms > 0) {
crm_timer_stop(transition_timer);
crm_timer_start(transition_timer);
} else if (too_many_st_failures() == FALSE) {
event = I_PE_CALC;
}
} else if (fsa_state == S_POLICY_ENGINE) {
register_fsa_action(A_PE_INVOKE);
}
break;
case tg_shutdown:
type = "shutdown";
if (is_set(fsa_input_register, R_SHUTDOWN)) {
event = I_STOP;
} else {
crm_err("We didn't ask to be shut down, yet our" " PE is telling us too.");
event = I_TERMINATE;
}
}
crm_debug("Transition %d status: %s - %s", graph->id, type, crm_str(graph->abort_reason));
graph->abort_reason = NULL;
graph->completion_action = tg_done;
clear_bit(fsa_input_register, R_IN_TRANSITION);
if (event != I_NULL) {
register_fsa_input(C_FSA_INTERNAL, event, NULL);
} else if (fsa_source) {
mainloop_set_trigger(fsa_source);
}
}
/*
A_ELECTION_CHECKアクション
CRM_OP_VOTEメッセージを受信してから流れる処理
voteハッシュテーブルに全てのアクティブなクラスタ構成メンバーからの受信をおこなったかどうかチェックし
全ての受信が行われている場合は、I_ELECTION_DCへ遷移する */
*/
void
do_election_check(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 voted_size = 0;
/* 現在のアクティブなメンバー数を取得する */
int num_members = crm_active_members();
if(voted) {
/* votedハッシュテーブルが存在する場合は、テーブルサイズを取得する */
voted_size = g_hash_table_size(voted);
}
/* in the case of #voted > #members, it is better to
* wait for the timeout and give the cluster time to
* stabilize
*/
if(fsa_state != S_ELECTION) {
/* S_ELECTION状態の場合はチェックしない */
crm_debug("Ignore election check: we not in an election");
} else if(voted_size >= num_members) {
/* voteハッシュテーブルのサイズ（メンバー数)が現在のアクティブなメンバー数以上の場合 */
/* we won and everyone has voted */
/* メンバーが揃ったのでelection_timeoutタイマーを止める */
crm_timer_stop(election_timeout);
/* アクティブメンバー数とvoteサイズが一致もしくは、voteサイズが大きい状態になったので、 */
/* メンバーは揃ったので、内部メッセージにI_ELECTION_DCをセットする */
register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL);
if(voted_size > num_members) {
char *data = NULL;
data = crm_strdup("member");
g_hash_table_foreach(crm_peer_cache, log_member_uname, data);
crm_free(data);
data = crm_strdup("voted");
g_hash_table_foreach(voted, log_node, data);
crm_free(data);
}
crm_debug("Destroying voted hash");
/* votedハッシュテーブルを破棄する */
g_hash_table_destroy(voted);
voted = NULL;
} else {
crm_debug("Still waiting on %d non-votes (%d total)",
num_members - voted_size, num_members);
}
return;
}
gboolean
crm_timer_popped(gpointer data)
{
fsa_timer_t *timer = (fsa_timer_t *) data;
if (timer == wait_timer
|| timer == recheck_timer
|| timer == transition_timer || timer == finalization_timer || timer == election_trigger) {
crm_info("%s (%s) just popped (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms);
timer->counter++;
} else {
crm_err("%s (%s) just popped in state %s! (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input),
fsa_state2string(fsa_state), timer->period_ms);
}
if (timer == election_trigger && election_trigger->counter > 5) {
crm_notice("We appear to be in an election loop, something may be wrong");
crm_write_blackbox(0, NULL);
election_trigger->counter = 0;
}
if (timer->repeat == FALSE) {
crm_timer_stop(timer); /* make it _not_ go off again */
}
if (timer->fsa_input == I_INTEGRATED) {
crm_info("Welcomed: %d, Integrated: %d",
crmd_join_phase_count(crm_join_welcomed),
crmd_join_phase_count(crm_join_integrated));
if (crmd_join_phase_count(crm_join_welcomed) == 0) {
/* If we don't even have ourself, start again */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __FUNCTION__);
} else {
register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL);
}
} else if (timer == recheck_timer && fsa_state != S_IDLE) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer == finalization_timer && fsa_state != S_FINALIZE_JOIN) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer->fsa_input != I_NULL) {
register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL);
}
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
return TRUE;
}
void
do_election_check(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 voted_size = 0;
int num_members = crm_active_peers();
if (voted) {
voted_size = g_hash_table_size(voted);
}
/* in the case of #voted > #members, it is better to
* wait for the timeout and give the cluster time to
* stabilize
*/
if (fsa_state != S_ELECTION) {
crm_debug("Ignore election check: we not in an election");
} else if (voted_size >= num_members) {
/* we won and everyone has voted */
crm_timer_stop(election_timeout);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL);
if (voted_size > num_members) {
GHashTableIter gIter;
const crm_node_t *node;
char *key = NULL;
g_hash_table_iter_init(&gIter, crm_peer_cache);
while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) & node)) {
if (crm_is_peer_active(node)) {
crm_err("member: %s proc=%.32x", node->uname, node->processes);
}
}
g_hash_table_iter_init(&gIter, voted);
while (g_hash_table_iter_next(&gIter, (gpointer *) & key, NULL)) {
crm_err("voted: %s", key);
}
}
crm_debug("Destroying voted hash");
g_hash_table_destroy(voted);
voted = NULL;
} else {
crm_debug("Still waiting on %d non-votes (%d total)",
num_members - voted_size, num_members);
}
return;
}
/* A_DC_TIMER_STOP, A_DC_TIMER_START,
* A_FINALIZE_TIMER_STOP, A_FINALIZE_TIMER_START
* A_INTEGRATE_TIMER_STOP, A_INTEGRATE_TIMER_START
*/
void
do_timer_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
gboolean timer_op_ok = TRUE;
if (action & A_DC_TIMER_STOP) {
timer_op_ok = crm_timer_stop(election_trigger);
} else if (action & A_FINALIZE_TIMER_STOP) {
timer_op_ok = crm_timer_stop(finalization_timer);
} else if (action & A_INTEGRATE_TIMER_STOP) {
timer_op_ok = crm_timer_stop(integration_timer);
/* } else if(action & A_ELECTION_TIMEOUT_STOP) { */
/* timer_op_ok = crm_timer_stop(election_timeout); */
}
/* dont start a timer that wasnt already running */
if (action & A_DC_TIMER_START && timer_op_ok) {
crm_timer_start(election_trigger);
if (AM_I_DC) {
/* there can be only one */
register_fsa_input(cause, I_ELECTION, NULL);
}
} else if (action & A_FINALIZE_TIMER_START) {
crm_timer_start(finalization_timer);
} else if (action & A_INTEGRATE_TIMER_START) {
crm_timer_start(integration_timer);
/* } else if(action & A_ELECTION_TIMEOUT_START) { */
/* crm_timer_start(election_timeout); */
}
}
gboolean
crm_timer_popped(gpointer data)
{
fsa_timer_t *timer = (fsa_timer_t *) data;
if (timer == wait_timer
|| timer == recheck_timer
|| timer == transition_timer || timer == finalization_timer || timer == election_trigger) {
crm_info("%s (%s) just popped (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms);
} else {
crm_err("%s (%s) just popped in state %s! (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input),
fsa_state2string(fsa_state), timer->period_ms);
}
if (timer->repeat == FALSE) {
crm_timer_stop(timer); /* make it _not_ go off again */
}
if (timer->fsa_input == I_INTEGRATED) {
crm_info("Welcomed: %d, Integrated: %d",
g_hash_table_size(welcomed_nodes), g_hash_table_size(integrated_nodes));
if (g_hash_table_size(welcomed_nodes) == 0) {
/* If we don't even have ourself, start again */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __FUNCTION__);
} else {
register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL);
}
} else if (timer == recheck_timer && fsa_state != S_IDLE) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer == finalization_timer && fsa_state != S_FINALIZE_JOIN) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer->fsa_input != I_NULL) {
register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL);
}
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
return TRUE;
}
void
do_election_check(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 voted_size = 0;
int num_members = crm_active_peers();
if (voted) {
voted_size = g_hash_table_size(voted);
}
/* in the case of #voted > #members, it is better to
* wait for the timeout and give the cluster time to
* stabilize
*/
if (fsa_state != S_ELECTION) {
crm_debug("Ignore election check: we not in an election");
} else if (voted_size >= num_members) {
/* we won and everyone has voted */
crm_timer_stop(election_timeout);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL);
if (voted_size > num_members) {
char *data = NULL;
data = strdup("member");
g_hash_table_foreach(crm_peer_cache, log_member_uname, data);
free(data);
data = strdup("voted");
g_hash_table_foreach(voted, log_node, data);
free(data);
}
crm_debug("Destroying voted hash");
g_hash_table_destroy(voted);
voted = NULL;
} else {
crm_debug("Still waiting on %d non-votes (%d total)",
num_members - voted_size, num_members);
}
return;
}
void
abort_transition_graph(int abort_priority, enum transition_action abort_action,
const char *abort_text, xmlNode * reason, const char *fn, int line)
{
int add[] = { 0, 0, 0 };
int del[] = { 0, 0, 0 };
int level = LOG_INFO;
xmlNode *diff = NULL;
xmlNode *change = NULL;
CRM_CHECK(transition_graph != NULL, return);
switch (fsa_state) {
case S_STARTING:
case S_PENDING:
case S_NOT_DC:
case S_HALT:
case S_ILLEGAL:
case S_STOPPING:
case S_TERMINATE:
crm_info("Abort %s suppressed: state=%s (complete=%d)",
abort_text, fsa_state2string(fsa_state), transition_graph->complete);
return;
default:
break;
}
/* Make sure any queued calculations are discarded ASAP */
free(fsa_pe_ref);
fsa_pe_ref = NULL;
if (transition_graph->complete == FALSE) {
if(update_abort_priority(transition_graph, abort_priority, abort_action, abort_text)) {
level = LOG_NOTICE;
}
}
if(reason) {
xmlNode *search = NULL;
for(search = reason; search; search = search->parent) {
if (safe_str_eq(XML_TAG_DIFF, TYPE(search))) {
diff = search;
break;
}
}
if(diff) {
xml_patch_versions(diff, add, del);
for(search = reason; search; search = search->parent) {
if (safe_str_eq(XML_DIFF_CHANGE, TYPE(search))) {
change = search;
break;
}
}
}
}
if(reason == NULL) {
do_crm_log(level, "Transition aborted: %s (source=%s:%d, %d)",
abort_text, fn, line, transition_graph->complete);
} else if(change == NULL) {
char *local_path = xml_get_path(reason);
do_crm_log(level, "Transition aborted by %s.%s: %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)",
TYPE(reason), ID(reason), abort_text, add[0], add[1], add[2], fn, line, local_path, transition_graph->complete);
free(local_path);
} else {
const char *kind = NULL;
const char *op = crm_element_value(change, XML_DIFF_OP);
const char *path = crm_element_value(change, XML_DIFF_PATH);
if(change == reason) {
if(strcmp(op, "create") == 0) {
reason = reason->children;
} else if(strcmp(op, "modify") == 0) {
reason = first_named_child(reason, XML_DIFF_RESULT);
if(reason) {
reason = reason->children;
}
}
}
kind = TYPE(reason);
if(strcmp(op, "delete") == 0) {
const char *shortpath = strrchr(path, '/');
do_crm_log(level, "Transition aborted by deletion of %s: %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)",
shortpath?shortpath+1:path, abort_text, add[0], add[1], add[2], fn, line, path, transition_graph->complete);
} else if (safe_str_eq(XML_CIB_TAG_NVPAIR, kind)) {
do_crm_log(level, "Transition aborted by %s, %s=%s: %s (%s cib=%d.%d.%d, source=%s:%d, path=%s, %d)",
crm_element_value(reason, XML_ATTR_ID),
crm_element_value(reason, XML_NVPAIR_ATTR_NAME),
crm_element_value(reason, XML_NVPAIR_ATTR_VALUE),
abort_text, op, add[0], add[1], add[2], fn, line, path, transition_graph->complete);
} else if (safe_str_eq(XML_LRM_TAG_RSC_OP, kind)) {
const char *magic = crm_element_value(reason, XML_ATTR_TRANSITION_MAGIC);
do_crm_log(level, "Transition aborted by %s '%s' on %s: %s (magic=%s, cib=%d.%d.%d, source=%s:%d, %d)",
crm_element_value(reason, XML_LRM_ATTR_TASK_KEY), op,
crm_element_value(reason, XML_LRM_ATTR_TARGET), abort_text,
magic, add[0], add[1], add[2], fn, line, transition_graph->complete);
} else if (safe_str_eq(XML_CIB_TAG_STATE, kind)
|| safe_str_eq(XML_CIB_TAG_NODE, kind)) {
const char *uname = crm_peer_uname(ID(reason));
do_crm_log(level, "Transition aborted by %s '%s' on %s: %s (cib=%d.%d.%d, source=%s:%d, %d)",
kind, op, uname ? uname : ID(reason), abort_text,
add[0], add[1], add[2], fn, line, transition_graph->complete);
} else {
do_crm_log(level, "Transition aborted by %s.%s '%s': %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)",
TYPE(reason), ID(reason), op?op:"change", abort_text, add[0], add[1], add[2], fn, line, path, transition_graph->complete);
}
}
if (transition_graph->complete) {
if (transition_timer->period_ms > 0) {
crm_timer_stop(transition_timer);
crm_timer_start(transition_timer);
} else {
register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL);
}
return;
}
mainloop_set_trigger(transition_trigger);
}
/* A_ELECTION_COUNT */
void
do_election_count_vote(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)
{
struct timeval your_age;
int age;
int election_id = -1;
int log_level = LOG_INFO;
gboolean use_born_on = FALSE;
gboolean done = FALSE;
gboolean we_loose = FALSE;
const char *op = NULL;
const char *vote_from = NULL;
const char *your_version = NULL;
const char *election_owner = NULL;
const char *reason = "unknown";
crm_node_t *our_node = NULL, *your_node = NULL;
ha_msg_input_t *vote = fsa_typed_data(fsa_dt_ha_msg);
static time_t last_election_loss = 0;
/* if the membership copy is NULL we REALLY shouldnt be voting
* the question is how we managed to get here.
*/
CRM_CHECK(msg_data != NULL, return);
CRM_CHECK(crm_peer_cache != NULL, return);
CRM_CHECK(vote != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
CRM_CHECK(vote->msg != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
your_age.tv_sec = 0;
your_age.tv_usec = 0;
op = crm_element_value(vote->msg, F_CRM_TASK);
vote_from = crm_element_value(vote->msg, F_CRM_HOST_FROM);
your_version = crm_element_value(vote->msg, F_CRM_VERSION);
election_owner = crm_element_value(vote->msg, F_CRM_ELECTION_OWNER);
crm_element_value_int(vote->msg, F_CRM_ELECTION_ID, &election_id);
crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_S, (int *)&(your_age.tv_sec));
crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_US, (int *)&(your_age.tv_usec));
CRM_CHECK(vote_from != NULL, vote_from = fsa_our_uname);
your_node = crm_get_peer(0, vote_from);
our_node = crm_get_peer(0, fsa_our_uname);
if (voted == NULL) {
crm_debug("Created voted hash");
voted = g_hash_table_new_full(crm_str_hash, g_str_equal,
g_hash_destroy_str, g_hash_destroy_str);
}
if (is_heartbeat_cluster()) {
use_born_on = TRUE;
} else if (is_classic_ais_cluster()) {
use_born_on = TRUE;
}
age = crm_compare_age(your_age);
if (cur_state == S_STARTING) {
reason = "Still starting";
we_loose = TRUE;
} else if (our_node == NULL || crm_is_peer_active(our_node) == FALSE) {
reason = "We are not part of the cluster";
log_level = LOG_ERR;
we_loose = TRUE;
} else if (election_id != current_election_id && crm_str_eq(fsa_our_uuid, election_owner, TRUE)) {
log_level = LOG_DEBUG_2;
reason = "Superceeded";
done = TRUE;
} else if (your_node == NULL || crm_is_peer_active(your_node) == FALSE) {
/* Possibly we cached the message in the FSA queue at a point that it wasn't */
reason = "Peer is not part of our cluster";
log_level = LOG_WARNING;
done = TRUE;
} else if (crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update the list of nodes that have voted */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if (crm_str_eq(vote_from, fsa_our_uname, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update ourselves in the list of nodes that have voted */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) < 0) {
reason = "Version";
we_loose = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) > 0) {
reason = "Version";
} else if (age < 0) {
reason = "Uptime";
we_loose = TRUE;
} else if (age > 0) {
reason = "Uptime";
/* TODO: Check for y(our) born < 0 */
} else if (use_born_on && your_node->born < our_node->born) {
reason = "Born";
we_loose = TRUE;
} else if (use_born_on && your_node->born > our_node->born) {
reason = "Born";
} else if (fsa_our_uname == NULL) {
reason = "Unknown host name";
we_loose = TRUE;
} else if (strcasecmp(fsa_our_uname, vote_from) > 0) {
reason = "Host name";
we_loose = TRUE;
} else {
reason = "Host name";
CRM_ASSERT(strcmp(fsa_our_uname, vote_from) != 0);
/* cant happen...
* } else if(strcasecmp(fsa_our_uname, vote_from) == 0) {
*
* default...
* } else { // strcasecmp(fsa_our_uname, vote_from) < 0
* we win
*/
}
if (done) {
do_crm_log(log_level + 1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)",
election_id, current_election_id, election_owner, op, vote_from, reason);
} else if (we_loose) {
xmlNode *novote = create_request(CRM_OP_NOVOTE, NULL, vote_from,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
do_crm_log(log_level, "Election %d (owner: %s) lost: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
update_dc(NULL);
crm_timer_stop(election_timeout);
if (fsa_input_register & R_THE_DC) {
crm_trace("Give up the DC to %s", vote_from);
register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL);
} else if (cur_state != S_STARTING) {
crm_trace("We werent the DC anyway");
register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL);
}
crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner);
crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id);
send_cluster_message(crm_get_peer(0, vote_from), crm_msg_crmd, novote, TRUE);
free_xml(novote);
fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local);
last_election_loss = time(NULL);
} else {
do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
if (last_election_loss) {
time_t tm_now = time(NULL);
if (tm_now - last_election_loss < (time_t) loss_dampen) {
crm_info("Election %d ignore: We already lost an election less than %ds ago (%s)",
election_id, loss_dampen, ctime(&last_election_loss));
update_dc(NULL);
return;
}
last_election_loss = 0;
}
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
g_hash_table_destroy(voted);
voted = NULL;
}
}
void
abort_transition_graph(int abort_priority, enum transition_action abort_action,
const char *abort_text, xmlNode * reason, const char *fn, int line)
{
const char *magic = NULL;
CRM_CHECK(transition_graph != NULL, return);
if (reason) {
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;
xmlNode *diff = get_xpath_object("//" F_CIB_UPDATE_RESULT "//diff", reason, LOG_DEBUG_2);
magic = crm_element_value(reason, XML_ATTR_TRANSITION_MAGIC);
if (diff) {
cib_diff_version_details(diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
if (crm_str_eq(TYPE(reason), XML_CIB_TAG_NVPAIR, TRUE)) {
crm_info
("%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, name=%s, value=%s, magic=%s, cib=%d.%d.%d) : %s",
fn, line, transition_graph->complete, TYPE(reason), ID(reason), NAME(reason),
VALUE(reason), magic ? magic : "NA", diff_add_admin_epoch, diff_add_epoch,
diff_add_updates, abort_text);
} else {
crm_info
("%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, magic=%s, cib=%d.%d.%d) : %s",
fn, line, transition_graph->complete, TYPE(reason), ID(reason),
magic ? magic : "NA", diff_add_admin_epoch, diff_add_epoch, diff_add_updates,
abort_text);
}
} else {
crm_info
("%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, magic=%s) : %s",
fn, line, transition_graph->complete, TYPE(reason), ID(reason),
magic ? magic : "NA", abort_text);
}
} else {
crm_info("%s:%d - Triggered transition abort (complete=%d) : %s",
fn, line, transition_graph->complete, abort_text);
}
switch (fsa_state) {
case S_STARTING:
case S_PENDING:
case S_NOT_DC:
case S_HALT:
case S_ILLEGAL:
case S_STOPPING:
case S_TERMINATE:
crm_info("Abort suppressed: state=%s (complete=%d)",
fsa_state2string(fsa_state), transition_graph->complete);
return;
default:
break;
}
if (magic == NULL && reason != NULL) {
crm_log_xml_debug(reason, "Cause");
}
/* Make sure any queued calculations are discarded ASAP */
free(fsa_pe_ref);
fsa_pe_ref = NULL;
if (transition_graph->complete) {
if (transition_timer->period_ms > 0) {
crm_timer_stop(transition_timer);
crm_timer_start(transition_timer);
} else if (too_many_st_failures() == FALSE) {
register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL);
} else {
register_fsa_input(C_FSA_INTERNAL, I_TE_SUCCESS, NULL);
}
return;
}
update_abort_priority(transition_graph, abort_priority, abort_action, abort_text);
mainloop_set_trigger(transition_trigger);
}
void
do_election_count_vote(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 election_id = -1;
int log_level = LOG_INFO;
gboolean done = FALSE;
gboolean we_loose = FALSE;
const char *op = NULL;
const char *vote_from = NULL;
const char *your_version = NULL;
const char *election_owner = NULL;
const char *reason = "unknown";
crm_node_t *our_node = NULL, *your_node = NULL;
ha_msg_input_t *vote = fsa_typed_data(fsa_dt_ha_msg);
static time_t last_election_win = 0;
static time_t last_election_loss = 0;
/* if the membership copy is NULL we REALLY shouldnt be voting
* the question is how we managed to get here.
*/
CRM_CHECK(msg_data != NULL, return);
CRM_CHECK(crm_peer_cache != NULL, return);
CRM_CHECK(vote != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
CRM_CHECK(vote->msg != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
/* 受信メッセージデータを取り出す */
op = crm_element_value(vote->msg, F_CRM_TASK);
vote_from = crm_element_value(vote->msg, F_CRM_HOST_FROM);
your_version = crm_element_value(vote->msg, F_CRM_VERSION);
election_owner = crm_element_value(vote->msg, F_CRM_ELECTION_OWNER);
crm_element_value_int(vote->msg, F_CRM_ELECTION_ID, &election_id);
CRM_CHECK(vote_from != NULL, vote_from = fsa_our_uname);
/* CRM_OP_VOTEメッセージの送信元のノード情報を取得する */
your_node = crm_get_peer(0, vote_from);
/* 自ノードのノード情報を取得する */
our_node = crm_get_peer(0, fsa_our_uname);
if(voted == NULL) {
crm_debug("Created voted hash");
/* votedハッシュテーブルが未作成の場合は作成する */
voted = g_hash_table_new_full(
g_str_hash, g_str_equal,
g_hash_destroy_str, g_hash_destroy_str);
}
if(cur_state == S_STARTING) {
/* 自ノードの状態が、まだ、S_STARTING状態の場合は、DCになれないのでCRM_OP_NOVOTEメッセージを送信する */
reason = "Still starting";
we_loose = TRUE;
} else if(our_node == NULL || crm_is_member_active(our_node) == FALSE) {
/* 自ノードがまだクラスタ構成として認識されていないか、アクティブでない場合は */
/* DCになれないのでCRM_OP_NOVOTEメッセージを送信する */
reason = "We are not part of the cluster";
log_level = LOG_ERR;
we_loose = TRUE;
} else if(your_node == NULL || crm_is_member_active(your_node) == FALSE) {
/* CRM_OP_VOTEメッセージの送信元のノードがクラスタ構成として認識されていないか、アクティブでない場合は */
/* ログのみを出力する */
reason = "Peer is not part of our cluster";
log_level = LOG_WARNING;
done = TRUE;
} else if(election_id != current_election_id
&& crm_str_eq(fsa_our_uuid, election_owner, TRUE)) {
/* 現在のelection_idと受信したelection_idが違う場合も、ログのみ出力する */
log_level = LOG_DEBUG_2;
reason = "Superceeded";
done = TRUE;
} else if(crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) {
/* DCになれないと思ったノードが送信したCRM_OP_NOVOTEメッセージの場合 */
char *op_copy = crm_strdup(op);
char *uname_copy = crm_strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update the list of nodes that have voted */
/* votedハッシュテーブルにノードデータをセットする */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if(crm_str_eq(vote_from, fsa_our_uname, TRUE)) {
/* 自ノードが送信したCRM_OP_VOTEメッセージを処理する場合 */
char *op_copy = crm_strdup(op);
char *uname_copy = crm_strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update ourselves in the list of nodes that have voted */
/* votedハッシュテーブルにノードデータをセットする */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if(compare_version(your_version, CRM_FEATURE_SET) < 0) {
/* 受信したCRM_OP_VOTEメッセージの送り元のversionがCRM_FEATURE_SETよりも小さい場合 */
/* 自ノードは、DCになれない */
reason = "Version";
we_loose = TRUE;
} else if(compare_version(your_version, CRM_FEATURE_SET) > 0) {
/* 受信したCRM_OP_VOTEメッセージの送り元のversionがCRM_FEATURE_SETよりも大きい場合、ログのみ出力する */
reason = "Version";
} else if(your_node->born < our_node->born) {
/* 受信したCRM_OP_VOTEメッセージの送り元の方がbornが自ノードよりも小さい場合 */
reason = "Age";
/* 自ノードは、DCになれない */
we_loose = TRUE;
} else if(your_node->born > our_node->born) {
/* 受信したCRM_OP_VOTEメッセージの送り元の方がbornが自ノードよりも大きい場合 */
/* 自ノードは、DCの候補 */
reason = "Age";
} else if(fsa_our_uname == NULL) {
/* 自ノードのノード名称がセットされていない場合 */
/* 自ノードは、DCになれない */
reason = "Unknown host name";
we_loose = TRUE;
} else if(strcasecmp(fsa_our_uname, vote_from) > 0) {
/* 自ノードのノード名が送信元...*/
/* 自ノードは、DCになれない */
reason = "Host name";
we_loose = TRUE;
} else {
/* その他の場合 */
reason = "Host name";
CRM_ASSERT(strcmp(fsa_our_uname, vote_from) != 0);
/* cant happen...
* } else if(strcasecmp(fsa_our_uname, vote_from) == 0) {
*
* default...
* } else { // strcasecmp(fsa_our_uname, vote_from) < 0
* we win
*/
}
if(done) {
do_crm_log(log_level+1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)",
election_id, current_election_id, election_owner, op, vote_from, reason);
} else if(we_loose) {
/* born値の比較などから、DCノードになれないと判断した場合 */
/* CRM_OP_NOVOTEメッセージをCRMD宛に生成する */
xmlNode *novote = create_request(
CRM_OP_NOVOTE, NULL, vote_from,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
do_crm_log(log_level+1, "Election %d (owner: %s) lost: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
/* DCノードにNULLをセットする */
update_dc(NULL);
/* election_timeoutタイマーを止める */
crm_timer_stop(election_timeout);
if(fsa_input_register & R_THE_DC) {
crm_debug_3("Give up the DC to %s", vote_from);
register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL);
} else if(cur_state != S_STARTING) {
crm_debug_3("We werent the DC anyway");
register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL);
}
/* CRM_OP_NOVOTEメッセージのF_CRM_ELECTION_OWNERに受信メッセージのelection_ownerをセットする */
crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner);
/* CRM_OP_NOVOTEメッセージのF_CRM_ELECTION_IDに受信メッセージのelection_idをセットする */
crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id);
/* CRM_OP_VOTEメッセージの送信元にCRM_OP_NOVOTEメッセージを送信する */
send_cluster_message(vote_from, crm_msg_crmd, novote, TRUE);
/* 送信メッセージを解放する */
free_xml(novote);
/* CIBのset_slave処理を実行する */
fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local);
last_election_loss = time(NULL);
last_election_win = 0;
} else {
do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
if(last_election_loss) {
time_t tm_now = time(NULL);
if(tm_now - last_election_loss < (time_t)loss_dampen) {
crm_info("Election %d ignore: We already lost an election less than %ds ago",
election_id, loss_dampen);
update_dc(NULL);
return;
}
last_election_loss = 0;
}
#if 0
/* Enabling this code can lead to multiple DCs during SimulStart.
* Specifically when a node comes up after our last 'win' vote.
*
* Fixing and enabling this functionality might become important when
* we start running realy big clusters, but for now leave it disabled.
*/
if(last_election_win) {
time_t tm_now = time(NULL);
if(tm_now - last_election_win < (time_t)win_dampen) {
crm_info("Election %d ignore: We already won an election less than %ds ago",
election_id, win_dampen);
return;
}
}
last_election_win = time(NULL);
#endif
/* I_ELECTIONへ */
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
g_hash_table_destroy(voted);
voted = NULL;
}
}
