Skip to content

Commit

Permalink
Feature: pacemakerd: keep tracking pacemakerd for liveness
Browse files Browse the repository at this point in the history
  • Loading branch information
@wenningerk
wenningerk committed Nov 18, 2021
1 parent 231cac4 commit 3f4fc68
Show file tree
Hide file tree
Showing 4 changed files with 40 additions and 48 deletions.
2 changes: 2 additions & 0 deletions daemons/pacemakerd/pacemakerd.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,8 @@ main(int argc, char **argv)
pcmk_ipc_api_t *old_instance = NULL;
qb_ipcs_service_t *ipcs = NULL;

subdaemon_check_progress = time(NULL);

crm_log_preinit(NULL, argc, argv);
mainloop_add_signal(SIGHUP, pcmk_ignore);
mainloop_add_signal(SIGQUIT, pcmk_sigquit);
Expand Down
1 change: 1 addition & 0 deletions daemons/pacemakerd/pacemakerd.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@ extern unsigned int shutdown_complete_state_reported_to;
extern gboolean shutdown_complete_state_reported_client_closed;
extern crm_trigger_t *shutdown_trigger;
extern crm_trigger_t *startup_trigger;
extern time_t subdaemon_check_progress;

gboolean mcp_read_config(void);

Expand Down
4 changes: 2 additions & 2 deletions daemons/pacemakerd/pcmkd_messages.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ pcmk_handle_ping_request(pcmk__client_t *c, xmlNode *msg, uint32_t id)
const char *value = NULL;
xmlNode *ping = NULL;
xmlNode *reply = NULL;
time_t pinged = time(NULL);
const char *from = crm_element_value(msg, F_CRM_SYS_FROM);

/* Pinged for status */
Expand All @@ -36,7 +35,8 @@ pcmk_handle_ping_request(pcmk__client_t *c, xmlNode *msg, uint32_t id)
value = crm_element_value(msg, F_CRM_SYS_TO);
crm_xml_add(ping, XML_PING_ATTR_SYSFROM, value);
crm_xml_add(ping, XML_PING_ATTR_PACEMAKERDSTATE, pacemakerd_state);
crm_xml_add_ll(ping, XML_ATTR_TSTAMP, (long long) pinged);
crm_xml_add_ll(ping, XML_ATTR_TSTAMP,
(long long) subdaemon_check_progress);
crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok");
reply = create_reply(msg, ping);
free_xml(ping);
Expand Down
81 changes: 35 additions & 46 deletions daemons/pacemakerd/pcmkd_subdaemons.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@ typedef struct pcmk_child_s {
bool active_before_startup;
} pcmk_child_t;

#define PCMK_PROCESS_CHECK_INTERVAL 5
#define PCMK_PROCESS_CHECK_INTERVAL 1
#define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */

/* Index into the array below */
Expand Down Expand Up @@ -82,6 +82,7 @@ static char *opts_vgrind[] = { NULL, NULL, NULL, NULL, NULL };

crm_trigger_t *shutdown_trigger = NULL;
crm_trigger_t *startup_trigger = NULL;
time_t subdaemon_check_progress;

/* When contacted via pacemakerd-api by a client having sbd in
* the name we assume it is sbd-daemon which wants to know
Expand All @@ -103,7 +104,6 @@ gboolean running_with_sbd = FALSE; /* local copy */
GMainLoop *mainloop = NULL;

static gboolean fatal_error = FALSE;
static bool global_keep_tracking = false;

static gboolean check_active_before_startup_processes(gpointer user_data);
static int child_liveness(pcmk_child_t *child);
Expand All @@ -127,44 +127,42 @@ pcmkd_cluster_connected(void)
static gboolean
check_active_before_startup_processes(gpointer user_data)
{
gboolean keep_tracking = FALSE;
static int next_child = 0;
int rc = child_liveness(&pcmk_children[next_child]);

for (int i = 0; i < PCMK__NELEM(pcmk_children); i++) {
if (!pcmk_children[i].active_before_startup) {
/* we are already tracking it as a child process. */
continue;
} else {
int rc = child_liveness(&pcmk_children[i]);

switch (rc) {
case pcmk_rc_ok:
break;
case pcmk_rc_ipc_unresponsive:
case pcmk_rc_ipc_pid_only: // This case: it was previously OK
if (pcmk_children[i].respawn) {
crm_err("%s[%lld] terminated%s", pcmk_children[i].name,
(long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[i].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
} else {
/* orderly shutdown */
crm_notice("%s[%lld] terminated%s", pcmk_children[i].name,
(long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[i].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
}
pcmk_process_exit(&(pcmk_children[i]));
continue;
default:
crm_exit(CRM_EX_FATAL);
break; /* static analysis/noreturn */
switch (rc) {
case pcmk_rc_ok:
next_child++;
subdaemon_check_progress = time(NULL);
break;
case pcmk_rc_ipc_unresponsive:
case pcmk_rc_ipc_pid_only: // This case: it was previously OK
if (pcmk_children[next_child].respawn) {
crm_err("%s[%lld] terminated%s",
pcmk_children[next_child].name,
(long long) PCMK__SPECIAL_PID_AS_0(
pcmk_children[next_child].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
} else {
/* orderly shutdown */
crm_notice("%s[%lld] terminated%s",
pcmk_children[next_child].name,
(long long) PCMK__SPECIAL_PID_AS_0(
pcmk_children[next_child].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
}
}
/* at least one of the processes found at startup
* is still going, so keep this recurring timer around */
keep_tracking = TRUE;
pcmk_process_exit(&(pcmk_children[next_child]));
break;
default:
crm_exit(CRM_EX_FATAL);
break; /* static analysis/noreturn */
}

global_keep_tracking = keep_tracking;
return keep_tracking;
if (next_child >= PCMK__NELEM(pcmk_children)) {
next_child = 0;
}

return true;
}

static gboolean
Expand Down Expand Up @@ -257,11 +255,6 @@ pcmk_process_exit(pcmk_child_t * child)
child->name, child->endpoint);
/* need to monitor how it evolves, and start new process if badly */
child->active_before_startup = true;
if (!global_keep_tracking) {
global_keep_tracking = true;
g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL,
check_active_before_startup_processes, NULL);
}

} else {
if (child->needs_cluster && !pcmkd_cluster_connected()) {
Expand Down Expand Up @@ -647,7 +640,6 @@ child_liveness(pcmk_child_t *child)
int
find_and_track_existing_processes(void)
{
bool tracking = false;
bool wait_in_progress;
int rc;
size_t i, rounds;
Expand Down Expand Up @@ -715,7 +707,6 @@ find_and_track_existing_processes(void)
pcmk_children[i].pid));
pcmk_children[i].respawn_count = -1; /* 0~keep watching */
pcmk_children[i].active_before_startup = true;
tracking = true;
break;
case pcmk_rc_ipc_pid_only:
if (pcmk_children[i].respawn_count == WAIT_TRIES) {
Expand Down Expand Up @@ -750,10 +741,8 @@ find_and_track_existing_processes(void)
pcmk_children[i].respawn_count = 0; /* restore pristine state */
}

if (tracking) {
g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL,
g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL,
check_active_before_startup_processes, NULL);
}
return pcmk_rc_ok;
}

Expand Down

0 comments on commit 3f4fc68

Please sign in to comment.