int
servant_cluster(const char *diskname, int mode, const void* argp)
{
enum cluster_type_e cluster_stack = get_cluster_type();
crm_system_name = strdup("sbd:cluster");
cl_log(LOG_INFO, "Monitoring %s cluster health", name_for_cluster_type(cluster_stack));
set_proc_title("sbd: watcher: Cluster");
sbd_membership_connect();
/* stonith_our_uname = cluster.uname; */
/* stonith_our_uuid = cluster.uuid; */
mainloop = g_main_new(FALSE);
notify_timer = g_timeout_add(timeout_loop * 1000, notify_timer_cb, NULL);
mainloop_add_signal(SIGTERM, cluster_shutdown);
mainloop_add_signal(SIGINT, cluster_shutdown);
g_main_run(mainloop);
g_main_destroy(mainloop);
clean_up(0);
return 0; /* never reached */
}
static void
sbd_membership_destroy(gpointer user_data)
{
cl_log(LOG_WARNING, "Lost connection to %s", name_for_cluster_type(get_cluster_type()));
set_servant_health(pcmk_health_unclean, LOG_ERR, "Cluster connection terminated");
notify_parent();
/* Attempt to reconnect, the watchdog will take the node down if the problem isn't transient */
sbd_membership_connect();
}
static void
sbd_membership_destroy(gpointer user_data)
{
cl_log(LOG_WARNING, "Lost connection to %s", name_for_cluster_type(get_cluster_type()));
if (get_cluster_type() != pcmk_cluster_unknown) {
#if SUPPORT_COROSYNC && CHECK_TWO_NODE
cmap_destroy();
#endif
}
set_servant_health(pcmk_health_unclean, LOG_ERR, "Cluster connection terminated");
notify_parent();
/* Attempt to reconnect, the watchdog will take the node down if the problem isn't transient */
sbd_membership_connect();
}
static gboolean
sbd_remote_check(gpointer user_data)
{
static int have_proc_pid = 0;
int running = 0;
cl_log(LOG_DEBUG, "Checking pacemaker remote connection: %d/%d", have_proc_pid, remoted_pid);
if(have_proc_pid == 0) {
char proc_path[PATH_MAX], exe_path[PATH_MAX];
/* check to make sure pid hasn't been reused by another process */
snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", (long unsigned int)getpid());
have_proc_pid = 1;
if(readlink(proc_path, exe_path, PATH_MAX - 1) < 0) {
have_proc_pid = -1;
}
}
if (remoted_pid <= 0) {
set_servant_health(pcmk_health_transient, LOG_WARNING, "No Pacemaker Remote connection");
goto notify;
} else if (kill(remoted_pid, 0) < 0 && errno == ESRCH) {
/* Not running */
} else if(have_proc_pid == -1) {
running = 1;
cl_log(LOG_DEBUG, "Poccess %ld is active", (long)remoted_pid);
} else {
int rc = 0;
char proc_path[PATH_MAX], exe_path[PATH_MAX], expected_path[PATH_MAX];
/* check to make sure pid hasn't been reused by another process */
snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", (long unsigned int)remoted_pid);
rc = readlink(proc_path, exe_path, PATH_MAX - 1);
if (rc < 0) {
crm_perror(LOG_ERR, "Could not read from %s", proc_path);
goto done;
}
exe_path[rc] = 0;
rc = snprintf(expected_path, sizeof(proc_path), "%s/pacemaker_remoted", SBINDIR);
expected_path[rc] = 0;
if (strcmp(exe_path, expected_path) == 0) {
cl_log(LOG_DEBUG, "Process %s (%ld) is active",
exe_path, (long)remoted_pid);
running = 1;
}
}
done:
if(running) {
set_servant_health(pcmk_health_online, LOG_INFO,
"Connected to Pacemaker Remote %lu", (long unsigned int)remoted_pid);
} else {
set_servant_health(pcmk_health_unclean, LOG_WARNING,
"Connection to Pacemaker Remote %lu lost", (long unsigned int)remoted_pid);
}
notify:
notify_parent();
if(running == 0) {
sbd_membership_connect();
}
return true;
}