void
sbd_cpg_membership_health_update()
{
if(cpg_membership_entries > 0) {
bool quorum_is_suspect =
(two_node && ever_seen_both && cpg_membership_entries == 1);
if (!quorum_is_suspect) {
set_servant_health(pcmk_health_online, LOG_INFO,
"Connected to %s (%u members)",
name_for_cluster_type(get_cluster_type()),
cpg_membership_entries
);
} else {
/* Alternative would be asking votequorum for number of votes.
* Using pacemaker's cpg as source for number of active nodes
* avoids binding to an additional library, is definitely
* less code to write and we wouldn't have to combine data
* from 3 sources (cmap, cpq & votequorum) in a potentially
* racy environment.
*/
set_servant_health(pcmk_health_noquorum, LOG_WARNING,
"Connected to %s but requires both nodes present",
name_for_cluster_type(get_cluster_type())
);
}
if (cpg_membership_entries > 1) {
ever_seen_both = true;
}
} else {
set_servant_health(pcmk_health_unclean, LOG_WARNING,
"Empty %s membership", name_for_cluster_type(get_cluster_type()));
}
}
static void
sbd_plugin_membership_dispatch(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
if(msg_len > 0) {
set_servant_health(pcmk_health_online, LOG_INFO,
"Connected to %s", name_for_cluster_type(get_cluster_type()));
} else {
set_servant_health(pcmk_health_unclean, LOG_WARNING,
"Broken %s message", name_for_cluster_type(get_cluster_type()));
}
notify_parent();
return;
}
void
sbd_cpg_membership_dispatch(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)
{
if(member_list_entries > 0) {
set_servant_health(pcmk_health_online, LOG_INFO,
"Connected to %s", name_for_cluster_type(get_cluster_type()));
} else {
set_servant_health(pcmk_health_unclean, LOG_WARNING,
"Empty %s membership", name_for_cluster_type(get_cluster_type()));
}
notify_parent();
}
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_connect(void)
{
bool connected = false;
cl_log(LOG_INFO, "Attempting cluster connection");
cluster.destroy = sbd_membership_destroy;
#if SUPPORT_PLUGIN
cluster.cpg.cpg_deliver_fn = sbd_plugin_membership_dispatch;
#endif
#if SUPPORT_COROSYNC
cluster.cpg.cpg_confchg_fn = sbd_cpg_membership_dispatch;
#endif
while(connected == false) {
enum cluster_type_e stack = get_cluster_type();
if(get_cluster_type() == pcmk_cluster_unknown) {
crm_debug("Attempting pacemaker remote connection");
/* Nothing is up, go looking for the pacemaker remote process */
if(find_pacemaker_remote() > 0) {
connected = true;
}
} else {
cl_log(LOG_INFO, "Attempting connection to %s", name_for_cluster_type(stack));
#if SUPPORT_COROSYNC && CHECK_TWO_NODE
if (sbd_get_two_node()) {
#endif
if(crm_cluster_connect(&cluster)) {
connected = true;
}
#if SUPPORT_COROSYNC && CHECK_TWO_NODE
}
#endif
}
if(connected == false) {
cl_log(LOG_INFO, "Failed, retrying in %ds", reconnect_msec / 1000);
sleep(reconnect_msec / 1000);
}
}
set_servant_health(pcmk_health_transient, LOG_INFO, "Connected, waiting for initial membership");
notify_parent();
notify_timer_cb(NULL);
}
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;
}
