Patch adds poll timer scheduler to be called 3 times per token timeout.
If poll timer was not called for more then 0.8 * token timeout, it means
corosync process was not scheduled and ether token_timeout should be
increased or load should be reduced (useful for VM, where host is
overcommitted so VM is not scheduled as expected).
Signed-off-by: Jan Friesse <jfriesse@xxxxxxxxxx>
---
exec/main.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 50 insertions(+), 0 deletions(-)
diff --git a/exec/main.c b/exec/main.c
index d23e244..42ffb7b 100644
--- a/exec/main.c
+++ b/exec/main.c
@@ -1295,6 +1295,49 @@ static struct coroipcs_init_state_v2 ipc_init_state_v2 = {
.stats_decrement_value = corosync_stats_decrement_value,
};
+struct scheduler_pause_timeout_data {
+ struct totem_config *totem_config;
+ poll_timer_handle handle;
+ unsigned long long tv_prev;
+ unsigned long long max_tv_diff;
+};
+
+static void timer_function_scheduler_timeout (void *data)
+{
+ struct scheduler_pause_timeout_data *timeout_data = (struct scheduler_pause_timeout_data *)data;
+ unsigned long long tv_current;
+ unsigned long long tv_diff;
+
+ tv_current = timerlist_nano_current_get ();
+
+ if (timeout_data->tv_prev == 0) {
+ /*
+ * Initial call -> just pretent everything is ok
+ */
+ timeout_data->tv_prev = tv_current;
+ timeout_data->max_tv_diff = 0;
+ }
+
+ tv_diff = tv_current - timeout_data->tv_prev;
+ timeout_data->tv_prev = tv_current;
+
+ if (tv_diff > timeout_data->max_tv_diff) {
+ log_printf (LOGSYS_LEVEL_WARNING, "Corosync main process was not scheduled for %0.4f ms "
+ "(threshold is %0.4f ms). Consider token timeout increase.",
+ (float)tv_diff / TIMERLIST_NS_IN_MSEC, (float)timeout_data->max_tv_diff / TIMERLIST_NS_IN_MSEC);
+ }
+
+ /*
+ * Set next threshold, because token_timeout can change
+ */
+ timeout_data->max_tv_diff = timeout_data->totem_config->token_timeout * TIMERLIST_NS_IN_MSEC * 0.8;
+ poll_timer_add (corosync_poll_handle,
+ timeout_data->totem_config->token_timeout / 3,
+ timeout_data,
+ timer_function_scheduler_timeout,
+ &timeout_data->handle);
+}
+
static void corosync_setscheduler (void)
{
#if defined(HAVE_PTHREAD_SETSCHEDPARAM) && defined(HAVE_SCHED_GET_PRIORITY_MAX) && defined(HAVE_SCHED_SETSCHEDULER)
@@ -1556,6 +1599,7 @@ error_close:
return (err);
}
+
int main (int argc, char **argv, char **envp)
{
const char *error_string;
@@ -1576,6 +1620,7 @@ int main (int argc, char **argv, char **envp)
char corosync_lib_dir[PATH_MAX];
hdb_handle_t object_runtime_handle;
enum e_ais_done flock_err;
+ struct scheduler_pause_timeout_data scheduler_pause_timeout_data;
/* default configuration
*/
@@ -1788,9 +1833,14 @@ int main (int argc, char **argv, char **envp)
serialize_unlock,
sched_priority);
+
corosync_poll_handle = poll_create ();
poll_low_fds_event_set(corosync_poll_handle, main_low_fds_event);
+ memset(&scheduler_pause_timeout_data, 0, sizeof(scheduler_pause_timeout_data));
+ scheduler_pause_timeout_data.totem_config = &totem_config;
+ timer_function_scheduler_timeout (&scheduler_pause_timeout_data);
+
/*
* Create exit pipe
*/
--
1.7.1
_______________________________________________
discuss mailing list
discuss@xxxxxxxxxxxx
http://lists.corosync.org/mailman/listinfo/discuss
