Fix: failure to launch agent thread is not reported

[lttng-tools.git] / src / bin / lttng-sessiond / main.c

diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c
index a4f9763ebf6c33b90bbe2cc0059490f0715eb418..cec8e835e04881c20c65787c19a560bc19756c58 100644 (file)
--- a/src/bin/lttng-sessiond/main.c
+++ b/src/bin/lttng-sessiond/main.c
@@ -304,7 +304,7 @@ struct lttng_ht *agent_apps_ht_by_sock = NULL;
  * NR_LTTNG_SESSIOND_READY must match the number of calls to
  * sessiond_notify_ready().
  */
-#define NR_LTTNG_SESSIOND_READY                5
+#define NR_LTTNG_SESSIOND_READY                6
 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
 
 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
@@ -4496,12 +4496,41 @@ static void *thread_manage_clients(void *data)
        }
 
        sessiond_notify_ready();
+
        ret = sem_post(&load_info->message_thread_ready);
        if (ret) {
                PERROR("sem_post message_thread_ready");
                goto error;
        }
 
+       /*
+        * Wait until all support threads are initialized before accepting
+        * commands.
+        */
+       while (uatomic_read(&lttng_sessiond_ready) != 0) {
+               fd_set read_fds;
+               struct timeval timeout;
+
+               FD_ZERO(&read_fds);
+               FD_SET(thread_quit_pipe[0], &read_fds);
+               memset(&timeout, 0, sizeof(timeout));
+               timeout.tv_usec = 1000;
+
+               /*
+                * If a support thread failed to launch, it may signal that
+                * we must exit and the sessiond would never be marked as
+                * "ready".
+                *
+                * The timeout is set to 1ms, which serves as a way to
+                * pace down this check.
+                */
+               ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
+                               &timeout);
+               if (ret > 0 || (ret < 0 && errno != EINTR)) {
+                       goto exit;
+               }
+       }
+
        /* This testpoint is after we signal readiness to the parent. */
        if (testpoint(sessiond_thread_manage_clients)) {
                goto error;
@@ -5717,9 +5746,9 @@ int main(int argc, char **argv)
        struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
                        *ust64_channel_monitor_pipe = NULL,
                        *kernel_channel_monitor_pipe = NULL;
-       bool notification_thread_running = false;
-       bool rotation_thread_running = false;
-       bool timer_thread_running = false;
+       bool notification_thread_launched = false;
+       bool rotation_thread_launched = false;
+       bool timer_thread_launched = false;
        struct lttng_pipe *ust32_channel_rotate_pipe = NULL,
                        *ust64_channel_rotate_pipe = NULL,
                        *kernel_channel_rotate_pipe = NULL;
@@ -6154,7 +6183,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_notification;
        }
-       notification_thread_running = true;
+       notification_thread_launched = true;
 
        /* Create timer thread. */
        ret = pthread_create(&timer_thread, default_pthread_attr(),
@@ -6166,7 +6195,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_notification;
        }
-       timer_thread_running = true;
+       timer_thread_launched = true;
 
        /* rotation_thread_data acquires the pipes' read side. */
        rotation_thread_handle = rotation_thread_handle_create(
@@ -6194,7 +6223,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_rotation;
        }
-       rotation_thread_running = true;
+       rotation_thread_launched = true;
 
        /* Create thread to manage the client socket */
        ret = pthread_create(&client_thread, default_pthread_attr(),
@@ -6400,7 +6429,7 @@ exit_init_data:
         * of the active session and channels at the moment of the teardown.
         */
        if (notification_thread_handle) {
-               if (notification_thread_running) {
+               if (notification_thread_launched) {
                        notification_thread_command_quit(
                                        notification_thread_handle);
                        ret = pthread_join(notification_thread, &status);
@@ -6414,7 +6443,7 @@ exit_init_data:
        }
 
        if (rotation_thread_handle) {
-               if (rotation_thread_running) {
+               if (rotation_thread_launched) {
                        ret = pthread_join(rotation_thread, &status);
                        if (ret) {
                                errno = ret;
@@ -6425,7 +6454,7 @@ exit_init_data:
                rotation_thread_handle_destroy(rotation_thread_handle);
        }
 
-       if (timer_thread_running) {
+       if (timer_thread_launched) {
                kill(getpid(), LTTNG_SESSIOND_SIG_EXIT);
                ret = pthread_join(timer_thread, &status);
                if (ret) {