Fix: remove consumer health poll update on startup
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
index 339b20d3ee433be2d17f33d3462b104ab6266448..84b3f20ed5420b03210e8c80c4f638f4f5072d3b 100644 (file)
@@ -222,12 +222,6 @@ enum consumerd_state {
 static enum consumerd_state ust_consumerd_state;
 static enum consumerd_state kernel_consumerd_state;
 
-/* Used for the health monitoring of the session daemon. See health.h */
-struct health_state health_thread_cmd;
-struct health_state health_thread_app_manage;
-struct health_state health_thread_app_reg;
-struct health_state health_thread_kernel;
-
 /*
  * Socket timeout for receiving and sending in seconds.
  */
@@ -705,6 +699,8 @@ static void *thread_manage_kernel(void *data)
 
        DBG("[thread] Thread manage kernel started");
 
+       health_register(HEALTH_TYPE_KERNEL);
+
        /*
         * This first step of the while is to clean this structure which could free
         * non NULL pointers so zero it before the loop.
@@ -715,14 +711,14 @@ static void *thread_manage_kernel(void *data)
                goto error_testpoint;
        }
 
-       health_code_update(&health_thread_kernel);
+       health_code_update();
 
        if (testpoint(thread_manage_kernel_before_loop)) {
                goto error_testpoint;
        }
 
        while (1) {
-               health_code_update(&health_thread_kernel);
+               health_code_update();
 
                if (update_poll_flag == 1) {
                        /* Clean events object. We are about to populate it again. */
@@ -750,9 +746,9 @@ static void *thread_manage_kernel(void *data)
 
                /* Poll infinite value of time */
        restart:
-               health_poll_update(&health_thread_kernel);
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
-               health_poll_update(&health_thread_kernel);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -775,7 +771,7 @@ static void *thread_manage_kernel(void *data)
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);
 
-                       health_code_update(&health_thread_kernel);
+                       health_code_update();
 
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
@@ -823,12 +819,12 @@ error_testpoint:
        utils_close_pipe(kernel_poll_pipe);
        kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
        if (err) {
-               health_error(&health_thread_kernel);
+               health_error();
                ERR("Health error occurred in %s", __func__);
                WARN("Kernel thread died unexpectedly. "
                                "Kernel tracing can continue but CPU hotplug is disabled.");
        }
-       health_exit(&health_thread_kernel);
+       health_unregister();
        DBG("Kernel thread dying");
        return NULL;
 }
@@ -869,24 +865,9 @@ static void *thread_manage_consumer(void *data)
 
        DBG("[thread] Manage consumer started");
 
-       /*
-        * Since the consumer thread can be spawned at any moment in time, we init
-        * the health to a poll status (1, which is a valid health over time).
-        * When the thread starts, we update here the health to a "code" path being
-        * an even value so this thread, when reaching a poll wait, does not
-        * trigger an error with an even value.
-        *
-        * Here is the use case we avoid.
-        *
-        * +1: the first poll update during initialization (main())
-        * +2 * x: multiple code update once in this thread.
-        * +1: poll wait in this thread (being a good health state).
-        * == even number which after the wait period shows as a bad health.
-        *
-        * In a nutshell, the following poll update to the health state brings back
-        * the state to an even value meaning a code path.
-        */
-       health_poll_update(&consumer_data->health);
+       health_register(HEALTH_TYPE_CONSUMER);
+
+       health_code_update();
 
        /*
         * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
@@ -907,18 +888,18 @@ static void *thread_manage_consumer(void *data)
                goto error;
        }
 
-       health_code_update(&consumer_data->health);
+       health_code_update();
 
        /* Inifinite blocking call, waiting for transmission */
 restart:
-       health_poll_update(&consumer_data->health);
+       health_poll_entry();
 
        if (testpoint(thread_manage_consumer)) {
                goto error;
        }
 
        ret = lttng_poll_wait(&events, -1);
-       health_poll_update(&consumer_data->health);
+       health_poll_exit();
        if (ret < 0) {
                /*
                 * Restart interrupted system call.
@@ -936,7 +917,7 @@ restart:
                revents = LTTNG_POLL_GETEV(&events, i);
                pollfd = LTTNG_POLL_GETFD(&events, i);
 
-               health_code_update(&consumer_data->health);
+               health_code_update();
 
                /* Thread quit pipe has been closed. Killing thread. */
                ret = check_thread_quit_pipe(pollfd, revents);
@@ -965,7 +946,7 @@ restart:
         */
        (void) utils_set_fd_cloexec(sock);
 
-       health_code_update(&consumer_data->health);
+       health_code_update();
 
        DBG2("Receiving code from consumer err_sock");
 
@@ -976,7 +957,7 @@ restart:
                goto error;
        }
 
-       health_code_update(&consumer_data->health);
+       health_code_update();
 
        if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
                consumer_data->cmd_sock =
@@ -1006,13 +987,13 @@ restart:
                goto error;
        }
 
-       health_code_update(&consumer_data->health);
+       health_code_update();
 
        /* Inifinite blocking call, waiting for transmission */
 restart_poll:
-       health_poll_update(&consumer_data->health);
+       health_poll_entry();
        ret = lttng_poll_wait(&events, -1);
-       health_poll_update(&consumer_data->health);
+       health_poll_exit();
        if (ret < 0) {
                /*
                 * Restart interrupted system call.
@@ -1030,7 +1011,7 @@ restart_poll:
                revents = LTTNG_POLL_GETEV(&events, i);
                pollfd = LTTNG_POLL_GETFD(&events, i);
 
-               health_code_update(&consumer_data->health);
+               health_code_update();
 
                /* Thread quit pipe has been closed. Killing thread. */
                ret = check_thread_quit_pipe(pollfd, revents);
@@ -1048,7 +1029,7 @@ restart_poll:
                }
        }
 
-       health_code_update(&consumer_data->health);
+       health_code_update();
 
        /* Wait for any kconsumerd error */
        ret = lttcomm_recv_unix_sock(sock, &code,
@@ -1099,10 +1080,10 @@ error:
        lttng_poll_clean(&events);
 error_poll:
        if (err) {
-               health_error(&consumer_data->health);
+               health_error();
                ERR("Health error occurred in %s", __func__);
        }
-       health_exit(&consumer_data->health);
+       health_unregister();
        DBG("consumer thread cleanup completed");
 
        return NULL;
@@ -1123,11 +1104,13 @@ static void *thread_manage_apps(void *data)
        rcu_register_thread();
        rcu_thread_online();
 
+       health_register(HEALTH_TYPE_APP_MANAGE);
+
        if (testpoint(thread_manage_apps)) {
                goto error_testpoint;
        }
 
-       health_code_update(&health_thread_app_manage);
+       health_code_update();
 
        ret = create_thread_poll_set(&events, 2);
        if (ret < 0) {
@@ -1143,16 +1126,16 @@ static void *thread_manage_apps(void *data)
                goto error;
        }
 
-       health_code_update(&health_thread_app_manage);
+       health_code_update();
 
        while (1) {
                DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
 
                /* Inifinite blocking call, waiting for transmission */
        restart:
-               health_poll_update(&health_thread_app_manage);
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
-               health_poll_update(&health_thread_app_manage);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -1170,7 +1153,7 @@ static void *thread_manage_apps(void *data)
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);
 
-                       health_code_update(&health_thread_app_manage);
+                       health_code_update();
 
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
@@ -1194,7 +1177,7 @@ static void *thread_manage_apps(void *data)
                                                goto error;
                                        }
 
-                                       health_code_update(&health_thread_app_manage);
+                                       health_code_update();
 
                                        /* Register applicaton to the session daemon */
                                        ret = ust_app_register(&ust_cmd.reg_msg,
@@ -1205,7 +1188,7 @@ static void *thread_manage_apps(void *data)
                                                break;
                                        }
 
-                                       health_code_update(&health_thread_app_manage);
+                                       health_code_update();
 
                                        /*
                                         * Validate UST version compatibility.
@@ -1219,7 +1202,7 @@ static void *thread_manage_apps(void *data)
                                                update_ust_app(ust_cmd.sock);
                                        }
 
-                                       health_code_update(&health_thread_app_manage);
+                                       health_code_update();
 
                                        ret = ust_app_register_done(ust_cmd.sock);
                                        if (ret < 0) {
@@ -1250,7 +1233,7 @@ static void *thread_manage_apps(void *data)
                                                                ust_cmd.sock);
                                        }
 
-                                       health_code_update(&health_thread_app_manage);
+                                       health_code_update();
 
                                        break;
                                }
@@ -1272,7 +1255,7 @@ static void *thread_manage_apps(void *data)
                                }
                        }
 
-                       health_code_update(&health_thread_app_manage);
+                       health_code_update();
                }
        }
 
@@ -1291,10 +1274,10 @@ error_testpoint:
         */
 
        if (err) {
-               health_error(&health_thread_app_manage);
+               health_error();
                ERR("Health error occurred in %s", __func__);
        }
-       health_exit(&health_thread_app_manage);
+       health_unregister();
        DBG("Application communication apps thread cleanup complete");
        rcu_thread_offline();
        rcu_unregister_thread();
@@ -1391,6 +1374,8 @@ static void *thread_registration_apps(void *data)
 
        DBG("[thread] Manage application registration started");
 
+       health_register(HEALTH_TYPE_APP_REG);
+
        if (testpoint(thread_registration_apps)) {
                goto error_testpoint;
        }
@@ -1428,9 +1413,9 @@ static void *thread_registration_apps(void *data)
 
                /* Inifinite blocking call, waiting for transmission */
        restart:
-               health_poll_update(&health_thread_app_reg);
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
-               health_poll_update(&health_thread_app_reg);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -1444,7 +1429,7 @@ static void *thread_registration_apps(void *data)
                nb_fd = ret;
 
                for (i = 0; i < nb_fd; i++) {
-                       health_code_update(&health_thread_app_reg);
+                       health_code_update();
 
                        /* Fetch once the poll data */
                        revents = LTTNG_POLL_GETEV(&events, i);
@@ -1496,7 +1481,7 @@ static void *thread_registration_apps(void *data)
                                                sock = -1;
                                                continue;
                                        }
-                                       health_code_update(&health_thread_app_reg);
+                                       health_code_update();
                                        ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
                                                        sizeof(struct ust_register_msg));
                                        if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
@@ -1514,7 +1499,7 @@ static void *thread_registration_apps(void *data)
                                                sock = -1;
                                                continue;
                                        }
-                                       health_code_update(&health_thread_app_reg);
+                                       health_code_update();
 
                                        ust_cmd->sock = sock;
                                        sock = -1;
@@ -1545,7 +1530,7 @@ static void *thread_registration_apps(void *data)
 exit:
 error:
        if (err) {
-               health_error(&health_thread_app_reg);
+               health_error();
                ERR("Health error occurred in %s", __func__);
        }
 
@@ -1573,7 +1558,7 @@ error_listen:
 error_create_poll:
 error_testpoint:
        DBG("UST Registration thread cleanup complete");
-       health_exit(&health_thread_app_reg);
+       health_unregister();
 
        return NULL;
 }
@@ -1948,9 +1933,7 @@ static int check_consumer_health(void)
 {
        int ret;
 
-       ret = health_check_state(&kconsumer_data.health) &&
-               health_check_state(&ustconsumer32_data.health) &&
-               health_check_state(&ustconsumer64_data.health);
+       ret = health_check_state(HEALTH_TYPE_CONSUMER);
 
        DBG3("Health consumer check %d", ret);
 
@@ -2104,7 +2087,7 @@ static int create_ust_session(struct ltt_session *session,
 
        DBG("Creating UST session");
 
-       lus = trace_ust_create_session(session->path, session->id, domain);
+       lus = trace_ust_create_session(session->path, session->id);
        if (lus == NULL) {
                ret = LTTNG_ERR_UST_SESS_FAIL;
                goto error;
@@ -3068,26 +3051,26 @@ restart:
 
                switch (msg.component) {
                case LTTNG_HEALTH_CMD:
-                       reply.ret_code = health_check_state(&health_thread_cmd);
+                       reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
                        break;
                case LTTNG_HEALTH_APP_MANAGE:
-                       reply.ret_code = health_check_state(&health_thread_app_manage);
+                       reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
                        break;
                case LTTNG_HEALTH_APP_REG:
-                       reply.ret_code = health_check_state(&health_thread_app_reg);
+                       reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
                        break;
                case LTTNG_HEALTH_KERNEL:
-                       reply.ret_code = health_check_state(&health_thread_kernel);
+                       reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
                        break;
                case LTTNG_HEALTH_CONSUMER:
                        reply.ret_code = check_consumer_health();
                        break;
                case LTTNG_HEALTH_ALL:
                        reply.ret_code =
-                               health_check_state(&health_thread_app_manage) &&
-                               health_check_state(&health_thread_app_reg) &&
-                               health_check_state(&health_thread_cmd) &&
-                               health_check_state(&health_thread_kernel) &&
+                               health_check_state(HEALTH_TYPE_APP_MANAGE) &&
+                               health_check_state(HEALTH_TYPE_APP_REG) &&
+                               health_check_state(HEALTH_TYPE_CMD) &&
+                               health_check_state(HEALTH_TYPE_KERNEL) &&
                                check_consumer_health();
                        break;
                default:
@@ -3162,11 +3145,13 @@ static void *thread_manage_clients(void *data)
 
        rcu_register_thread();
 
+       health_register(HEALTH_TYPE_CMD);
+
        if (testpoint(thread_manage_clients)) {
                goto error_testpoint;
        }
 
-       health_code_update(&health_thread_cmd);
+       health_code_update();
 
        ret = lttcomm_listen_unix_sock(client_sock);
        if (ret < 0) {
@@ -3199,16 +3184,16 @@ static void *thread_manage_clients(void *data)
                goto error;
        }
 
-       health_code_update(&health_thread_cmd);
+       health_code_update();
 
        while (1) {
                DBG("Accepting client command ...");
 
                /* Inifinite blocking call, waiting for transmission */
        restart:
-               health_poll_update(&health_thread_cmd);
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
-               health_poll_update(&health_thread_cmd);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -3226,7 +3211,7 @@ static void *thread_manage_clients(void *data)
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);
 
-                       health_code_update(&health_thread_cmd);
+                       health_code_update();
 
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
@@ -3246,7 +3231,7 @@ static void *thread_manage_clients(void *data)
 
                DBG("Wait for client response");
 
-               health_code_update(&health_thread_cmd);
+               health_code_update();
 
                sock = lttcomm_accept_unix_sock(client_sock);
                if (sock < 0) {
@@ -3282,7 +3267,7 @@ static void *thread_manage_clients(void *data)
                cmd_ctx->llm = NULL;
                cmd_ctx->session = NULL;
 
-               health_code_update(&health_thread_cmd);
+               health_code_update();
 
                /*
                 * Data is received from the lttng client. The struct
@@ -3303,7 +3288,7 @@ static void *thread_manage_clients(void *data)
                        continue;
                }
 
-               health_code_update(&health_thread_cmd);
+               health_code_update();
 
                // TODO: Validate cmd_ctx including sanity check for
                // security purpose.
@@ -3336,7 +3321,7 @@ static void *thread_manage_clients(void *data)
                        continue;
                }
 
-               health_code_update(&health_thread_cmd);
+               health_code_update();
 
                DBG("Sending response (size: %d, retcode: %s)",
                                cmd_ctx->lttng_msg_size,
@@ -3355,7 +3340,7 @@ static void *thread_manage_clients(void *data)
 
                clean_command_ctx(&cmd_ctx);
 
-               health_code_update(&health_thread_cmd);
+               health_code_update();
        }
 
 exit:
@@ -3382,11 +3367,11 @@ error_testpoint:
        }
 
        if (err) {
-               health_error(&health_thread_cmd);
+               health_error();
                ERR("Health error occurred in %s", __func__);
        }
 
-       health_exit(&health_thread_cmd);
+       health_unregister();
 
        DBG("Client thread dying");
 
@@ -4167,26 +4152,6 @@ int main(int argc, char **argv)
 
        cmd_init();
 
-       /* Init all health thread counters. */
-       health_init(&health_thread_cmd);
-       health_init(&health_thread_kernel);
-       health_init(&health_thread_app_manage);
-       health_init(&health_thread_app_reg);
-
-       /*
-        * Init health counters of the consumer thread. We do a quick hack here to
-        * the state of the consumer health is fine even if the thread is not
-        * started. Once the thread starts, the health state is updated with a poll
-        * value to set a health code path. This is simply to ease our life and has
-        * no cost what so ever.
-        */
-       health_init(&kconsumer_data.health);
-       health_poll_update(&kconsumer_data.health);
-       health_init(&ustconsumer32_data.health);
-       health_poll_update(&ustconsumer32_data.health);
-       health_init(&ustconsumer64_data.health);
-       health_poll_update(&ustconsumer64_data.health);
-
        /* Check for the application socket timeout env variable. */
        env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
        if (env_app_timeout) {
This page took 0.032594 seconds and 4 git commands to generate.