X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Frotation-thread.c;h=e6b9b6a1e72790702a7d96e1412dd3bcadc43662;hp=1704777a09ece70aef872745fd0ffc0c451ba939;hb=d73ee93f00ad2bf67dc79f2e3e09f74e2fb3d01f;hpb=d086f507d02078aed618ab291a0bc4a634958fa3 diff --git a/src/bin/lttng-sessiond/rotation-thread.c b/src/bin/lttng-sessiond/rotation-thread.c index 1704777a0..e6b9b6a1e 100644 --- a/src/bin/lttng-sessiond/rotation-thread.c +++ b/src/bin/lttng-sessiond/rotation-thread.c @@ -33,6 +33,7 @@ #include #include +#include #include "rotation-thread.h" #include "lttng-sessiond.h" @@ -52,6 +53,8 @@ */ struct cds_lfht *channel_pending_rotate_ht; +struct lttng_notification_channel *rotate_notification_channel = NULL; + struct rotation_thread_state { struct lttng_poll_event events; }; @@ -142,7 +145,9 @@ struct rotation_thread_handle *rotation_thread_handle_create( struct lttng_pipe *ust64_channel_rotate_pipe, struct lttng_pipe *kernel_channel_rotate_pipe, int thread_quit_pipe, - struct rotation_thread_timer_queue *rotation_timer_queue) + struct rotation_thread_timer_queue *rotation_timer_queue, + struct notification_thread_handle *notification_thread_handle, + sem_t *notification_thread_ready) { struct rotation_thread_handle *handle; @@ -183,6 +188,8 @@ struct rotation_thread_handle *rotation_thread_handle_create( } handle->thread_quit_pipe = thread_quit_pipe; handle->rotation_timer_queue = rotation_timer_queue; + handle->notification_thread_handle = notification_thread_handle; + handle->notification_thread_ready = notification_thread_ready; end: return handle; @@ -256,6 +263,9 @@ void fini_thread_state(struct rotation_thread_state *state) { lttng_poll_clean(&state->events); cds_lfht_destroy(channel_pending_rotate_ht, NULL); + if (rotate_notification_channel) { + lttng_notification_channel_destroy(rotate_notification_channel); + } } static @@ -281,6 +291,25 @@ int init_thread_state(struct rotation_thread_handle *handle, goto end; } + /* + * We wait until the notification thread is ready to create the + * notification channel and add it to the poll_set. + */ + sem_wait(handle->notification_thread_ready); + rotate_notification_channel = lttng_notification_channel_create( + lttng_session_daemon_notification_endpoint); + if (!rotate_notification_channel) { + ERR("[rotation-thread] Could not create notification channel"); + ret = -1; + goto end; + } + ret = lttng_poll_add(&state->events, rotate_notification_channel->socket, + LPOLLIN | LPOLLERR); + if (ret < 0) { + ERR("[rotation-thread] Failed to add notification fd to pollset"); + goto end; + } + end: return ret; } @@ -327,7 +356,7 @@ int handle_channel_rotation_pipe(int fd, uint32_t revents, } DBG("[rotation-thread] Received notification for chan %" PRIu64 - ", domain %d\n", key, domain); + ", domain %d", key, domain); channel_info = lookup_channel_pending(key, domain); if (!channel_info) { @@ -355,7 +384,7 @@ int handle_channel_rotation_pipe(int fd, uint32_t revents, time_t now = time(NULL); if (now == (time_t) -1) { - session->rotation_status = LTTNG_ROTATION_STATUS_ERROR; + session->rotation_state = LTTNG_ROTATION_STATE_ERROR; ret = LTTNG_ERR_UNK; goto end_unlock_session; } @@ -366,8 +395,19 @@ int handle_channel_rotation_pipe(int fd, uint32_t revents, goto end_unlock_session; } session->rotate_pending = false; - session->rotation_status = LTTNG_ROTATION_STATUS_COMPLETED; session->last_chunk_start_ts = session->current_chunk_start_ts; + if (session->rotate_pending_relay) { + ret = sessiond_timer_rotate_pending_start( + session, + DEFAULT_ROTATE_PENDING_RELAY_TIMER); + if (ret) { + ERR("Failed to enable rotate pending timer"); + ret = -1; + goto end_unlock_session; + } + } else { + session->rotation_state = LTTNG_ROTATION_STATE_COMPLETED; + } DBG("Rotation completed for session %s", session->name); } @@ -383,6 +423,323 @@ end: return ret; } +/* + * Process the rotate_pending check, called with session lock held. + */ +static +int rotate_pending_relay_timer(struct ltt_session *session) +{ + int ret; + + DBG("[rotation-thread] Check rotate pending on session %" PRIu64, + session->id); + ret = relay_rotate_pending(session, session->rotate_count - 1); + if (ret < 0) { + ERR("[rotation-thread] Check relay rotate pending"); + goto end; + } + if (ret == 0) { + DBG("[rotation-thread] Rotation completed on the relay for " + "session %" PRIu64, session->id); + /* + * Now we can clear the pending flag in the session. New + * rotations can start now. + */ + session->rotate_pending_relay = false; + session->rotation_state = LTTNG_ROTATION_STATE_COMPLETED; + } else if (ret == 1) { + DBG("[rotation-thread] Rotation still pending on the relay for " + "session %" PRIu64, session->id); + ret = sessiond_timer_rotate_pending_start(session, + DEFAULT_ROTATE_PENDING_RELAY_TIMER); + if (ret) { + ERR("Re-enabling rotate pending timer"); + ret = -1; + goto end; + } + } + + ret = 0; + +end: + return ret; +} + +/* + * Process the rotate_timer, called with session lock held. + */ +static +int rotate_timer(struct ltt_session *session) +{ + int ret; + + /* + * Complete _at most_ one scheduled rotation on a stopped session. + */ + if (!session->active && session->rotate_timer_enabled && + session->rotated_after_last_stop) { + ret = 0; + goto end; + } + + /* Ignore this timer if a rotation is already in progress. */ + if (session->rotate_pending || session->rotate_pending_relay) { + ret = 0; + goto end; + } + + DBG("[rotation-thread] Rotate timer on session %s", session->name); + + ret = cmd_rotate_session(session, NULL); + if (ret == -LTTNG_ERR_ROTATION_PENDING) { + DBG("Scheduled rotation aborted since a rotation is already in progress"); + ret = 0; + goto end; + } else if (ret != LTTNG_OK) { + ERR("[rotation-thread] Automatic time-triggered rotation failed with error code %i", ret); + ret = -1; + goto end; + } + + ret = 0; + +end: + return ret; +} + +static +int handle_rotate_timer_pipe(uint32_t revents, + struct rotation_thread_handle *handle, + struct rotation_thread_state *state, + struct rotation_thread_timer_queue *queue) +{ + int ret = 0; + int fd = lttng_pipe_get_readfd(queue->event_pipe); + struct ltt_session *session; + char buf[1]; + + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ret = lttng_poll_del(&state->events, fd); + if (ret) { + ERR("[rotation-thread] Failed to remove consumer " + "rotate pending pipe from poll set"); + } + goto end; + } + + ret = lttng_read(fd, buf, 1); + if (ret != 1) { + ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd); + ret = -1; + goto end; + } + + for (;;) { + struct sessiond_rotation_timer *timer_data; + + /* + * Take the queue lock only to pop elements from the list. + */ + pthread_mutex_lock(&queue->lock); + if (cds_list_empty(&queue->list)) { + pthread_mutex_unlock(&queue->lock); + break; + } + timer_data = cds_list_first_entry(&queue->list, + struct sessiond_rotation_timer, head); + cds_list_del(&timer_data->head); + pthread_mutex_unlock(&queue->lock); + + /* + * session lock to lookup the session ID. + */ + session_lock_list(); + session = session_find_by_id(timer_data->session_id); + if (!session) { + DBG("[rotation-thread] Session %" PRIu64 " not found", + timer_data->session_id); + /* + * This is a non-fatal error, and we cannot report it to the + * user (timer), so just print the error and continue the + * processing. + */ + session_unlock_list(); + free(timer_data); + continue; + } + + /* + * Take the session lock and release the session_list lock. + */ + session_lock(session); + session_unlock_list(); + + if (timer_data->signal == LTTNG_SESSIOND_SIG_ROTATE_PENDING) { + ret = rotate_pending_relay_timer(session); + } else if (timer_data->signal == LTTNG_SESSIOND_SIG_ROTATE_TIMER) { + ret = rotate_timer(session); + } else { + ERR("Unknown signal in rotate timer %d", timer_data->signal); + ret = -1; + } + session_unlock(session); + free(timer_data); + if (ret) { + ERR("Error processing timer"); + goto end; + } + } + + ret = 0; + +end: + return ret; +} + +int handle_condition( + const struct lttng_condition *condition, + const struct lttng_evaluation *evaluation, + struct notification_thread_handle *notification_thread_handle) +{ + int ret = 0; + const char *condition_session_name = NULL; + enum lttng_condition_type condition_type; + enum lttng_condition_status condition_status; + enum lttng_evaluation_status evaluation_status; + uint64_t consumed; + struct ltt_session *session; + + condition_type = lttng_condition_get_type(condition); + + if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) { + ret = -1; + ERR("[rotation-thread] Condition type and session usage type are not the same"); + goto end; + } + + /* Fetch info to test */ + condition_status = lttng_condition_session_consumed_size_get_session_name( + condition, &condition_session_name); + if (condition_status != LTTNG_CONDITION_STATUS_OK) { + ERR("[rotation-thread] Session name could not be fetched"); + ret = -1; + goto end; + } + evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation, + &consumed); + if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) { + ERR("[rotation-thread] Failed to get evaluation"); + ret = -1; + goto end; + } + + session_lock_list(); + session = session_find_by_name(condition_session_name); + if (!session) { + ret = -1; + session_unlock_list(); + ERR("[rotation-thread] Session \"%s\" not found", + condition_session_name); + goto end; + } + session_lock(session); + session_unlock_list(); + + ret = unsubscribe_session_consumed_size_rotation(session, + notification_thread_handle); + if (ret) { + goto end; + } + + ret = cmd_rotate_session(session, NULL); + if (ret == -LTTNG_ERR_ROTATION_PENDING) { + DBG("Rotate already pending, subscribe to the next threshold value"); + ret = 0; + } else if (ret != LTTNG_OK) { + ERR("[rotation-thread] Failed to rotate on size notification with error: %s", + lttng_strerror(ret)); + ret = -1; + goto end_unlock; + } + ret = subscribe_session_consumed_size_rotation(session, + consumed + session->rotate_size, + notification_thread_handle); + if (ret) { + ERR("[rotation-thread] Failed to subscribe to session consumed size condition"); + goto end_unlock; + } + ret = 0; + +end_unlock: + session_unlock(session); +end: + return ret; +} + +static +int handle_notification_channel(int fd, uint32_t revents, + struct rotation_thread_handle *handle, + struct rotation_thread_state *state) +{ + int ret; + bool notification_pending; + struct lttng_notification *notification = NULL; + enum lttng_notification_channel_status status; + const struct lttng_evaluation *notification_evaluation; + const struct lttng_condition *notification_condition; + + status = lttng_notification_channel_has_pending_notification( + rotate_notification_channel, ¬ification_pending); + if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) { + ERR("[rotation-thread ]Error occured while checking for pending notification"); + ret = -1; + goto end; + } + + if (!notification_pending) { + ret = 0; + goto end; + } + + /* Receive the next notification. */ + status = lttng_notification_channel_get_next_notification( + rotate_notification_channel, + ¬ification); + + switch (status) { + case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK: + break; + case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED: + /* Not an error, we will wait for the next one */ + ret = 0; + goto end;; + case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED: + ERR("Notification channel was closed"); + ret = -1; + goto end; + default: + /* Unhandled conditions / errors. */ + ERR("Unknown notification channel status"); + ret = -1; + goto end; + } + + notification_condition = lttng_notification_get_condition(notification); + notification_evaluation = lttng_notification_get_evaluation(notification); + + ret = handle_condition(notification_condition, notification_evaluation, + handle->notification_thread_handle); + +end: + lttng_notification_destroy(notification); + if (ret != 0) { + goto end; + } + + + return ret; +} + void *thread_rotation(void *data) { int ret; @@ -440,13 +797,27 @@ void *thread_rotation(void *data) if (fd == handle->thread_quit_pipe) { DBG("[rotation-thread] Quit pipe activity"); goto exit; + } else if (fd == lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe)) { + ret = handle_rotate_timer_pipe(revents, + handle, &state, handle->rotation_timer_queue); + if (ret) { + ERR("[rotation-thread] Failed to handle rotation timer pipe event"); + goto error; + } } else if (fd == handle->ust32_consumer || fd == handle->ust64_consumer || fd == handle->kernel_consumer) { ret = handle_channel_rotation_pipe(fd, revents, handle, &state); if (ret) { - ERR("[rotation-thread] Handle channel rotation pipe"); + ERR("[rotation-thread] Failed to handle channel rotation pipe"); + goto error; + } + } else if (fd == rotate_notification_channel->socket) { + ret = handle_notification_channel(fd, revents, + handle, &state); + if (ret) { + ERR("[rotation-thread] Error occured while handling activity on notification channel socket"); goto error; } }