X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=a6b408a90740943138e43ad69160d34a5395baa6;hp=e426469a30ebe4125b70a0bec537af3d1c8a553c;hb=26296c48a2c7f6a7e90c0412c11cc2b1c5c2f287;hpb=0a6518b0685cb9e07cf156a4c882dc6ec40db35a diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index e426469a3..a6b408a90 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -57,6 +57,7 @@ #include "utils.h" #include "lttng-relayd.h" #include "live.h" +#include "health-relayd.h" /* command line options */ char *opt_output_path; @@ -67,6 +68,8 @@ static struct lttng_uri *live_uri; const char *progname; +const char *tracing_group_name = DEFAULT_TRACING_GROUP; + /* * Quit pipe for all threads. This permits a single cancellation point * for all threads when receiving an event on the pipe. @@ -85,6 +88,7 @@ static int dispatch_thread_exit; static pthread_t listener_thread; static pthread_t dispatcher_thread; static pthread_t worker_thread; +static pthread_t health_thread; static uint64_t last_relay_stream_id; static uint64_t last_relay_session_id; @@ -114,6 +118,9 @@ struct lttng_ht *viewer_streams_ht; /* Global hash table that stores relay index object. */ struct lttng_ht *indexes_ht; +/* Relayd health monitoring */ +struct health_app *health_relayd; + /* * usage function on stderr */ @@ -127,6 +134,7 @@ void usage(void) fprintf(stderr, " -D, --data-port URL Data port listening.\n"); fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); } static @@ -140,6 +148,7 @@ int parse_args(int argc, char **argv) { "control-port", 1, 0, 'C', }, { "data-port", 1, 0, 'D', }, { "daemonize", 0, 0, 'd', }, + { "group", 1, 0, 'g', }, { "help", 0, 0, 'h', }, { "output", 1, 0, 'o', }, { "verbose", 0, 0, 'v', }, @@ -148,7 +157,7 @@ int parse_args(int argc, char **argv) while (1) { int option_index = 0; - c = getopt_long(argc, argv, "dhv" "C:D:o:", + c = getopt_long(argc, argv, "dhv" "C:D:o:g:", long_options, &option_index); if (c == -1) { break; @@ -184,6 +193,9 @@ int parse_args(int argc, char **argv) case 'd': opt_daemon = 1; break; + case 'g': + tracing_group_name = optarg; + break; case 'h': usage(); exit(EXIT_FAILURE); @@ -281,18 +293,26 @@ void cleanup(void) static int notify_thread_pipe(int wpipe) { - int ret; + ssize_t ret; - do { - ret = write(wpipe, "!", 1); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != 1) { + ret = lttng_write(wpipe, "!", 1); + if (ret < 1) { PERROR("write poll pipe"); } return ret; } +static void notify_health_quit_pipe(int *pipe) +{ + ssize_t ret; + + ret = lttng_write(pipe[1], "4", 1); + if (ret < 1) { + PERROR("write relay health quit"); + } +} + /* * Stop all threads by closing the thread quit pipe. */ @@ -308,6 +328,8 @@ void stop_threads(void) ERR("write error on thread quit pipe"); } + notify_health_quit_pipe(health_quit_pipe); + /* Dispatch thread */ CMM_STORE_SHARED(dispatch_thread_exit, 1); futex_nto1_wake(&relay_cmd_queue.futex); @@ -513,6 +535,10 @@ void *relay_thread_listener(void *data) DBG("[thread] Relay listener started"); + health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); + + health_code_update(); + control_sock = relay_init_sock(control_uri); if (!control_sock) { goto error_sock_control; @@ -544,10 +570,14 @@ void *relay_thread_listener(void *data) } while (1) { + health_code_update(); + DBG("Listener accepting connections"); restart: + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -562,6 +592,8 @@ restart: DBG("Relay new connection received"); for (i = 0; i < nb_fd; i++) { + health_code_update(); + /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); @@ -656,8 +688,10 @@ error_sock_relay: lttcomm_destroy_sock(control_sock); error_sock_control: if (err) { - DBG("Thread exited with error"); + health_error(); + ERR("Health error occurred in %s", __func__); } + health_unregister(health_relayd); DBG("Relay listener thread cleanup complete"); stop_threads(); return NULL; @@ -669,17 +703,26 @@ error_sock_control: static void *relay_thread_dispatcher(void *data) { - int ret; + int err = -1; + ssize_t ret; struct cds_wfq_node *node; struct relay_command *relay_cmd = NULL; DBG("[thread] Relay dispatcher started"); + health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); + + health_code_update(); + while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + health_code_update(); + /* Atomically prepare the queue futex */ futex_nto1_prepare(&relay_cmd_queue.futex); do { + health_code_update(); + /* Dequeue commands */ node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); if (node == NULL) { @@ -696,22 +739,30 @@ void *relay_thread_dispatcher(void *data) * call is blocking so we can be assured that the data will be read * at some point in time or wait to the end of the world :) */ - do { - ret = write(relay_cmd_pipe[1], relay_cmd, - sizeof(struct relay_command)); - } while (ret < 0 && errno == EINTR); + ret = lttng_write(relay_cmd_pipe[1], relay_cmd, + sizeof(struct relay_command)); free(relay_cmd); - if (ret < 0 || ret != sizeof(struct relay_command)) { + if (ret < sizeof(struct relay_command)) { PERROR("write cmd pipe"); goto error; } } while (node != NULL); /* Futex wait on queue. Blocking call on futex() */ + health_poll_entry(); futex_nto1_wait(&relay_cmd_queue.futex); + health_poll_exit(); } + /* Normal exit, no error */ + err = 0; + error: + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_relayd); DBG("Dispatch thread dying"); stop_threads(); return NULL; @@ -749,8 +800,6 @@ void deferred_free_stream(struct rcu_head *head) struct relay_stream *stream = caa_container_of(head, struct relay_stream, rcu_node); - ctf_trace_try_destroy(stream->ctf_trace); - free(stream->path_name); free(stream->channel_name); free(stream); @@ -770,8 +819,7 @@ void deferred_free_session(struct rcu_head *head) * RCU read side lock MUST be acquired. If NO close_stream_check() was called * BEFORE the stream lock MUST be acquired. */ -static void destroy_stream(struct relay_stream *stream, - struct lttng_ht *ctf_traces_ht) +static void destroy_stream(struct relay_stream *stream) { int delret; struct relay_viewer_stream *vstream; @@ -799,15 +847,27 @@ static void destroy_stream(struct relay_stream *stream, * lookup failure on the live thread side of a stream indicates * that the viewer stream index received value should be used. */ + pthread_mutex_lock(&stream->viewer_stream_rotation_lock); vstream->total_index_received = stream->total_index_received; + vstream->tracefile_count_last = stream->tracefile_count_current; + vstream->close_write_flag = 1; + pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); } + /* Cleanup index of that stream. */ + relay_index_destroy_by_stream_id(stream->stream_handle); + iter.iter.node = &stream->stream_n.node; delret = lttng_ht_del(relay_streams_ht, &iter); assert(!delret); iter.iter.node = &stream->ctf_trace_node.node; - delret = lttng_ht_del(ctf_traces_ht, &iter); + delret = lttng_ht_del(stream->ctf_traces_ht, &iter); assert(!delret); + + if (stream->ctf_trace) { + ctf_trace_try_destroy(stream->ctf_trace); + } + call_rcu(&stream->rcu_node, deferred_free_stream); DBG("Closed tracefile %d from close stream", stream->fd); } @@ -839,10 +899,10 @@ void relay_delete_session(struct relay_command *cmd, } stream = caa_container_of(node, struct relay_stream, stream_n); if (stream->session == cmd->session) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); + cmd->session->stream_count--; + assert(cmd->session->stream_count >= 0); } - /* Cleanup index of that stream. */ - relay_index_destroy_by_stream_id(stream->stream_handle); } /* Make this session not visible anymore. */ @@ -903,6 +963,8 @@ int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, session->id = ++last_relay_session_id; session->sock = cmd->sock; + session->minor = cmd->minor; + session->major = cmd->major; cmd->session = session; reply.session_id = htobe64(session->id); @@ -1027,6 +1089,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, stream->ctf_trace->metadata_stream = stream; } ctf_trace_assign(cmd->ctf_traces_ht, stream); + stream->ctf_traces_ht = cmd->ctf_traces_ht; lttng_ht_node_init_ulong(&stream->stream_n, (unsigned long) stream->stream_handle); @@ -1035,6 +1098,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); + session->stream_count++; DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, stream->stream_handle); @@ -1111,9 +1175,11 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); stream->close_flag = 1; + session->stream_count--; + assert(session->stream_count >= 0); if (close_stream_check(stream)) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); } end_unlock: @@ -1184,7 +1250,7 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, */ static int write_padding_to_file(int fd, uint32_t size) { - int ret = 0; + ssize_t ret = 0; char *zeros; if (size == 0) { @@ -1198,10 +1264,8 @@ static int write_padding_to_file(int fd, uint32_t size) goto end; } - do { - ret = write(fd, zeros, size); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != size) { + ret = lttng_write(fd, zeros, size); + if (ret < size) { PERROR("write padding to file"); } @@ -1219,6 +1283,7 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, struct relay_command *cmd) { int ret = htobe32(LTTNG_OK); + ssize_t size_ret; struct relay_session *session = cmd->session; struct lttcomm_relayd_metadata_payload *metadata_struct; struct relay_stream *metadata_stream; @@ -1275,11 +1340,9 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end_unlock; } - do { - ret = write(metadata_stream->fd, metadata_struct->payload, - payload_size); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != payload_size) { + size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, + payload_size); + if (size_ret < payload_size) { ERR("Relay error writing metadata on file"); ret = -1; goto end_unlock; @@ -1831,17 +1894,99 @@ end: return ret; } +/* + * Handle index for a data stream. + * + * RCU read side lock MUST be acquired. + * + * Return 0 on success else a negative value. + */ +static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, + int rotate_index) +{ + int ret = 0, index_created = 0; + uint64_t stream_id, data_offset; + struct relay_index *index, *wr_index = NULL; + + assert(stream); + + stream_id = stream->stream_handle; + /* Get data offset because we are about to update the index. */ + data_offset = htobe64(stream->tracefile_size_current); + + /* + * Lookup for an existing index for that stream id/sequence number. If on + * exists, the control thread already received the data for it thus we need + * to write it on disk. + */ + index = relay_index_find(stream_id, net_seq_num); + if (!index) { + /* A successful creation will add the object to the HT. */ + index = relay_index_create(stream_id, net_seq_num); + if (!index) { + ret = -1; + goto error; + } + index_created = 1; + } + + if (rotate_index || stream->index_fd < 0) { + index->to_close_fd = stream->index_fd; + ret = index_create_file(stream->path_name, stream->channel_name, + relayd_uid, relayd_gid, stream->tracefile_size, + stream->tracefile_count_current); + if (ret < 0) { + /* This will close the stream's index fd if one. */ + relay_index_free_safe(index); + goto error; + } + stream->index_fd = ret; + } + index->fd = stream->index_fd; + index->index_data.offset = data_offset; + + if (index_created) { + /* + * Try to add the relay index object to the hash table. If an object + * already exist, destroy back the index created and set the data. + */ + relay_index_add(index, &wr_index); + if (wr_index) { + /* Copy back data from the created index. */ + wr_index->fd = index->fd; + wr_index->to_close_fd = index->to_close_fd; + wr_index->index_data.offset = data_offset; + free(index); + } + } else { + /* The index already exists so write it on disk. */ + wr_index = index; + } + + /* Do we have a writable ready index to write on disk. */ + if (wr_index) { + ret = relay_index_write(wr_index->fd, wr_index); + if (ret < 0) { + goto error; + } + stream->total_index_received++; + } + +error: + return ret; +} + /* * relay_process_data: Process the data received on the data socket */ static int relay_process_data(struct relay_command *cmd) { - int ret = 0, rotate_index = 0, index_created = 0; + int ret = 0, rotate_index = 0; + ssize_t size_ret; struct relay_stream *stream; - struct relay_index *index, *wr_index = NULL; struct lttcomm_relayd_data_hdr data_hdr; - uint64_t stream_id, data_offset; + uint64_t stream_id; uint64_t net_seq_num; uint32_t data_size; @@ -1901,10 +2046,54 @@ int relay_process_data(struct relay_command *cmd) if (stream->tracefile_size > 0 && (stream->tracefile_size_current + data_size) > stream->tracefile_size) { + struct relay_viewer_stream *vstream; + uint64_t new_id; + + new_id = (stream->tracefile_count_current + 1) % + stream->tracefile_count; + /* + * When we wrap-around back to 0, we start overwriting old + * trace data. + */ + if (!stream->tracefile_overwrite && new_id == 0) { + stream->tracefile_overwrite = 1; + } + pthread_mutex_lock(&stream->viewer_stream_rotation_lock); + if (stream->tracefile_overwrite) { + stream->oldest_tracefile_id = + (stream->oldest_tracefile_id + 1) % + stream->tracefile_count; + } + vstream = live_find_viewer_stream_by_id(stream->stream_handle); + if (vstream) { + /* + * The viewer is reading a file about to be + * overwritten. Close the FDs it is + * currently using and let it handle the fault. + */ + if (vstream->tracefile_count_current == new_id) { + pthread_mutex_lock(&vstream->overwrite_lock); + vstream->abort_flag = 1; + pthread_mutex_unlock(&vstream->overwrite_lock); + DBG("Streaming side setting abort_flag on stream %s_%lu\n", + stream->channel_name, new_id); + } else if (vstream->tracefile_count_current == + stream->tracefile_count_current) { + /* + * The reader and writer were in the + * same trace file, inform the viewer + * that no new index will ever be added + * to this file. + */ + vstream->close_write_flag = 1; + } + } ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, stream->tracefile_size, stream->tracefile_count, relayd_uid, relayd_gid, stream->fd, &(stream->tracefile_count_current), &stream->fd); + stream->total_index_received = 0; + pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); if (ret < 0) { ERR("Rotating stream output file"); goto end_rcu_unlock; @@ -1914,81 +2103,20 @@ int relay_process_data(struct relay_command *cmd) rotate_index = 1; } - /* Get data offset because we are about to update the index. */ - data_offset = htobe64(stream->tracefile_size_current); - /* - * Lookup for an existing index for that stream id/sequence number. If on - * exists, the control thread already received the data for it thus we need - * to write it on disk. + * Index are handled in protocol version 2.4 and above. Also, snapshot and + * index are NOT supported. */ - index = relay_index_find(stream_id, net_seq_num); - if (!index) { - /* A successful creation will add the object to the HT. */ - index = relay_index_create(stream->stream_handle, net_seq_num); - if (!index) { - goto end_rcu_unlock; - } - index_created = 1; - } - - if (rotate_index || stream->index_fd < 0) { - index->to_close_fd = stream->index_fd; - ret = index_create_file(stream->path_name, stream->channel_name, - relayd_uid, relayd_gid, stream->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - /* This will close the stream's index fd if one. */ - relay_index_free_safe(index); - goto end_rcu_unlock; - } - stream->index_fd = ret; - } - index->fd = stream->index_fd; - index->index_data.offset = data_offset; - - if (index_created) { - /* - * Try to add the relay index object to the hash table. If an object - * already exist, destroy back the index created and set the data. - */ - relay_index_add(index, &wr_index); - if (wr_index) { - /* Copy back data from the created index. */ - wr_index->fd = index->fd; - wr_index->to_close_fd = index->to_close_fd; - wr_index->index_data.offset = data_offset; - free(index); - } - } else { - /* The index already exists so write it on disk. */ - wr_index = index; - } - - /* Do we have a writable ready index to write on disk. */ - if (wr_index) { - /* Starting at 2.4, create the index file if none available. */ - if (cmd->minor >= 4 && stream->index_fd < 0) { - ret = index_create_file(stream->path_name, stream->channel_name, - relayd_uid, relayd_gid, stream->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - goto end_rcu_unlock; - } - stream->index_fd = ret; - } - - ret = relay_index_write(wr_index->fd, wr_index); + if (stream->session->minor >= 4 && !stream->session->snapshot) { + ret = handle_index_data(stream, net_seq_num, rotate_index); if (ret < 0) { goto end_rcu_unlock; } - stream->total_index_received++; } - do { - ret = write(stream->fd, data_buffer, data_size); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != data_size) { + /* Write data to stream output fd. */ + size_ret = lttng_write(stream->fd, data_buffer, data_size); + if (size_ret < data_size) { ERR("Relay error writing data to file"); ret = -1; goto end_rcu_unlock; @@ -2007,7 +2135,7 @@ int relay_process_data(struct relay_command *cmd) /* Check if we need to close the FD */ if (close_stream_check(stream)) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); } end_rcu_unlock: @@ -2034,24 +2162,30 @@ int relay_add_connection(int fd, struct lttng_poll_event *events, struct lttng_ht *relay_connections_ht) { struct relay_command *relay_connection; - int ret; + ssize_t ret; relay_connection = zmalloc(sizeof(struct relay_command)); if (relay_connection == NULL) { PERROR("Relay command zmalloc"); goto error; } - do { - ret = read(fd, relay_connection, sizeof(struct relay_command)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret < sizeof(struct relay_command)) { + ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); + if (ret < sizeof(struct relay_command)) { PERROR("read relay cmd pipe"); goto error_read; } - relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); - if (!relay_connection->ctf_traces_ht) { - goto error_read; + /* + * Only used by the control side and the reference is copied inside each + * stream from that connection. Thus a destroy HT must be done after every + * stream has been destroyed. + */ + if (relay_connection->type == RELAY_CONTROL) { + relay_connection->ctf_traces_ht = lttng_ht_new(0, + LTTNG_HT_TYPE_STRING); + if (!relay_connection->ctf_traces_ht) { + goto error_read; + } } lttng_ht_node_init_ulong(&relay_connection->sock_n, @@ -2076,7 +2210,6 @@ void deferred_free_connection(struct rcu_head *head) struct relay_command *relay_connection = caa_container_of(head, struct relay_command, rcu_node); - lttng_ht_destroy(relay_connection->ctf_traces_ht); lttcomm_destroy_sock(relay_connection->sock); free(relay_connection); } @@ -2090,12 +2223,13 @@ void relay_del_connection(struct lttng_ht *relay_connections_ht, ret = lttng_ht_del(relay_connections_ht, iter); assert(!ret); + if (relay_connection->type == RELAY_CONTROL) { relay_delete_session(relay_connection, sessions_ht); + lttng_ht_destroy(relay_connection->ctf_traces_ht); } - call_rcu(&relay_connection->rcu_node, - deferred_free_connection); + call_rcu(&relay_connection->rcu_node, deferred_free_connection); } /* @@ -2119,6 +2253,10 @@ void *relay_thread_worker(void *data) rcu_register_thread(); + health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); + + health_code_update(); + /* table of connections indexed on socket */ relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); if (!relay_connections_ht) { @@ -2145,9 +2283,13 @@ restart: while (1) { int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; + health_code_update(); + /* Infinite blocking call, waiting for transmission */ DBG3("Relayd worker thread polling..."); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -2170,6 +2312,8 @@ restart: uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + health_code_update(); + /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -2272,6 +2416,9 @@ restart: if (last_seen_data_fd >= 0) { for (i = 0; i < nb_fd; i++) { int pollfd = LTTNG_POLL_GETFD(&events, i); + + health_code_update(); + if (last_seen_data_fd == pollfd) { idx = i; break; @@ -2285,6 +2432,8 @@ restart: uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + health_code_update(); + /* Skip the command pipe. It's handled in the first loop. */ if (pollfd == relay_cmd_pipe[0]) { continue; @@ -2334,6 +2483,9 @@ restart: last_seen_data_fd = -1; } + /* Normal exit, no error */ + ret = 0; + exit: error: lttng_poll_clean(&events); @@ -2341,6 +2493,8 @@ error: /* empty the hash table and free the memory */ rcu_read_lock(); cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { + health_code_update(); + node = lttng_ht_iter_get_node_ulong(&iter); if (node) { relay_connection = caa_container_of(node, @@ -2349,15 +2503,9 @@ error: &iter, relay_connection, sessions_ht); } } -error_poll_create: - { - struct relay_index *index; - cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index, index_n.node) { - relay_index_delete(index); - } - lttng_ht_destroy(indexes_ht); - } rcu_read_unlock(); +error_poll_create: + lttng_ht_destroy(indexes_ht); indexes_ht_error: lttng_ht_destroy(relay_connections_ht); relay_connections_ht_error: @@ -2368,8 +2516,13 @@ relay_connections_ht_error: } DBG("Worker thread cleanup complete"); free(data_buffer); - stop_threads(); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_relayd); rcu_unregister_thread(); + stop_threads(); return NULL; } @@ -2484,6 +2637,26 @@ int main(int argc, char **argv) goto exit_relay_ctx_viewer_streams; } + /* Initialize thread health monitoring */ + health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); + if (!health_relayd) { + PERROR("health_app_create error"); + goto exit_health_app_create; + } + + ret = utils_create_pipe(health_quit_pipe); + if (ret < 0) { + goto error_health_pipe; + } + + /* Create thread to manage the client socket */ + ret = pthread_create(&health_thread, NULL, + thread_manage_health, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create health"); + goto health_error; + } + /* Setup the dispatcher thread */ ret = pthread_create(&dispatcher_thread, NULL, relay_thread_dispatcher, (void *) NULL); @@ -2508,31 +2681,52 @@ int main(int argc, char **argv) goto exit_listener; } - ret = live_start_threads(live_uri, relay_ctx); + ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe); if (ret != 0) { ERR("Starting live viewer threads"); + goto exit_live; } -exit_listener: +exit_live: ret = pthread_join(listener_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } -exit_worker: +exit_listener: ret = pthread_join(worker_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } -exit_dispatcher: +exit_worker: ret = pthread_join(dispatcher_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } + +exit_dispatcher: + ret = pthread_join(health_thread, &status); + if (ret != 0) { + PERROR("pthread_join health thread"); + goto error; /* join error, exit without cleanup */ + } + + /* + * Stop live threads only after joining other threads. + */ + live_stop_threads(); + +health_error: + utils_close_pipe(health_quit_pipe); + +error_health_pipe: + health_app_destroy(health_relayd); + +exit_health_app_create: lttng_ht_destroy(viewer_streams_ht); exit_relay_ctx_viewer_streams: @@ -2545,7 +2739,6 @@ exit_relay_ctx_sessions: free(relay_ctx); exit: - live_stop_threads(); cleanup(); if (!ret) { exit(EXIT_SUCCESS);