X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=a3eef148810e3219d1978edeacfe1e2434d46690;hp=4de6613a0eff3e8847d31cc65c90228cc467adf8;hb=ad7051c0f2e1eb584c623794bf4b83548e3ccadc;hpb=c617c0c651432f9d5ae7adf4c5c1a5fd92ad828e diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 4de6613a0..a3eef1488 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -1460,19 +1460,51 @@ end_no_session: */ static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) + struct relay_command *cmd, struct lttng_ht *streams_ht) { int ret; + uint64_t stream_id; + struct relay_stream *stream; + struct lttng_ht_iter iter; + struct lttcomm_relayd_quiescent_control msg; struct lttcomm_relayd_generic_reply reply; DBG("Checking quiescent state on control socket"); + if (!cmd->session || cmd->version_check_done == 0) { + ERR("Trying to check for data before version check"); + ret = -1; + goto end_no_session; + } + + ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); + if (ret < sizeof(msg)) { + ERR("Relay didn't receive valid begin data_pending struct size: %d", + ret); + ret = -1; + goto end_no_session; + } + + stream_id = be64toh(msg.stream_id); + + rcu_read_lock(); + cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) { + if (stream->stream_handle == stream_id) { + stream->data_pending_check_done = 1; + DBG("Relay quiescent control pending flag set to %" PRIu64, + stream_id); + break; + } + } + rcu_read_unlock(); + reply.ret_code = htobe32(LTTNG_OK); ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay data quiescent control ret code failed"); } +end_no_session: return ret; } @@ -1643,7 +1675,7 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, ret = relay_data_pending(recv_hdr, cmd, streams_ht); break; case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, cmd); + ret = relay_quiescent_control(recv_hdr, cmd, streams_ht); break; case RELAYD_BEGIN_DATA_PENDING: ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht); @@ -1841,8 +1873,8 @@ void relay_del_connection(struct lttng_ht *relay_connections_ht, static void *relay_thread_worker(void *data) { - int i, ret, pollfd, err = -1; - uint32_t revents, nb_fd; + int ret, err = -1, last_seen_data_fd = -1; + uint32_t nb_fd; struct relay_command *relay_connection; struct lttng_poll_event events; struct lttng_ht *relay_connections_ht; @@ -1877,9 +1909,11 @@ void *relay_thread_worker(void *data) goto error; } +restart: while (1) { + int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; + /* Infinite blocking call, waiting for transmission */ - restart: DBG3("Relayd worker thread polling..."); ret = lttng_poll_wait(&events, -1); if (ret < 0) { @@ -1894,10 +1928,15 @@ void *relay_thread_worker(void *data) nb_fd = ret; + /* + * Process control. The control connection is prioritised so we don't + * starve it with high throughout put tracing data on the data + * connection. + */ for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ - revents = LTTNG_POLL_GETEV(&events, i); - pollfd = LTTNG_POLL_GETFD(&events, i); + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -1919,7 +1958,7 @@ void *relay_thread_worker(void *data) goto error; } } - } else if (revents > 0) { + } else if (revents) { rcu_read_lock(); lttng_ht_lookup(relay_connections_ht, (void *)((unsigned long) pollfd), @@ -1939,12 +1978,18 @@ void *relay_thread_worker(void *data) relay_del_connection(relay_connections_ht, streams_ht, &iter, relay_connection); + if (last_seen_data_fd == pollfd) { + last_seen_data_fd = last_notdel_data_fd; + } } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { DBG("Socket %d hung up", pollfd); relay_cleanup_poll_connection(&events, pollfd); relay_del_connection(relay_connections_ht, streams_ht, &iter, relay_connection); + if (last_seen_data_fd == pollfd) { + last_seen_data_fd = last_notdel_data_fd; + } } else if (revents & LPOLLIN) { /* control socket */ if (relay_connection->type == RELAY_CONTROL) { @@ -1966,34 +2011,101 @@ void *relay_thread_worker(void *data) ret = relay_process_control(&recv_hdr, relay_connection, streams_ht); - /* - * there was an error in processing a control - * command: clear the session - * */ if (ret < 0) { + /* Clear the session on error. */ relay_cleanup_poll_connection(&events, pollfd); relay_del_connection(relay_connections_ht, streams_ht, &iter, relay_connection); DBG("Connection closed with %d", pollfd); } + seen_control = 1; } - /* data socket */ - } else if (relay_connection->type == RELAY_DATA) { - ret = relay_process_data(relay_connection, streams_ht); - /* connection closed */ - if (ret < 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); - DBG("Data connection closed with %d", pollfd); - } + } else { + /* + * Flag the last seen data fd not deleted. It will be + * used as the last seen fd if any fd gets deleted in + * this first loop. + */ + last_notdel_data_fd = pollfd; + } + } + rcu_read_unlock(); + } + } + + /* + * The last loop handled a control request, go back to poll to make + * sure we prioritise the control socket. + */ + if (seen_control) { + continue; + } + + if (last_seen_data_fd >= 0) { + for (i = 0; i < nb_fd; i++) { + int pollfd = LTTNG_POLL_GETFD(&events, i); + if (last_seen_data_fd == pollfd) { + idx = i; + break; + } + } + } + + /* Process data connection. */ + for (i = idx + 1; i < nb_fd; i++) { + /* Fetch the poll data. */ + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Skip the command pipe. It's handled in the first loop. */ + if (pollfd == relay_cmd_pipe[0]) { + continue; + } + + if (revents) { + rcu_read_lock(); + lttng_ht_lookup(relay_connections_ht, + (void *)((unsigned long) pollfd), + &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + /* Skip it. Might be removed before. */ + rcu_read_unlock(); + continue; + } + relay_connection = caa_container_of(node, + struct relay_command, sock_n); + + if (revents & LPOLLIN) { + if (relay_connection->type != RELAY_DATA) { + continue; + } + + ret = relay_process_data(relay_connection, streams_ht); + /* connection closed */ + if (ret < 0) { + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); + DBG("Data connection closed with %d", pollfd); + /* + * Every goto restart call sets the last seen fd where + * here we don't really care since we gracefully + * continue the loop after the connection is deleted. + */ + } else { + /* Keep last seen port. */ + last_seen_data_fd = pollfd; + rcu_read_unlock(); + goto restart; } } rcu_read_unlock(); } } + last_seen_data_fd = -1; } exit: