Fix: rotation may never complete in per-PID buffering mode
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
index 5e1ff896c4e37102bf895819717eda89aefbbba3..a81a497a9f8a8fa042ba28f29bbf0e86e73dc92e 100644 (file)
@@ -32,6 +32,7 @@
 #include <unistd.h>
 #include <urcu/list.h>
 #include <signal.h>
+#include <stdbool.h>
 
 #include <bin/lttng-consumerd/health-consumerd.h>
 #include <common/common.h>
@@ -146,7 +147,8 @@ static struct lttng_consumer_channel *allocate_channel(uint64_t session_id,
  */
 static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
                struct lttng_consumer_channel *channel,
-               struct lttng_consumer_local_data *ctx, int *_alloc_ret)
+               struct lttng_consumer_local_data *ctx, int *_alloc_ret,
+               uint64_t trace_archive_id)
 {
        int alloc_ret;
        struct lttng_consumer_stream *stream = NULL;
@@ -165,7 +167,8 @@ static struct lttng_consumer_stream *allocate_stream(int cpu, int key,
                        cpu,
                        &alloc_ret,
                        channel->type,
-                       channel->monitor);
+                       channel->monitor,
+                       trace_archive_id);
        if (stream == NULL) {
                switch (alloc_ret) {
                case -ENOENT:
@@ -266,7 +269,8 @@ end:
  * Return 0 on success else a negative value.
  */
 static int create_ust_streams(struct lttng_consumer_channel *channel,
-               struct lttng_consumer_local_data *ctx)
+               struct lttng_consumer_local_data *ctx,
+               uint64_t trace_archive_id)
 {
        int ret, cpu = 0;
        struct ustctl_consumer_stream *ustream;
@@ -297,7 +301,8 @@ static int create_ust_streams(struct lttng_consumer_channel *channel,
                }
 
                /* Allocate consumer stream object. */
-               stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret);
+               stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret,
+                               trace_archive_id);
                if (!stream) {
                        goto error_alloc;
                }
@@ -406,9 +411,7 @@ error_shm_open:
        return -1;
 }
 
-static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
-               struct ustctl_consumer_channel_attr *attr,
-               int cpu)
+static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu)
 {
        char shm_path[PATH_MAX];
        int ret;
@@ -463,7 +466,7 @@ static int create_ust_channel(struct lttng_consumer_channel *channel,
                goto error_alloc;
        }
        for (i = 0; i < nr_stream_fds; i++) {
-               stream_fds[i] = open_ust_stream_fd(channel, attr, i);
+               stream_fds[i] = open_ust_stream_fd(channel, i);
                if (stream_fds[i] < 0) {
                        ret = -1;
                        goto error_open;
@@ -539,11 +542,11 @@ error:
 }
 
 /*
- * Send channel to sessiond.
+ * Send channel to sessiond and relayd if applicable.
  *
  * Return 0 on success or else a negative value.
  */
-static int send_sessiond_channel(int sock,
+static int send_channel_to_sessiond_and_relayd(int sock,
                struct lttng_consumer_channel *channel,
                struct lttng_consumer_local_data *ctx, int *relayd_error)
 {
@@ -563,6 +566,8 @@ static int send_sessiond_channel(int sock,
                        health_code_update();
 
                        /* Try to send the stream to the relayd if one is available. */
+                       DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd",
+                                       stream->key, channel->name);
                        ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
                        if (ret < 0) {
                                /*
@@ -638,9 +643,10 @@ error:
  * Return 0 on success or else, a negative value is returned and the channel
  * MUST be destroyed by consumer_del_channel().
  */
-static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
+static int ask_channel(struct lttng_consumer_local_data *ctx,
                struct lttng_consumer_channel *channel,
-               struct ustctl_consumer_channel_attr *attr)
+               struct ustctl_consumer_channel_attr *attr,
+               uint64_t trace_archive_id)
 {
        int ret;
 
@@ -681,7 +687,7 @@ static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
        }
 
        /* Open all streams for this channel. */
-       ret = create_ust_streams(channel, ctx);
+       ret = create_ust_streams(channel, ctx, trace_archive_id);
        if (ret < 0) {
                goto end;
        }
@@ -985,7 +991,8 @@ end:
  * Returns 0 on success, < 0 on error
  */
 static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
-               struct lttng_consumer_local_data *ctx)
+               struct lttng_consumer_local_data *ctx,
+               uint64_t trace_archive_id)
 {
        int ret = 0;
        struct lttng_consumer_channel *metadata_channel;
@@ -1025,7 +1032,7 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
         * The metadata stream is NOT created in no monitor mode when the channel
         * is created on a sessiond ask channel command.
         */
-       ret = create_ust_streams(metadata_channel, ctx);
+       ret = create_ust_streams(metadata_channel, ctx, trace_archive_id);
        if (ret < 0) {
                goto error;
        }
@@ -1080,7 +1087,8 @@ error:
  * Returns 0 on success, < 0 on error
  */
 static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
-               uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
+               uint64_t nb_packets_per_stream,
+               struct lttng_consumer_local_data *ctx)
 {
        int ret;
        unsigned use_relayd = 0;
@@ -1491,7 +1499,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 
                health_code_update();
 
-               ret = ask_channel(ctx, sock, channel, &attr);
+               ret = ask_channel(ctx, channel, &attr,
+                               msg.u.ask_channel.trace_archive_id);
                if (ret < 0) {
                        goto end_channel_error;
                }
@@ -1576,8 +1585,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 
                health_code_update();
 
-               /* Send everything to sessiond. */
-               ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
+               /* Send the channel to sessiond (and relayd, if applicable). */
+               ret = send_channel_to_sessiond_and_relayd(sock, channel, ctx,
+                               &relayd_err);
                if (ret < 0) {
                        if (relayd_err) {
                                /*
@@ -1745,7 +1755,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        ret = snapshot_metadata(msg.u.snapshot_channel.key,
                                        msg.u.snapshot_channel.pathname,
                                        msg.u.snapshot_channel.relayd_id,
-                                       ctx);
+                                       ctx,
+                                       msg.u.snapshot_channel.trace_archive_id);
                        if (ret < 0) {
                                ERR("Snapshot metadata failed");
                                ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
@@ -1918,44 +1929,41 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                }
                goto end_msg_sessiond;
        }
-       case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
+       case LTTNG_CONSUMER_ROTATE_CHANNEL:
        {
-               int channel_rotate_pipe;
-               int flags;
-
-               ret_code = LTTCOMM_CONSUMERD_SUCCESS;
-               /* Successfully received the command's type. */
-               ret = consumer_send_status_msg(sock, ret_code);
+               /*
+                * Sample the rotate position of all the streams in this channel.
+                */
+               ret = lttng_consumer_rotate_channel(msg.u.rotate_channel.key,
+                               msg.u.rotate_channel.pathname,
+                               msg.u.rotate_channel.relayd_id,
+                               msg.u.rotate_channel.metadata,
+                               msg.u.rotate_channel.new_chunk_id,
+                               ctx);
                if (ret < 0) {
-                       goto error_fatal;
+                       ERR("Rotate channel failed");
+                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
                }
 
-               ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe, 1);
-               if (ret != sizeof(channel_rotate_pipe)) {
-                       ERR("Failed to receive channel rotate pipe");
-                       goto error_fatal;
-               }
+               health_code_update();
 
-               DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
-               ctx->channel_rotate_pipe = channel_rotate_pipe;
-               /* Set the pipe as non-blocking. */
-               ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
-               if (ret == -1) {
-                       PERROR("fcntl get flags of the channel rotate pipe");
-                       goto error_fatal;
+               ret = consumer_send_status_msg(sock, ret_code);
+               if (ret < 0) {
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
                }
-               flags = ret;
 
-               ret = fcntl(channel_rotate_pipe, F_SETFL, flags | O_NONBLOCK);
-               if (ret == -1) {
-                       PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
-                       goto error_fatal;
-               }
-               DBG("Channel rotate pipe set as non-blocking");
-               ret_code = LTTCOMM_CONSUMERD_SUCCESS;
-               ret = consumer_send_status_msg(sock, ret_code);
+               /*
+                * Rotate the streams that are ready right now.
+                * FIXME: this is a second consecutive iteration over the
+                * streams in a channel, there is probably a better way to
+                * handle this, but it needs to be after the
+                * consumer_send_status_msg() call.
+                */
+               ret = lttng_consumer_rotate_ready_streams(
+                               msg.u.rotate_channel.key, ctx);
                if (ret < 0) {
-                       goto error_fatal;
+                       ERR("Rotate channel failed");
                }
                break;
        }
@@ -1970,7 +1978,77 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                msg.u.rotate_rename.relayd_id);
                if (ret < 0) {
                        ERR("Rotate rename failed");
+                       ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
+               }
+
+               health_code_update();
+
+               ret = consumer_send_status_msg(sock, ret_code);
+               if (ret < 0) {
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
+               }
+               break;
+       }
+       case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_LOCAL:
+       {
+               int pending;
+               uint32_t pending_reply;
+
+               DBG("Perform local check of pending rotation for session id %" PRIu64,
+                               msg.u.check_rotation_pending_local.session_id);
+               pending = lttng_consumer_check_rotation_pending_local(
+                               msg.u.check_rotation_pending_local.session_id,
+                               msg.u.check_rotation_pending_local.chunk_id);
+               if (pending < 0) {
+                       ERR("Local rotation pending check failed with code %i", pending);
+                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+               } else {
+                       pending_reply = !!pending;
+               }
+
+               health_code_update();
+
+               ret = consumer_send_status_msg(sock, ret_code);
+               if (ret < 0) {
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
+               }
+
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' flag as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
+               /* Send back returned value to session daemon */
+               ret = lttcomm_send_unix_sock(sock, &pending_reply,
+                               sizeof(pending_reply));
+               if (ret < 0) {
+                       PERROR("Failed to send rotation pending return code");
+                       goto error_fatal;
+               }
+               break;
+       }
+       case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY:
+       {
+               int pending;
+               uint32_t pending_reply;
+
+               DBG("Perform relayd check of pending rotation for session id %" PRIu64,
+                               msg.u.check_rotation_pending_relay.session_id);
+               pending = lttng_consumer_check_rotation_pending_relay(
+                               msg.u.check_rotation_pending_relay.session_id,
+                               msg.u.check_rotation_pending_relay.relayd_id,
+                               msg.u.check_rotation_pending_relay.chunk_id);
+               if (pending < 0) {
+                       ERR("Relayd rotation pending check failed with code %i", pending);
                        ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+               } else {
+                       pending_reply = !!pending;
                }
 
                health_code_update();
@@ -1980,6 +2058,23 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        /* Somehow, the session daemon is not responding anymore. */
                        goto end_nosignal;
                }
+
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' flag as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
+               /* Send back returned value to session daemon */
+               ret = lttcomm_send_unix_sock(sock, &pending_reply,
+                               sizeof(pending_reply));
+               if (ret < 0) {
+                       PERROR("Failed to send rotation pending return code");
+                       goto error_fatal;
+               }
                break;
        }
        case LTTNG_CONSUMER_MKDIR:
@@ -1993,7 +2088,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                msg.u.mkdir.relayd_id);
                if (ret < 0) {
                        ERR("consumer mkdir failed");
-                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+                       ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
                }
 
                health_code_update();
@@ -2594,10 +2689,10 @@ end:
  * Return 0 on success else a negative value.
  */
 int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
-               struct lttng_consumer_local_data *ctx)
+               struct lttng_consumer_local_data *ctx, bool *rotated)
 {
        unsigned long len, subbuf_size, padding;
-       int err, write_index = 1;
+       int err, write_index = 1, rotation_ret;
        long ret = 0;
        struct ustctl_consumer_stream *ustream;
        struct ctf_packet_index index;
@@ -2629,7 +2724,21 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                readlen = lttng_read(stream->wait_fd, &dummy, 1);
                if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
                        ret = readlen;
-                       goto end;
+                       goto error;
+               }
+       }
+
+       /*
+        * If the stream was flagged to be ready for rotation before we extract the
+        * next packet, rotate it now.
+        */
+       if (stream->rotate_ready) {
+               DBG("Rotate stream before extracting data");
+               rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
+               if (rotation_ret < 0) {
+                       ERR("Stream rotation error");
+                       ret = -1;
+                       goto error;
                }
        }
 
@@ -2644,7 +2753,7 @@ retry:
                if (stream->metadata_flag) {
                        ret = commit_one_metadata_packet(stream);
                        if (ret <= 0) {
-                               goto end;
+                               goto error;
                        }
                        ustctl_flush_buffer(stream->ustream, 1);
                        goto retry;
@@ -2659,7 +2768,7 @@ retry:
                 */
                DBG("Reserving sub buffer failed (everything is normal, "
                                "it is due to concurrency) [ret: %d]", err);
-               goto end;
+               goto error;
        }
        assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
 
@@ -2669,7 +2778,7 @@ retry:
                if (ret < 0) {
                        err = ustctl_put_subbuf(ustream);
                        assert(err == 0);
-                       goto end;
+                       goto error;
                }
 
                /* Update the stream's sequence and discarded events count. */
@@ -2678,7 +2787,7 @@ retry:
                        PERROR("kernctl_get_events_discarded");
                        err = ustctl_put_subbuf(ustream);
                        assert(err == 0);
-                       goto end;
+                       goto error;
                }
        } else {
                write_index = 0;
@@ -2696,6 +2805,7 @@ retry:
        assert(len >= subbuf_size);
 
        padding = len - subbuf_size;
+
        /* write the subbuffer to the tracefile */
        ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
        /*
@@ -2727,13 +2837,13 @@ retry:
        if (!stream->metadata_flag) {
                ret = notify_if_more_data(stream, ctx);
                if (ret < 0) {
-                       goto end;
+                       goto error;
                }
        }
 
        /* Write index if needed. */
        if (!write_index) {
-               goto end;
+               goto rotate;
        }
 
        if (stream->chan->live_timer_interval && !stream->metadata_flag) {
@@ -2758,17 +2868,35 @@ retry:
                }
 
                if (err < 0) {
-                       goto end;
+                       goto error;
                }
        }
 
        assert(!stream->metadata_flag);
        err = consumer_stream_write_index(stream, &index);
        if (err < 0) {
-               goto end;
+               goto error;
        }
 
-end:
+rotate:
+       /*
+        * After extracting the packet, we check if the stream is now ready to be
+        * rotated and perform the action immediately.
+        */
+       rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
+       if (rotation_ret == 1) {
+               rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
+               if (rotation_ret < 0) {
+                       ERR("Stream rotation error");
+                       ret = -1;
+                       goto error;
+               }
+       } else if (rotation_ret < 0) {
+               ERR("Checking if stream is ready to rotate");
+               ret = -1;
+               goto error;
+       }
+error:
        return ret;
 }
 
@@ -3013,7 +3141,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
        request.key = channel->key;
 
        DBG("Sending metadata request to sessiond, session id %" PRIu64
-                       ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
+                       ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64,
                        request.session_id, request.session_id_per_pid, request.uid,
                        request.key);
 
This page took 0.032796 seconds and 5 git commands to generate.