Handle the tracefile_size/count in live
authorJulien Desfossez <jdesfossez@efficios.com>
Wed, 13 Nov 2013 16:00:10 +0000 (11:00 -0500)
committerDavid Goulet <dgoulet@efficios.com>
Wed, 13 Nov 2013 16:23:42 +0000 (11:23 -0500)
Signed-off-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: David Goulet <dgoulet@efficios.com>
src/bin/lttng-relayd/live.c
src/bin/lttng-relayd/lttng-relayd.h
src/bin/lttng-relayd/lttng-viewer.h
src/bin/lttng-relayd/main.c

index b0cc63128a0910ae708af1de6b5af4646400a125..e51ff5cdf53495d4848a2867cfc952e9f2534bc5 100644 (file)
@@ -616,15 +616,15 @@ static int open_index(struct relay_viewer_stream *stream)
        char fullpath[PATH_MAX];
        struct lttng_packet_index_file_hdr hdr;
 
-       if (stream->tracefile_size > 0) {
-               /* For now we don't support on-disk ring buffer. */
-               ret = -1;
-               goto end;
+       if (stream->tracefile_count > 0) {
+               ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s_%"
+                               PRIu64 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
+                               stream->channel_name, stream->tracefile_count_current);
+       } else {
+               ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s"
+                               DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
+                               stream->channel_name);
        }
-
-       ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s"
-                       DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
-                       stream->channel_name);
        if (ret < 0) {
                PERROR("snprintf index path");
                goto error;
@@ -663,7 +663,6 @@ static int open_index(struct relay_viewer_stream *stream)
        ret = 0;
 
 error:
-end:
        return ret;
 }
 
@@ -691,25 +690,60 @@ int init_viewer_stream(struct relay_stream *stream, int seek_last)
                ret = -1;
                goto error;
        }
-
-       viewer_stream->read_fd = -1;
-       viewer_stream->index_read_fd = -1;
        viewer_stream->session_id = stream->session->id;
        viewer_stream->stream_handle = stream->stream_handle;
        viewer_stream->path_name = strndup(stream->path_name,
                        LTTNG_VIEWER_PATH_MAX);
        viewer_stream->channel_name = strndup(stream->channel_name,
                        LTTNG_VIEWER_NAME_MAX);
-       viewer_stream->total_index_received = stream->total_index_received;
-       viewer_stream->tracefile_size = stream->tracefile_size;
        viewer_stream->tracefile_count = stream->tracefile_count;
        viewer_stream->metadata_flag = stream->metadata_flag;
+       if (seek_last) {
+               viewer_stream->tracefile_count_current =
+                       stream->tracefile_count_current;
+       } else {
+               viewer_stream->tracefile_count_current =
+                       stream->oldest_tracefile_id;
+       }
+
+       /*
+        * The deletion of this ctf_trace object is only done in a call RCU of the
+        * relay stream making it valid as long as we have the read side lock.
+        */
+       viewer_stream->ctf_trace = stream->ctf_trace;
+       uatomic_inc(&viewer_stream->ctf_trace->refcount);
 
-       if (seek_last && viewer_stream->total_index_received > 0) {
+       lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
+       lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
+
+       viewer_stream->index_read_fd = -1;
+       viewer_stream->read_fd = -1;
+
+       /*
+        * This is to avoid a race between the initialization of this object and
+        * the close of the given stream. If the stream is unable to find this
+        * viewer stream when closing, this copy will at least take the latest
+        * value.
+        * We also need that for the seek_last.
+        */
+       viewer_stream->total_index_received = stream->total_index_received;
+
+       /*
+        * If we never received an index for the current stream, delay
+        * the opening of the index, otherwise open it right now.
+        */
+       if (viewer_stream->tracefile_count_current ==
+                       stream->tracefile_count_current &&
+                       viewer_stream->total_index_received == 0) {
+               viewer_stream->index_read_fd = -1;
+       } else {
                ret = open_index(viewer_stream);
                if (ret < 0) {
                        goto error;
                }
+       }
+
+       if (seek_last && viewer_stream->index_read_fd > 0) {
                ret = lseek(viewer_stream->index_read_fd,
                                viewer_stream->total_index_received *
                                        sizeof(struct lttng_packet_index),
@@ -721,23 +755,73 @@ int init_viewer_stream(struct relay_stream *stream, int seek_last)
                        viewer_stream->total_index_received;
        }
 
-       /*
-        * This is to avoid a race between the initialization of this object and
-        * the close of the given stream. If the stream is unable to find this
-        * viewer stream when closing, this copy will at least take the latest
-        * value.
-        */
-       viewer_stream->total_index_received = stream->total_index_received;
+       ret = 0;
+
+error:
+       return ret;
+}
+
+/*
+ * Rotate a stream to the next tracefile.
+ *
+ * Returns 0 on success, a negative value on error.
+ */
+static
+int rotate_viewer_stream(struct relay_viewer_stream *viewer_stream,
+               struct relay_stream *stream)
+{
+       int ret;
+       uint64_t tracefile_id;
+
+       assert(viewer_stream);
+
+       tracefile_id = (viewer_stream->tracefile_count_current + 1) %
+               viewer_stream->tracefile_count;
 
+       if (stream) {
+               pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
+       }
        /*
-        * The deletion of this ctf_trace object is only done in a call RCU of the
-        * relay stream making it valid as long as we have the read side lock.
+        * The writer and the reader are not working in the same
+        * tracefile, we can read up to EOF, we don't care about the
+        * total_index_received.
         */
-       viewer_stream->ctf_trace = stream->ctf_trace;
-       uatomic_inc(&viewer_stream->ctf_trace->refcount);
+       if (!stream || (stream->tracefile_count_current != tracefile_id)) {
+               viewer_stream->close_write_flag = 1;
+       } else {
+               /*
+                * We are opening a file that is still open in write, make
+                * sure we limit our reading to the number of indexes
+                * received.
+                */
+               viewer_stream->close_write_flag = 0;
+               if (stream) {
+                       viewer_stream->total_index_received =
+                               stream->total_index_received;
+               }
+       }
+       viewer_stream->tracefile_count_current = tracefile_id;
+       pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
 
-       lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
-       lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
+       if (viewer_stream->abort_flag == 0) {
+               ret = close(viewer_stream->index_read_fd);
+               if (ret < 0) {
+                       PERROR("close index file");
+               }
+               ret = close(viewer_stream->read_fd);
+               if (ret < 0) {
+                       PERROR("close tracefile");
+               }
+       } else {
+               viewer_stream->abort_flag = 0;
+       }
+
+       viewer_stream->read_fd = -1;
+
+       ret = open_index(viewer_stream);
+       if (ret < 0) {
+               goto error;
+       }
 
        ret = 0;
 
@@ -1033,23 +1117,42 @@ int viewer_get_next_index(struct relay_command *cmd,
 
        rstream = relay_stream_find_by_id(vstream->stream_handle);
        if (rstream) {
+               if (vstream->abort_flag) {
+                       /* Rotate on abort (overwrite). */
+                       DBG("Viewer rotate because of overwrite");
+                       ret = rotate_viewer_stream(vstream, rstream);
+                       if (ret < 0) {
+                               goto end_unlock;
+                       }
+               }
                if (rstream->beacon_ts_end != -1ULL &&
                                vstream->last_sent_index == rstream->total_index_received) {
                        viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
                        viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
                        goto send_reply;
                }
-
-               if (rstream->total_index_received <= vstream->last_sent_index) {
+               /*
+                * Reader and writer are working in the same tracefile, so we care
+                * about the number of index received and sent. Otherwise, we read
+                * up to EOF.
+                */
+               pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
+               if (rstream->tracefile_count_current == vstream->tracefile_count_current
+                               && rstream->total_index_received <= vstream->last_sent_index
+                               && !vstream->close_write_flag) {
+                       pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
                        /* No new index to send, retry later. */
                        viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
                        goto send_reply;
                }
-       } else if (!rstream &&
+               pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
+       } else if (!rstream && vstream->close_write_flag &&
                        vstream->total_index_received == vstream->last_sent_index) {
-               /* Last index sent and stream closed */
+               /* Last index sent and current tracefile closed in write */
                viewer_index.status = htobe32(VIEWER_INDEX_HUP);
                goto send_reply;
+       } else {
+               vstream->close_write_flag = 1;
        }
 
        if (!vstream->ctf_trace->metadata_received ||
@@ -1061,8 +1164,30 @@ int viewer_get_next_index(struct relay_command *cmd,
        ret = lttng_read(vstream->index_read_fd, &packet_index,
                        sizeof(packet_index));
        if (ret < sizeof(packet_index)) {
-               PERROR("Relay reading index file");
-               viewer_index.status = htobe32(VIEWER_INDEX_ERR);
+               /*
+                * The tracefile is closed in write, so we read up to EOF.
+                */
+               if (vstream->close_write_flag == 1) {
+                       viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
+                       /* Rotate on normal EOF */
+                       ret = rotate_viewer_stream(vstream, rstream);
+                       if (ret < 0) {
+                               goto end_unlock;
+                       }
+               } else {
+                       /*
+                        * If the read fd was closed by the streaming side, the
+                        * abort_flag will be set to 1, otherwise it is an error.
+                        */
+                       if (vstream->abort_flag != 1) {
+                               PERROR("Relay reading index file");
+                               viewer_index.status = htobe32(VIEWER_INDEX_ERR);
+                               goto send_reply;
+                       } else {
+                               viewer_index.status = htobe32(VIEWER_INDEX_HUP);
+                       }
+               }
+               goto send_reply;
        } else {
                viewer_index.status = htobe32(VIEWER_INDEX_OK);
                vstream->last_sent_index++;
@@ -1155,8 +1280,14 @@ int viewer_get_packet(struct relay_command *cmd)
        if (stream->read_fd < 0) {
                char fullpath[PATH_MAX];
 
-               ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
-                               stream->channel_name);
+               if (stream->tracefile_count > 0) {
+                       ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
+                                       stream->channel_name,
+                                       stream->tracefile_count_current);
+               } else {
+                       ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
+                                       stream->channel_name);
+               }
                if (ret < 0) {
                        goto error;
                }
@@ -1185,14 +1316,32 @@ int viewer_get_packet(struct relay_command *cmd)
 
        ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
        if (ret < 0) {
-               PERROR("lseek");
-               goto error;
+               /*
+                * If the read fd was closed by the streaming side, the
+                * abort_flag will be set to 1, otherwise it is an error.
+                */
+               if (stream->abort_flag == 0) {
+                       PERROR("lseek");
+                       goto error;
+               }
+               reply.status = htobe32(VIEWER_GET_PACKET_EOF);
+               goto send_reply;
        }
        read_len = lttng_read(stream->read_fd, data, len);
        if (read_len < len) {
-               PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
-                               stream->read_fd, be64toh(get_packet_info.offset));
-               goto error;
+               /*
+                * If the read fd was closed by the streaming side, the
+                * abort_flag will be set to 1, otherwise it is an error.
+                */
+               if (stream->abort_flag == 0) {
+                       PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
+                                       stream->read_fd,
+                                       be64toh(get_packet_info.offset));
+                       goto error;
+               } else {
+                       reply.status = htobe32(VIEWER_GET_PACKET_EOF);
+                       goto send_reply;
+               }
        }
        reply.status = htobe32(VIEWER_GET_PACKET_OK);
        reply.len = htobe32(len);
@@ -1522,13 +1671,13 @@ void viewer_del_streams(uint64_t session_id)
                        continue;
                }
 
-               if (stream->read_fd > 0) {
+               if (stream->read_fd >= 0) {
                        ret = close(stream->read_fd);
                        if (ret < 0) {
                                PERROR("close read_fd");
                        }
                }
-               if (stream->index_read_fd > 0) {
+               if (stream->index_read_fd >= 0) {
                        ret = close(stream->index_read_fd);
                        if (ret < 0) {
                                PERROR("close index_read_fd");
index 6fdcca29189edb7e70583326fcebf43d7f1fa9df..8039a7ec703d01f4672d32abcc9448ed49fb017a 100644 (file)
@@ -96,6 +96,8 @@ struct relay_stream {
        uint64_t tracefile_size_current;
        uint64_t tracefile_count;
        uint64_t tracefile_count_current;
+       /* To inform the viewer up to where it can go back in time. */
+       uint64_t oldest_tracefile_id;
 
        uint64_t total_index_received;
        struct relay_viewer_stream *viewer_stream;
@@ -124,12 +126,24 @@ struct relay_stream {
         * timestamp end, when it is active, this field == -1ULL.
         */
        uint64_t beacon_ts_end;
+       /*
+        * To protect the update of the close_write_flag and the checks of
+        * the tracefile_count_current.
+        * It is taken before checking whenever we need to know if the
+        * writer and reader are working in the same tracefile.
+        */
+       pthread_mutex_t viewer_stream_rotation_lock;
 
        /* Information telling us when to close the stream  */
        unsigned int close_flag:1;
        /* Indicate if the stream was initialized for a data pending command. */
        unsigned int data_pending_check_done:1;
        unsigned int metadata_flag:1;
+       /*
+        * To detect when we start overwriting old data, it is used to
+        * update the oldest_tracefile_id.
+        */
+       unsigned int tracefile_overwrite:1;
 };
 
 /*
@@ -147,8 +161,6 @@ struct relay_viewer_stream {
        char *channel_name;
        uint64_t last_sent_index;
        uint64_t total_index_received;
-       uint64_t tracefile_size;
-       uint64_t tracefile_size_current;
        uint64_t tracefile_count;
        uint64_t tracefile_count_current;
        struct lttng_ht_node_u64 stream_n;
@@ -156,6 +168,16 @@ struct relay_viewer_stream {
        struct ctf_trace *ctf_trace;
        /* Information telling us if the stream is a metadata stream. */
        unsigned int metadata_flag:1;
+       /*
+        * Information telling us that the stream is closed in write, so
+        * we don't expect new indexes and we can read up to EOF.
+        */
+       unsigned int close_write_flag:1;
+       /*
+        * If the streaming side closes a FD in use in the viewer side,
+        * it sets this flag to inform that it is a normal error.
+        */
+       unsigned int abort_flag:1;
 };
 
 /*
index 79aecd6b3f41d3145a49ce25d8aa327cfd93550b..fcaa30c49ac0b405a404c9295fa40084a2e8bbb6 100644 (file)
@@ -60,12 +60,14 @@ enum lttng_viewer_next_index_return_code {
        VIEWER_INDEX_HUP        = 3, /* Index closed (trace destroyed). */
        VIEWER_INDEX_ERR        = 4, /* Unknow error. */
        VIEWER_INDEX_INACTIVE   = 5, /* Inactive stream beacon. */
+       VIEWER_INDEX_EOF        = 6, /* End of index file. */
 };
 
 enum lttng_viewer_get_packet_return_code {
        VIEWER_GET_PACKET_OK            = 1,
        VIEWER_GET_PACKET_RETRY         = 2,
        VIEWER_GET_PACKET_ERR           = 3,
+       VIEWER_GET_PACKET_EOF           = 4,
 };
 
 enum lttng_viewer_get_metadata_return_code {
index 0005714c55abf9bdd9a3347bcb96f99304b2ccca..5782175e6e7f15b57753cc29dedebd54db7a51da 100644 (file)
@@ -2039,10 +2039,62 @@ 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) {
+                               vstream->abort_flag = 1;
+                               vstream->close_write_flag = 1;
+
+                               ret = close(vstream->read_fd);
+                               if (ret < 0) {
+                                       PERROR("close index");
+                               }
+
+                               ret = close(vstream->index_read_fd);
+                               if (ret < 0) {
+                                       PERROR("close tracefile");
+                               }
+                               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);
+               pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
                if (ret < 0) {
                        ERR("Rotating stream output file");
                        goto end_rcu_unlock;
This page took 0.033204 seconds and 4 git commands to generate.