Fix: relayd: failure to open chunk files concurrently with session clear

[lttng-tools.git] / src / bin / lttng-relayd / live.cpp

diff --git a/src/bin/lttng-relayd/live.cpp b/src/bin/lttng-relayd/live.cpp
index 92c3bf4a7831d8ac6195009e08af8dc5e6fceee0..9cdc1c8d751d78b0876d1cd35c88026b72b4c6cf 100644 (file)
--- a/src/bin/lttng-relayd/live.cpp
+++ b/src/bin/lttng-relayd/live.cpp
@@ -1167,6 +1167,16 @@ int viewer_get_new_streams(struct relay_connection *conn)
         * the viewer's point of view.
         */
        pthread_mutex_lock(&session->lock);
+       /*
+        * If a session rotation is ongoing, do not attempt to open any
+        * stream, because the chunk can be in an intermediate state
+        * due to directory renaming.
+        */
+       if (session->ongoing_rotation) {
+               DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
+               response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_NO_NEW);
+               goto send_reply_unlock;
+       }
        ret = make_viewer_streams(session,
                        conn->viewer_session,
                        LTTNG_VIEWER_SEEK_BEGINNING, &nb_total, &nb_unsent,
@@ -1307,6 +1317,17 @@ int viewer_attach_session(struct relay_connection *conn)
                goto send_reply;
        }
 
+       /*
+        * If a session rotation is ongoing, do not attempt to open any
+        * stream, because the chunk can be in an intermediate state
+        * due to directory renaming.
+        */
+       if (session->ongoing_rotation) {
+               DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
+               send_streams = 0;
+               goto send_reply;
+       }
+
        ret = make_viewer_streams(session,
                        conn->viewer_session, seek_type,
                        &nb_streams, NULL, NULL, &closed);
@@ -1661,9 +1682,9 @@ int viewer_get_next_index(struct relay_connection *conn)
         * This allows clients to consume all the packets of a trace chunk
         * after a session's destruction.
         */
-       if (conn->viewer_session->current_trace_chunk != vstream->stream_file.trace_chunk &&
+       if (!lttng_trace_chunk_ids_equal(conn->viewer_session->current_trace_chunk, vstream->stream_file.trace_chunk) &&
                        !(rstream->completed_rotation_count == vstream->last_seen_rotation_count + 1 && !rstream->trace_chunk)) {
-               DBG("Viewer session and viewer stream chunk differ: "
+               DBG("Viewer session and viewer stream chunk IDs differ: "
                                "vsession chunk %p vstream chunk %p",
                                conn->viewer_session->current_trace_chunk,
                                vstream->stream_file.trace_chunk);
@@ -2006,11 +2027,11 @@ int viewer_get_metadata(struct relay_connection *conn)
                 * an error.
                 */
                if (vstream->metadata_sent > 0) {
-                       vstream->stream->no_new_metadata_notified = true;
-                       if (vstream->stream->closed) {
+                       if (vstream->stream->closed && vstream->stream->no_new_metadata_notified) {
                                /* Release ownership for the viewer metadata stream. */
                                viewer_stream_put(vstream);
                        }
+                       vstream->stream->no_new_metadata_notified = true;
                }
                goto send_reply;
        }
@@ -2032,8 +2053,8 @@ int viewer_get_metadata(struct relay_connection *conn)
        }
 
        if (conn->viewer_session->current_trace_chunk &&
-                       conn->viewer_session->current_trace_chunk !=
-                                       vstream->stream_file.trace_chunk) {
+                       !lttng_trace_chunk_ids_equal(conn->viewer_session->current_trace_chunk,
+                                       vstream->stream_file.trace_chunk)) {
                bool acquired_reference;
 
                DBG("Viewer session and viewer stream chunk differ: "