[MM-57211] Fix issues with multiple transcription jobs per call #657
Conversation
| @@ -109,7 +109,7 @@ func (p *Plugin) saveRecordingMetadata(postID, recID, trID string) error { | |||
| trID: tm.toMap(), | |||
| } | |||
| } else { | |||
| recordings[trID] = tm.toMap() | |||
There was a problem hiding this comment.
This is the bug, an innocent typo.
| @@ -17,21 +17,21 @@ import ( | |||
|
|
|||
| var callRecordingActionRE = regexp.MustCompile(`^\/calls\/([a-z0-9]+)/recording/(start|stop|publish)$`) | |||
|
|
|||
| const recordingJobStartTimeout = 2 * time.Minute | |||
There was a problem hiding this comment.
Lowering this a bit since 2 minutes is just too much for a user to be waiting.
| // This is needed as we don't yet handle wsEventCallTranscriptionState on | ||
| // the client since jobs are coupled. | ||
| recClientState.Err = "failed to start transcriber job: timed out waiting for bot to join call" | ||
| p.publishWebSocketEvent(wsEventCallRecordingState, map[string]interface{}{ | ||
| "callID": callID, | ||
| "recState": clientState.toMap(), | ||
| }, &model.WebsocketBroadcast{ChannelId: callID, ReliableClusterSend: true}) |
There was a problem hiding this comment.
This is something I forgot to add originally. If the transcribing job fails first then we wouldn't notify the host as we are not handling the below wsEventCallTranscriptionState event.
| @@ -26,7 +26,7 @@ func (p *Plugin) transcriptionJobTimeoutChecker(callID, jobID string) { | |||
|
|
|||
| trState, err := state.getTranscription() | |||
| if err != nil { | |||
| p.LogError("failed to get transcription state", "error", err.Error()) | |||
| p.LogWarn("failed to get transcription state", "err", err.Error(), "callID", callID, "jobID", jobID) | |||
There was a problem hiding this comment.
Not asking for a change, but just curious why the change to LogWarn? If we return, it seems like an error makes sense?
There was a problem hiding this comment.
Good question. The reason is that in most cases that's not really a sign of a problem. For example it will happen in any of these cases:
- If host ends a call before the bot gets a chance to join.
- If host stops and restart a recording before the first one started.
- If any of the two coupled jobs (transcriptions and recording) fails as we also stop the other but don't explicitly cancel the timeout checker.
There was a problem hiding this comment.
Ahh, okay, that makes sense. Thanks.
streamer45
added
3: Reviews Complete
Summary
In case of multiple transcription jobs per call we were not writing the metadata correctly which would cause failure to attach and render the captions.
Ticket Link
https://mattermost.atlassian.net/browse/MM-57211