What's Changed
Bug Fixes
- Fixed an issue where uploading large or slow files could result in the video never appearing in the library after upload completed successfully
- Fixed a race condition that could result in multiple workers attempting to transcode the same file
- Fixed an issue where transcoding the same video multiple times simultaneously could cause corrupted output or failed transcodes
- Fixed an issue where transcoding would silently fail if the job was interrupted by a container restart. Affected jobs are now automatically retried on next startup
Improvements
- Transcoding jobs are now queued in the database, so manually triggered transcodes from any session are no longer lost if the server handles the request on a different internal worker
- Video library scans no longer skip or get blocked while a long transcode is in progress
- Servers with high core-count CPUs (e.g. 32-core Threadripper) no longer spawn an excessive number of worker processes, which could exhaust container PID limits and prevent startup. By default, Fireshare will us at most 4 workers with 8 threads each. If you want more or less workers than the defaults worker count can now be tuned via environment variables (
GUNICORN_WORKERS,GUNICORN_WORKER_CAP,GUNICORN_THREADS).