Every replica runs the transcode cleanup task against the same shared transcode
directory, so one pod deletes segments another pod is still streaming.
- read the active sessions from ITranscodeSessionStore before deleting
- keep files matching an active session manifest path or segment prefix
- abort the sweep when the store cannot be reached
Explicitly set StandardErrorEncoding and StandardOutputEncoding to
Encoding.UTF8 when invoking the FFmpeg subprocess.
This prevents log encoding issues and character corruption on Windows
environments that default to non-UTF8 ANSI code pages.
This fixes garbled metadata and font names in the FFmpeg logs.
Signed-off-by: nyanmisaka <nst799610810@gmail.com>
Address review feedback:
- Replace typeof(Person).FullName with IItemTypeLookup.BaseItemKindNames
- Replace foreach+ToListAsync with PartitionEagerAsync for batched
iteration with built-in progress reporting
- Check HasImage/HasOverview on the loaded domain Person object
instead of projecting from the DB query
Instead of re-checking image/overview on the domain object after loading,
project the values directly from the database query as part of the
anonymous type selection. This avoids redundant checks since the DB
already has this information.
Instead of loading all people names and checking each one in memory,
query the database directly for Person items that need refresh:
- Missing primary image OR missing overview
- Not refreshed within the last 30 days
This reduces the operation from N+1 queries (1 for all names + 1 per
person to load) to a single filtered query returning only the IDs that
need work.
Move image refresh logic from PeopleValidator (which runs during library
scans) into PeopleValidationTask (the "Refresh People" scheduled task).
This keeps library scans fast while ensuring the scheduled task fetches
missing images from remote providers like TMDB.
People missing a Primary image or overview get refreshed with
MetadataRefreshMode.Default instead of ValidationOnly, with a 30-day
cooldown to avoid hammering providers for people they have no data for.
Fixes jellyfin#8103
This commit integrates remaining performance changes:
- Add batch user data fetching in DtoService to reduce N+1 queries
- Add GetNextUpEpisodesBatch in TVSeriesManager for efficient batch retrieval
- Update Video/Movie/BoxSet to use LibraryManager for alternate versions
- Transition LinkedChild to use ItemId instead of Path (obsolete Path/LibraryItemId)
- Update providers and controllers for LinkedChildren-based references
- Add NextUpEpisodeBatchResult for batched episode queries
- Integrate IDescendantQueryProvider in SqliteDatabaseProvider
Follow TrickPlay example of running other background ffmpeg tasks as ProcessPriorityClass.BelowNormal:
- Keyframe extraction
- Media info probing during library scans
- Audio normalization
- Image extraction
unkin-agent
Cody Robibero