Skip to content

v2.7.24 — Bazarr now hears about subtitles from Requeue, Regenerate and Scan, and large libraries no longer stall the app

Latest

Choose a tag to compare

@github-actions github-actions released this 22 Oct 08:05
· 2 commits to main since this release
2cbf5df

Bazarr now hears about subtitles from Requeue, Regenerate and Scan, and large libraries no longer stall the app.

Fixed

  • The Dashboard stalled the whole app on large libraries (#621). To show one count and the most recently probed file, the "discovered" tile loaded every row of the probe cache into memory on each 30-second refresh. At about 40,000 files that took over a second of Python work, and everything else waiting on subarr at that moment hitched with it, which showed up as a stream of loop_lag warnings. The tile now asks the database for just those two things: about 1 ms on a 40,000-row table, measured, against 750-1,200 ms before, using a new database index (applied automatically on upgrade). The tile's size, which had always read as 0, now shows the file's real size. The Review page's verification lookups also moved off the request loop. Thanks to @caracturus for the report and the stack traces that pinned it down.
  • Short app-wide pauses while the library was being scanned (#627). A week of loop_lag warnings on a 6,600-file install traced 28 of 34 stalls to database and file-system work done on the thread that answers every request: saving the coverage snapshot after each refresh, checking each file while probing new media, and counting probed files for telemetry. That work now runs in the background, so the UI and API stay responsive while it happens. Nothing to configure.
  • Bazarr was never told about a subtitle from Requeue, Regenerate or a Scan-tab submit (#629). Those three actions queued the file without its Sonarr or Radarr ids, so when the job finished subarr could neither upload the subtitle to Bazarr nor ask Bazarr to rescan, and said nothing about skipping it. Bazarr kept listing the episode or movie as wanted. A file that still needs a subtitle now gets its ids when it is queued, from the library subarr already holds. Any other job that finishes without ids, such as a file that already had a subtitle, now asks Sonarr or Radarr for them before writing back to Bazarr, and keeps them on its history row. If neither knows the file (a folder, or a file renamed since), the log says that Bazarr was not notified and why.