v1.11.0-rc.3 (Release Candidate)
Pre-releaseThird release candidate for 1.11.0.
Changed since rc.2 — one fix. The log fingerprint now also lands in a log file
that already exists. On rc.2 it was written only into a brand-new file, so an
instance upgrading with a populated /config/sublarr.log stayed unidentifiable
until its log happened to rotate — exactly the case the feature exists for. Found
by inspecting the deployed rc.2 container rather than by a test: every test used a
fresh file, so the suite was green while the feature did nothing in practice.
The RC as a whole is about making a log worth sending.
- Every log file carries a line naming version, platform, container, DB backend,
deployment mode and rotation window — at startup and on every rotation. - The request id appears in ordinary log lines, not only on the error line.
- The log rotation setting finally applies. It could be saved, the UI showed
it, the API promised it took effect on restart — and nothing read it; the handler
was pinned to 5 MB x 3. At DEBUG that window reached back about twenty minutes. - An unconfigured provider logs at INFO, not WARNING. In one real user log, 91% of
all WARNING lines were that single benign class. - The foreign-tracks sweep runs in batches and resumes where it stopped. Ships
disabled. - The support bundle no longer omits rotated log files past the third, and long
scheduler jobs are no longer killed by a worker timeout set below their runtime.
BREAKING: GET /api/v1/logs/download returns the anonymised support bundle
(ZIP) instead of the raw log. Pass ?raw=1 for the old behaviour.
Multi-arch (amd64 + arm64). Full backend suite: 6184 passed, 0 failed.