Rename the old CHAR(36) id to uuid (still used for public URLs and file
naming) and add a real auto-increment id as the primary key. Track
input/output file size and conversion duration per job. Cleanup no
longer deletes rows; it marks cleaned_at and skips already-cleaned
expired jobs on later runs.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Also adds vitest.config.js with fileParallelism: false. Every DB-backed
test file wipes and reseeds the shared conversion_jobs table in
beforeEach; running test files in parallel (Vitest's default) let one
file's DELETE race another file's just-inserted row against the same
live MariaDB instance, causing intermittent cross-file failures.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Regenerates the sample.png test fixture via sharp itself: the
hand-encoded base64 PNG from the plan had valid magic bytes (enough to
fool file-type's signature check) but was malformed past the header,
which libpng rejected as soon as sharp actually tried to decode it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Also configures the MariaDB pool with timezone: 'auto', since the
default 'local' mode sends dates without timezone conversion and
silently broke expires_at comparisons whenever the app host and DB
server clocks differ (caught by the findExpiredJobs test).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
anthonyandClaude Sonnet 5