pacnpal/compressatorium v3.7.5

Release v3.7.5 - Post-sync auto-rematch + hardened error paths

on GitHub

v3.7.5 - Post-sync auto-rematch + hardened match-job error handling

✨ New Features

• DAT re-syncs now auto-rematch previously-scanned files. When a DAT re-sync wipes the match cache, the sync now snapshots the pre-wipe path list and enqueues a background /dat/match-batch/job against those paths. Before this, GET /dat/stats would read matched=0, scanned=0 immediately after a sync until the user manually browsed every affected directory. The stats counter now refreshes on its own within a few minutes of sync completion.
• schedule_match_job() helper. Extracted from the /dat/match-batch/job HTTP handler so non-HTTP callers (like the post-sync hook) can reuse the same concurrency-guarded scheduling path. Works with or without a FastAPI BackgroundTasks.

🐛 Bug Fixes & Hardening

• Failure counting in _run_match_job. Previously a dropped volume produced a misleading "complete, 0 matched" signal. Two new counters (errors + skips) now track real failures vs. policy skips. When errors == total the job flips to failed with a "check volume accessibility" error message.
• Mid-loop exception preserves counter context. Error message now includes processed N/M, K error(s) so operators know how far the job got.
• Programmer-error tracebacks surface. Two broad except Exception blocks now use logger.exception (ERROR level with full traceback) rather than logger.warning.
• Background match-task exceptions surface too. Exceptions past _run_match_job's own try/finally are now routed through the project logger instead of only appearing as asyncio's "Task exception was never retrieved" warning on GC.
• list_match_paths failure no longer leaks staged DATs. Snapshot failure is now best-effort; the DAT import stays durable.
• Rematch ImportError no longer silently swallowed. A missing-module deployment bug now fails the sync loudly.
• schedule_match_job scheduling failure rolls back _active_match_job_id. Without this, future match jobs would return HTTP 409 forever after a failed schedule.
• Canonical internal imports. Removed three try: from services.x / except ImportError: from app.services.x fallback pairs that could silently absorb real circular-import bugs. Production-only feature-gate imports in archive.py / disc_id.py untouched.

🧪 Tests

308 tests pass (+14 new since v3.7.3), covering the rematch hook, error counting, threshold behavior, traceback logging, scheduling-failure rollback, and snapshot-failure resilience.

⚠️ Upgrade Notes

• No action required. After restart, stale-DAT self-heal (v3.7.3) triggers if needed, and the follow-up rematch now happens automatically.
• Jobs panel: after a sync you'll see a "DAT Match" external job tick through your previously-scanned files. If a volume is offline while it runs, the job finishes with failed + "all N file(s) failed — check volume accessibility" instead of a misleading "0 matched" success.