v31.0

docling-project/doclingv31.0May 19, 2026by pierre-b

AI Summary

Contains database schema optimizations and fixes to prevent trigger loops, correct recurring task timing, and improve connection pool health checks.

Key Highlights

Optimized trigger function for segment recomputation to prevent self-loops.
Recurring tasks now correctly write and compare `timeout_after` in UTC.
Connection pool health check now uses an isolated context to prevent premature eviction.

New Features

Trigger optimization for contact timeline events.
UTC time fix for HTTP dispatched recurring tasks.
Isolated context for database pool health checks.

Full Release Notes

### Database Schema Changes

- Migration v31.0 updates the `queue_contact_for_segment_recomputation` trigger function on every workspace database to short-circuit when the inserted `contact_timeline` row is itself a segment membership event (`kind IN ('segment.joined', 'segment.left')`).

### Fixes

- **Fix**: `queue_contact_for_segment_recomputation` trigger no longer re-enqueues contacts when the inserted `contact_timeline` event is itself a segment membership change (`segment.joined`/`segment.left`). Removes a self-loop where every membership write re-queued the same contact.
- **Fix**: Recurring tasks dispatched via HTTP now write `timeout_after` in UTC. The column is `TIMESTAMP WITHOUT TIME ZONE` and the scheduler compares it against `time.Now().UTC()`; on non-UTC hosts the local-time value caused the task to appear "still running" for the host's UTC offset. Same fix applied to the broadcast-pause `next_run_after`.
- **Fix**: `GetWorkspaceConnection`'s pool health check now uses an isolated context for `pool.PingContext` instead of the caller's. A caller-context cancellation no longer triggers pool eviction.