Platform

This release fixes the abort of jobs, report runs, and assistant chats, adds a Celery health probe to the web portal, and fixes several background job and replication problems.

Reliable job and report aborts

Improvement for administrators.

Aborting a job run, a report run, or an AI assistant chat now takes effect whether the task is queued or already running. Before, an abort of a queued task was lost and the task ran to completion, aborting wiped the stored progress, and an assistant reply was saved anyway. The run shows the new Aborting status until the worker confirms Aborted, and the progress and results accumulated so far are kept. Clicking Abort again on a run that no worker owns closes it as aborted, which repairs runs orphaned by a dead worker. A queued report run shows Starting with an Abort button before pickup.

A report run right after Abort, before the worker confirms.

See Jobs console.

Celery health probe endpoint

New feature for system administrators.

GET /health/celery/probe enqueues a task and waits for a worker to process it, so a monitor can tell a live web process from a working job pipeline. Access is limited to the addresses and networks listed in observability.celery_probe.hosts. The default production configuration allows 127.0.0.1 and ::1. If the key is empty or missing, the probe rejects every request. The probe and the existing /health and /health/ready endpoints have configurable rate limits. See System status.

Fixes

Import and call sync jobs (Webex, NICE inContact, Twilio) and the transcription results check no longer process the same record twice when a retry is due immediately.
Merge-mode replication keeps evaluation reports on the target that a restricted token skipped, instead of deleting them as intentionally omitted.
Delete and Delete Errors work on the incoming replication errors page.
An oversized upload is rejected before it is read into memory, with an inline "File is too large" error instead of a server error.