render: cap startToClose at 4h — the tasks default (~1h) reaped live renders

Liveness is the 1200s heartbeat; the cap only bounds a live-but-stuck render. Observed: activities reaped at ~68m mid-render, queue re-ran the same job six times, results stranded on the node.
2026-07-16 19:31:10 -07:00
1 changed files with 4 additions and 0 deletions
@@ -129,6 +129,10 @@ def dispatch_if_worker(request, org_id: str, prompt_id: str, prompt: dict) -> bo
            # sample) can exceed 600s; the worker heartbeats each step, so this is
            # the ceiling for a genuinely long single render, not idle slack.
            "heartbeatTimeout": "1200s",
+            # Liveness is the heartbeat above; this cap only bounds a live-but-stuck
+            # render. Unset, the tasks default (~1h) reaped real renders mid-run and
+            # the queue re-ran them for hours.
+            "startToCloseTimeout": "14400s",
            "input": {
                "prompt": prompt,
                "org": org_id,