Summary
With per-session isolated task queues (isolateAdvWorkerTaskQueues), a change workflow is pinned to the task queue of the session that created it (advance-<projectId>-sess_<ptr>). When that session ends (or its worker exits), no live worker polls that queue, so any tool that must query/signal the workflow fails fast with ServiceError: Failed to query Workflow — even though the workflow itself is healthy and the Temporal server is fine.
Incident evidence (2026-07-23)
- Change
fixDurableProofFallback created in session sess_ZH2cWlqT; workflow Running, healthy (HistoryLength 140, StateTransitionCount 82).
- After an unrelated session restart,
temporal task-queue describe on advance-bdf259aa162ae192af5b18899ccdc653b085528d-sess_ZH2cWlqT showed zero pollers.
adv_gate_status / adv_change_show returned ServiceError: Failed to query Workflow (fast, not timeout). adv_change_list still worked (reads server-side search attributes, no worker query).- Manual recovery:
adv_worktree_resume changeId: fixDurableProofFallback re-hosted it onto the live session; queries then succeeded.
Expected
A live worker should auto-adopt an orphaned session queue whose owning session is gone, so healthy workflows remain queryable/signalable without a manual adv_worktree_resume. Unsafe/ambiguous ownership must remain fail-closed (no cross-session hijack of a queue still served by a live peer).
Status
This is the target behavior of the in-flight change autoAdoptOrphanSessionQueues (Epic hardenTemporalReliability, execution phase, 1/6 tasks done). Filing this issue as a GitHub-visible reproduction + evidence record and to track closure against that change. Not a competing proposal.
Related
isolateAdvWorkerTaskQueues (archived) — introduced per-session queues.list-orphan-session-queues.ts, orphan-queue-adopter.ts (source).- Companion incident: stale plugin-host generation opaque-timeout (separate issue, same 2026-07-23 session).
Summary
With per-session isolated task queues (
isolateAdvWorkerTaskQueues), a change workflow is pinned to the task queue of the session that created it (advance-<projectId>-sess_<ptr>). When that session ends (or its worker exits), no live worker polls that queue, so any tool that must query/signal the workflow fails fast withServiceError: Failed to query Workflow— even though the workflow itself is healthy and the Temporal server is fine.Incident evidence (2026-07-23)
fixDurableProofFallbackcreated in sessionsess_ZH2cWlqT; workflow Running, healthy (HistoryLength 140, StateTransitionCount 82).temporal task-queue describeonadvance-bdf259aa162ae192af5b18899ccdc653b085528d-sess_ZH2cWlqTshowed zero pollers.adv_gate_status/adv_change_showreturnedServiceError: Failed to query Workflow(fast, not timeout).adv_change_liststill worked (reads server-side search attributes, no worker query).adv_worktree_resume changeId: fixDurableProofFallbackre-hosted it onto the live session; queries then succeeded.Expected
A live worker should auto-adopt an orphaned session queue whose owning session is gone, so healthy workflows remain queryable/signalable without a manual
adv_worktree_resume. Unsafe/ambiguous ownership must remain fail-closed (no cross-session hijack of a queue still served by a live peer).Status
This is the target behavior of the in-flight change
autoAdoptOrphanSessionQueues(EpichardenTemporalReliability, execution phase, 1/6 tasks done). Filing this issue as a GitHub-visible reproduction + evidence record and to track closure against that change. Not a competing proposal.Related
isolateAdvWorkerTaskQueues(archived) — introduced per-session queues.list-orphan-session-queues.ts,orphan-queue-adopter.ts(source).