Skip to main content
← Back to list
01Issue
BugShippedSwamp CLIPublic
AssigneesNone

Relationships

↔ sibling #4693

#4703 serve: automatic continuation follow-ups (scheduler rebuild overlap, reconciliation retry, nested refusal re-judge)

Opened by stack72 · 11/13/2026· Shipped 11/14/2026

Follow-up from #4693 (automatic continuation of interrupted runs).

When serve continues an interrupted scheduled run, it registers the run with the scheduler (ScheduledExecutionService.adoptContinuedRun) so the next fire of that workflow is skipped as an overlap. That registration lives only in the in-memory continued map of the current scheduler instance.

Anything that stops or rebuilds the scheduler drops it: swamp serve reload toggling scheduling, scheduler disable then enable, or a config reload that rebuilds the service through ScheduledExecutionLifecycle. The continued run keeps running, because serve's active run registry drives it, but the next cron fire finds the workflow in neither running nor continued and starts a second run beside it. The same gap applies when scheduling is off or the instance is a follower at the time of the continuation, and later turns on or becomes leader.

Expected: while a continued scheduled run is active, no fire of the same workflow starts, across scheduler rebuilds.

Suggested fix: when the scheduler starts or is rebuilt, re-adopt active workflow-resume entries from the active run registry whose run has triggerSource schedule, or have the overlap check ask the registry directly.

Second gap from the same review: a continuation the reconciliation tick starts for the runs it just interrupted is not retried if it falls through for a passing reason (the record was still being pushed, so the currency check failed; serve was draining). The run stays interrupted until the next boot or leader change. A first fix that judged every stamped run on each tick was reverted, because it re-read the whole interrupted-run history every tick. A bounded retry set (runs a pass could not act on, read by id) would close it.

Third gap: a run refused as nested_run_unsettled is judged again only at the next boot or leader change. On a single long-running serve the parent stays interrupted after its nested run settles (its refusal names swamp workflow recover). The nested-run settle path, or the retry set above, should re-judge such parents. The design doc's Recovery section says later passes judge it again; it should say the next boot or leader change until this lands.

02Bog Flow
✓OPEN✓TRIAGED✓IN PROGRESS✓SHIPPED+ 9 MOREPR_LINKED+ 1 MORESESSION_SUMMARIZED

Shipped

11/14/2026, 10:40:39 AM

Click a lifecycle step above to view its details.

03Sludge Pulse
stack72 linked sibling of #469311/13/2026, 1:38:26 PM

Sign in to post a ripple.