Relationships
↔ sibling #4577#4794 Docs: a planned swamp serve restart hands in-flight runs over instead of interrupting them
Opened by stack72 · 11/16/2026· Shipped 11/19/2026
swamp-club#4737 changes what a planned swamp serve restart (SIGTERM, a rolling restart included) does to in-flight workflow runs. Serve stops starting new steps, lets running steps finish within the shutdown drain timeout, then parks each run at a step boundary (recorded interrupted with reason server_restart and no unknown step). The new serve process continues parked runs under the same run id once the old process has gone. Clients attached with workflow run --server keep streaming and see the run finish, and the dashboard shows the run as parked, then continuing. A step still running at the drain timeout is interrupted as before and continued only when it has a guard; otherwise it waits for swamp workflow recover. autoContinue false keeps a parked run until resumed by hand.
Pages to update: content/manual/reference/swamp-serve.md (shutdown and the drain timeout), content/manual/how-to/swamp-serve.md (rolling restarts), and content/manual/reference/workflows.md (autoContinue and the server_restart interruption reason).
Shipped
Click a lifecycle step above to view its details.
Sign in to post a ripple.