← Back to list
01Issue
BugShippedSwamp CLIPublic
Assigneesstack72
Relationships
#4518 Serve restart after kill -9 leaves the killed run reading running; only CLI recover's dead-owner check settles it
Opened by stack72 · 11/8/2026· Shipped 11/8/2026
What happens
After kill -9 of a native swamp serve mid-run, restarting serve does not mark the killed workflow run interrupted. The serve log shows the boot pass ran (Boot: reaping stale runs via tracker), but the run record still reads status: running 60 seconds after the restart. The run is only settled later, when a CLI swamp workflow recover <wf> --acknowledge-unknown finds its owner dead and marks it with interrupt_reason: owner_process_dead.
Expected: serve's boot reaper marks the run interrupted (server_crash), so a plain workflow recover or a serve-side recovery path sees it without relying on the CLI's dead-owner check.
Reproduction
- Scratch repo,
swamp serve --no-schedule --port <p>(auth none) and twoswamp worker connectprocesses. swamp workflow run <wf> --server http://127.0.0.1:<p>with a step that sleeps several seconds.- While the step runs:
kill -9the serve process (and the workers). - Start serve again on the same repo and watch the run record: it stays
status: running. - Stop serve;
swamp workflow recover <wf> --acknowledge-unknownrecovers it (via the dead-owner settle), and a resume through serve completes.
Notes
- Seen with a fan-out workflow during the swamp-club#4516 live check, and the same check for swamp-club#4506 hit it too (no interrupted run found after restart). Neither change touches the reaper (
src/serve/orphaned_run_reaper.ts,src/serve/boot_reconciliation.ts), so it predates both. - Not yet checked: whether it is specific to fan-out runs or to native serve on macOS, and how the tracker-based reaper decides a run is stale.
02Bog Flow
Shipped
Click a lifecycle step above to view its details.
03Sludge Pulse
Sign in to post a ripple.