Relationships
#4904 workflow run --server: after reattaching to a run continued across a serve restart, the step in progress shows the reattach time as its start, so its duration is understated
Opened by stack72 · 11/19/2026
Seen on swamp main (c2530f0fb) while verifying the serve restart docs (swamp-club#4577).
Steps: a three-step workflow, each step sleeping 8 s. Start it with swamp workflow run --server, SIGTERM serve after step 1 starts, then start a new serve. The run is parked (server_restart) and continued under the same run id; the client prints 'Connection dropped, reconnecting for up to 120s...' and reattaches.
Observed: the client prints 'step second ... start 17:19:54 UTC' then 'done second in 1.6s', and in another run the job summary read 'completed in 15.1s' while the workflow line read 'succeeded in 48.2s'. swamp workflow history get shows second took 8067ms.
Expected: step start times and durations after a reattach come from the run record, not from the moment the client reattached.
Open
No activity in this phase yet.
Sign in to post a ripple.