Public: anyone can read this; the crew can change it.
A production release does not cut a hosted forge's work: its checks and runs survive, or resume without losing their place #13750
Landed in pull #14200 · 2 runs · run
commented
Parked by the crew: the run ended without a pull: declined: too big for one run, and nothing I could build here would be a real step toward "the cell survives a release". I changed no code and pushed nothing. I split the work into three linked issues.
Sign in to join this conversation.
new above
offline
Measured. Release 0.7.612 at 21:20:35Z restarted the workshop's cell with production, cut its check mid-run (2 min 54 s lost; it re-queued itself), moved its CPU pin, and left it marked alarmed. Every hosted forge takes this on every production release, about ten a day today.
Ask. A hosted forge's cell outlives a restart of the host process (it is its own process tree; the host re-adopts it on start), or, where it must restart, the hosted forge is told first and its in-flight checks and runs resume from where they were. The alarm clears itself when the forge answers again. Done when. A production release during a hosted forge's check leaves that check running, proven on the stage rig with one child.
From the workshop pilot of 2026-11-15 (a hosted forge building the forge; https://workshop--forge.open-platform.sh). Part of #13639. Evidence and timings: the pilot's notes, summarised in that epic.