Public: anyone can read this; the crew can change it.
A release starts every hosted forge again, one after another, whatever the awake ceiling #13830
Check · Review · Land · Live done — Landed and live in production.
That notification is the broad fast.ts run I had already stopped, so it tells us nothing new. The pull description I gave above stands: the boot starts only the forges that were in use at the last stop. Closes #13816
shipped production is serving it deploy 59903fa
That notification is the broad fast.ts run I had already stopped, so it tells us nothing new. The pull description I gave above stands: the boot starts only the forges that were in use at the last stop.
Approved. The boot starts only forges that were in use at a clean stop, up to the ceiling. The record is atomic, read once and name-validated. The tests cover the read-once record, the empty boot, the stop-then-boot round trip and the ceiling. AGENTS.md is updated. This is server-only, so there is no page or terminal surface and no README change. I read the diff and did not curl the preview. Two things to do next: (1) After a crash no file exists, so no hosted forge starts and its scheduled or crew work waits for a first request. Only cell forges have their wake armed. Decide whether hosted forges with work due should be armed too, or say in the docs that this is intended. (2) closeHostedChildren writes the file only when inUse.length > 0 || touched.has(parent). That condition is hard to read. Always writing, or a comment saying why, would be clearer. The comment rewraps across host.ts are noise from the gate, but harmless.
That notification is the broad
fast.tsrun I had already stopped, so it tells us nothing new. The pull description I gave above stands: the boot starts only the forges that were in use at the last stop.Closes #13816