forge: bump forgejo-nsc to 82e3da07 (stop-monitor startup window fix + daemon TTL drain) #56
No reviewers
Labels
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
key-store/key.store!56
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "forge/forgejo-nsc-stop-monitor"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Pulls in key-store/forgejo-nsc#1 (merged as
82e3da07).Why: on the forge, 125 of 176 Namespace runner launches in the 24 h to 2026-08-16 10:00Z were destroyed 2–5 s after
nsc run --waitreturned. Reproduced as root: for the first ~3–6 snsc describe --output jsonreports the resource with nocontainerkey, andinstanceStopped()treated that as "all containers stopped". Also fixes the related hazard where daemon-mode runners accepted jobs right before their TTL (job 97045 / run 17183).What changes on the forge after deploy (
forgejo-nsc-dispatcher.service):nsc describeevery 10 s; only a tombstone / stopped container (or "not found" after a container was seen or 2 m grace) ends the wait; transient describe errors retry. New Info logs:runner first stop check(raw JSON),runner instance phase changed,runner instance reported gone.FORGEJO_RUNNER_TTL_DEADLINE_EPOCH+FORGEJO_RUNNER_DRAIN_SECONDS(TTL/3 clamped 5m..30m → 30 m for the 90 m amd64/arm64-large targets, 5 m for 15 m small) → SIGTERMforgejo-runner daemonat deadline−drain withshutdown_timeout = drain−60 s. Override per target viaenv.FORGEJO_RUNNER_DRAIN_SECONDS.go.mod/go.sumunchanged upstream → samevendorHash; verified locally withnix build .#forgejo-nsc-dispatcher.Rollout: deploy the forge, then re-run the journalctl census (launched → destroyed within 15 s should drop to ~0 while jobs still get runners).
93766a2f589abe479091