Skip to content

fix: recover model instances after host reboot - #13

Merged
dmytrogajewski merged 1 commit into
mainfrom
fix/reboot-instance-recovery
Oct 4, 2026
Merged

dmytrogajewski merged 1 commit into
mainfrom
fix/reboot-instance-recovery

Conversation

@dmytrogajewski

Copy link
Copy Markdown
Contributor

When Docker fails to restart a model during host boot, Sparkplane now recognizes the exited container without a live endpoint or process and recreates the exact generation using its existing engine and artifact identities. Recovery remains deferred while admission telemetry is missing or memory pressure is present.

Agent launch now waits for an existing instance with running intent instead of requesting another model copy and failing aggregate admission. Explicit instance selection is honored, dry runs report wait, and cold-start waits use the bounded engine startup deadline.

Validation: make lint, make test, make test-client, and make audit. Regression coverage includes exited-container inspection, recovery over real typed Unix IPC, admission deferral, and pinned HTTPS waits that reject changed generations and stopped intent.

@dmytrogajewski
dmytrogajewski merged commit 6404d59 into main Oct 4, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants