carlos / platform Public

I found this during the v0.26.1 flagship roll on 2026-10-13. It's from before the roll and isn't a regression. It has no owner yet: platform-1e confirmed it isn't the claim's.

What happens: on annie-welding (i-092c0c1eea75723cb, eu-west-1), the pre-roll v0.25.1 agent logged layout: ... refused -> db is missing at <path> for four hosts:

• lpay.bde
• probe1.bab
• two p-*.bdy hosts

I haven't captured the first-seen timestamp. Whoever takes this should run journalctl -u carlos-instance-control | grep 'db is missing' on the box.

Where it comes from: internal/layout/pass.go:354 (migrateHost). The check runs before quiesce, so a host whose database file isn't on this box is refused without being stopped. That is by design (see the comment above it).

To find out:

1. Do these four hosts have a local database at all? They could be hibernated with the DB only in the bucket, test leftovers (probe1, p-*), or registry rows for apps that are gone.
2. If the refusal is the correct steady state, should the pass skip such hosts quietly instead of logging a refusal every pass?
3. If any of them is a real, live app whose database was lost, that is urgent.

I'll leave this unclaimed for whoever owns the layout migration.