Resolved. A self-hosted sfp-server host (running on Podman) became unreachable. Root cause: after the host was moved to the Docker Compose v2 provider, a leftover Compose v1 stack remained, and a duplicate logging container began crash-looping on a shared file lock. The resulting restart storm exhausted the host's container/init machinery until it dropped off the network and could only be recovered via the serial console. This was not an out-of-memory event and not the Postgres collation warning — both were ruled out from the host's own logs; the database was repaired with no data loss. The leftover v1 stack has been removed and the host is stable, with no recurrence. To prevent this happening again we are hardening sfp server init/update to enforce the Compose v2 provider on Podman and to remove any leftover v1 stack during migration.
