immich: 0 postgres backends killed by signal 9; 3 server worker exits
immich-retry: 0 postgres backends killed by signal 9; 3 server worker exits
NOTE: the saved compose logs above do NOT carry the postgres kill lines (0 matches — the 400 kB tail is the server's output).
The kill counts below come from a LIVE read, transcribed here by the session (not re-readable: the bench is torn down).
Read live on the bench during the retry (2026-09-30 14:35 CEST), immich-postgres cgroup memory.events: oom 413, oom_kill 36;
HostConfig.Memory 536870912 (512 MiB); OOMKilled true. Image = the ladder's tested digest (sha256:1a078b23…), PostgreSQL 16.10.
The server's first start (reverse-geocoding import) dropped its DB connection ("terminating connection because of crash of another
server process") and never became healthy in the harness's 420 s. The same template on 9202 (14:31, fresh install, kept data moved
aside) seeded in 35 s and its update ended done; its replaced container read oom_kill 0 — the FIRST container's counter is gone.
