## 2026-09-15T09:21:32Z (a) restart case at 128M: docker oom events, last 15 min
container now: false restarting restarts=11
81:          memory: 1280M
134:          memory: 1280M
## 2026-09-15T09:25:05Z (b) cap restored: 1342177280 starting restarts=10 oom=false
memory hog exec rc=137 (137 = killed)
## 2026-09-15T09:25:10Z (c) child-process OOM inside the RUNNING container:
container: oomkilled=false status=restarting restarts=11
docker oom events since 2026-09-15T09:25:05Z:
## INTERPRETATION (2026-09-15)
- In this unprivileged LXC guest (Docker 29.8.0), NEITHER Docker signal reported the kills: State.OOMKilled stayed false in the
  restart shape (128M cap, 11 restarts) and in the child-process shape (memory hog rc=137, container then restarting);
  `docker events --filter event=oom` returned nothing for either window.
- The controller v0.243.0 check reads State.OOMKilled only. It covers the shape BIGNIGHT measured on VM 333
  (oomkilled=true, restarts=0, container running) and is NOT proven live here. Filed as a row: a reliable OOM signal
  (cgroup memory.events oom_kill read by the agent, or a restart-count trend) is a new mechanism.
- MISTAKE, stated: restoring the cap with `sed s/memory: 128M/memory: 1280M/` also rewrote the redis service's 128M cap
  (compose line 134). Moot: the throwaway stack was removed right after (teardown-9202.txt); the catalog template in git
  was never touched.
