Optional checks introduced in 0.0.68, also visible on the live demo.
1. Enable the measurements
monit-docker --event-window 300 stats --rsc oom_events --rsc starts_recent --rsc pids_current --rsc pids_limit --rsc pids_percentThese checks are opt-in. Explicit --rsc options replace the default selection: add CPU and memory resources if you need them too. The event window is a global option, before stats or serve. Rules referencing these resources also enable their collection.
2. Interpret the results
oom_events counts Docker OOM events in the window. starts_recent counts starts, including the initial start, manual restarts and Docker restart policies. PID usage includes both processes and threads; its percentage uses the configured container PID limit, not an ancestor cgroup limit. An unlimited or unavailable limit has no percentage.
3. Recognize incomplete history
Docker retains at most 256 events. When that buffer does not cover the requested window, counts are unavailable and the interface displays “Event history incomplete”. History is not durable across Docker daemon restarts, even when no truncation is detected. These are window gauges, not lifetime counters; do not apply rate() to them.
4. Add optional alerts
The source example examples/monitoring/alerts.runtime.yml covers OOM, frequent starts, PID pressure and incomplete history. Mount it in Prometheus and add it to rule_files. Match its 300-second window to your agent configuration. The public demo displays measurements and keeps manual actions in its journal; it does not deliberately exhaust memory to demonstrate OOM.