Added
--min-container-age / MIN_CONTAINER_AGE (default 30s): suppresses metric emission for containers younger than the configured threshold, reducing high-cardinality series from short-lived job/cronjob pods. Ported from upstream coroot-node-agent; the gate falls back to the cgroup directory mtime when taskstats hasn't populated startedAt, so containers whose initial taskstats lookup failed (PID-vs-TGID races, restricted netlink caps, missing CONFIG_TASKSTATS_NETLINK) still emit metrics instead of being permanently filtered.
--instrumentation-delay / INSTRUMENTATION_DELAY (default 30s): delays attaching language-runtime probes (Python GIL, Node.js event loop, etc.) to newly started processes, avoiding probe churn on processes that exit during startup. Ported from upstream coroot-node-agent.
Upgrade notes
The 30s defaults change behavior for existing deployments: containers younger than 30s will be suppressed from metric output, and language-runtime probes attach 30s after process start. Set either flag (or env var) to 0 to restore pre-1.2.1 behavior.