Adding services, containers and applications

Systemd enumeration discovers new service states and restart counters privately. The public portal shows a failed-service count without identities. Per-timer outcomes require an explicit timer/unit allowlist and are not assumed from generic service state.

If a supported container runtime is installed later, add a compatible cAdvisor exporter using a reviewed opt-in configuration. Discovery then follows running containers. Do not install a runtime solely for telemetry or give the public portal access to the Docker socket. Container names, labels and image identities remain private. The initial collector does not claim container support before this integration is installed.

Application request rate, errors, latency, database internals and traces require compatible exporters or instrumentation. Configure OpenTelemetry or a supported Prometheus endpoint explicitly, with bounded labels and authentication as appropriate. Never scrape arbitrary discovered ports. Continuous eBPF profiling and packet capture are excluded.

New public metrics require a catalog entry, finite dimensions, collector aggregation, boundary tests and a reviewed release. Update the wiki’s pinned catalog only after the portal reference is committed. Schema additions alone do not make a hardware measurement supported.