Monitoring
Use three layers of checks.
Process health
Section titled “Process health”Check Beacon /healthz, sidecar health and container/process state.
Overlay health
Section titled “Overlay health”Track peer count, routes, circuit failures, descriptor lookup/publish errors, reconnect loops and version skew.
End-to-end health
Section titled “End-to-end health”Run a synthetic KnotRoute client in another failure domain and request a known .knot service. This covers discovery, routing, directory and rendezvous rather than just one process.
Management health
Section titled “Management health”Agent heartbeat and job status help operate the fleet, but an offline Agent does not necessarily mean the underlying data plane is offline.