Operations

Metrics and alerts

Three processes export Prometheus-compatible metrics: runtime daemon, paired secret kernel, and operation gateway. This page names stable metric families and alert intents; concrete thresholds belong to the operator profile and deployment SLO.

Runtime daemon

Runtime metrics cover process liveness, interface schema pins, kernel reachability, admission persistence, replay behavior, and publication/export gates.

  • Liveness and start time establish process continuity.
  • Interface and state schema versions pin the runtime contract for dashboards and automation.
  • Kernel reachability counters distinguish runtime health from paired-kernel health.
  • Admission counters track submissions, idempotency replays, failures, rate limiting, signed decisions, denials, and decision verification failures.
  • Publication and external-export counters track attempts, successes, and failures at the disclosure boundary.

Secret kernel

Kernel metrics are private-network signals for the paired runtime and operator monitoring plane.

  • Process liveness, start time, request count, response count, error count, and last request time.
  • Outcome counters by kind: done, pending, abort, and none.
  • Backend profile labels let operators separate DKLS23 and FROST behavior in mixed environments.

Recommended alerts

Each alert should name the owning plane, matching the evidence model:

  • Runtime process unavailable — route to node operations.
  • Paired kernel unreachable — route to secret-boundary operations.
  • Admission decision verification failures — route to policy authority and trust-bundle owners.
  • Sustained admission rate limiting — route to gateway intake and client-owner review.
  • Publication gate failures — route to the publication plane owner and inspect denial evidence.
  • Kernel errors or abort outcomes — correlate runtime reason codes with kernel-side outcome counters.