Skip to content

Metrics

This page lists Casebox’s own metrics. The server serves Prometheus metrics at /metrics on port 8081, and sends metrics and traces over OTLP when OTEL_EXPORTER_OTLP_ENDPOINT is set. No metric carries a person, a repository or a workspace name.

Metric Kind Tags Shows
casebox.ingest.lag histogram, s Receipt time minus the newest event’s time, per uploaded batch
casebox.spool.backlog gauge Events the CLIs reported as not uploaded, over the last hour
casebox.jobs.queued gauge kind Queued jobs
casebox.jobs.oldest gauge, s kind How long the oldest queued job has waited
casebox.jobs.lease_expiries counter kind Leases that expired: a worker stopped before it answered
casebox.workers.seen gauge Workers seen in the last 10 minutes
casebox.runs counter outcome Evaluation runs completed, or failed to run
casebox.sandbox.start histogram, s provider Time to prepare and start an agent’s sandbox
casebox.spend.month gauge, USD org Evaluation spend this month
casebox.budget.month gauge, USD org The monthly limit
casebox.steering.unclassified gauge org Share of the last 30 days’ interventions left unclassified

The server also publishes Deedbox’s metrics: deedbox.consumer.lag.seconds shows how far each projection and workflow is behind.

With OTEL_EXPORTER_OTLP_ENDPOINT set, casebox worker sends a span per job, casebox.worker.job.duration (kind, outcome) and casebox.worker.sandbox.start (provider).

QueueBox serves its own metrics at /metrics on port 9090, among them queuebox_outbox_messages_total by status: dead counts side effects it gave up on. See the runbooks.