Metrics
The operator and the media browser each serve Prometheus metrics on
port 9200, and the catalog pod’s Corrosion agent serves its own on its
own telemetry port, under liken milestone
65
’s
contract. The base applies with no Prometheus in the cluster. An owner
who runs the prometheus-operator adds the deploy/monitoring component
beside the base, which holds a PodMonitor for each of the three.
| Component | Metric | Type | Why |
|---|---|---|---|
| library-operator | liken_build_info{component, version} |
gauge | the release each process runs, always 1 |
| library-operator | library_reconcile_duration_seconds{kind} |
histogram | how long one reconcile pass over one object took |
| library-operator | library_reconcile_errors_total{kind} |
counter | reconcile passes that ended in an error |
| library-operator | library_watch_restarts_total{kind} |
counter | watches the API server closed that the operator reopened |
| library-operator | library_run_duration_seconds{library, worker} |
histogram | a worker whose run got slow |
| library-operator | library_run_last_success_timestamp_seconds{library, worker} |
gauge | a worker that stopped finishing |
| library-operator | library_items{library, kind} |
gauge | catalog size over time |
| library-operator | library_fact_gap{library, fact} |
gauge | a fact whose gap never closes |
| library-operator | library_attempts_total{library, fact, result} |
counter | a fact that stopped finding anything |
| library-operator | library_provider_requests_total{library, provider, status} |
counter | a provider that started refusing |
| library-operator | library_provider_request_seconds_total{library, provider} |
counter | the time one provider costs a run |
| library-operator | library_trailer_fetches_total{library, site, result} |
counter | a site whose downloads started failing |
| library-operator | library_trailer_fetch_bytes_total{library, site} |
counter | what a library’s trailers cost the volume |
| media-browser | library_browser_frame_seconds |
histogram | the covered-spin problem as a graph |
| media-browser | library_browser_art_cache_bytes |
gauge | the RSS problem as a graph |
| catalog | upstream corro_* |
upstream | sync and change health |
The worker label names the runs row a run wrote. scan is a full
walk and rescan a walk of the folders webhooks named, each timed from
the walk’s start to its end. enrich is the whole of a Library’s
Job, from the start its close container writes to the hand-off, so it
covers the walk and every phase. cleanup is the sweep of a deleted
Library. The counters read the tallies every phase container writes,
and they carry no worker.
resources:
- https://github.com/liken-sh/liken//library-operator/deploy?ref=<ref>
components:
- https://github.com/liken-sh/liken//library-operator/deploy/monitoring?ref=<ref>