Observability

Today: Prometheus metrics on cellpd, structured logs on stdout, and per-version celld process output. Planned (not shipped): OTLP trace/log export and a version-scoped query API behind the same admin token — backends would be pluggable (memory, Jaeger, or your own LGTM stack).

cellp will not ship a SaaS analytics product or an in-tree log search engine.

Metrics

Scrape cellpd at the root path (not under /v1):

yaml

scrape_configs:
  - job_name: cellpd
    static_configs:
      - targets: ["cellpd:8790"]
    metrics_path: /metrics

You will see HTTP counters/histograms, orchestrator queue depth, and gateway upstream health. Metric names are defined in the cellp Go module (internal/metrics).

bash

curl -s http://127.0.0.1:8790/metrics | head

Logs (today)

Component	Where
cellpd	Docker / systemd / stdout — API, saga, archive reaper
celld	Per-version process — `console.log` and binding CLI
RustFS	Its own metrics/logs

Archive reaper lines contain orch: archive reaper.

There is no wrangler tail equivalent. Use process logs and metrics until OTLP live tail ships.

Health

bash

curl -sf http://127.0.0.1:8790/v1/health
curl -sf http://127.0.0.1:8790/v1/health/deep

Deep health is the right probe for “can I deploy?” — registry, object store, runtimes, queue. Returns 503 when the deploy queue exceeds CELLP_QUEUE_MAX.

GET /v1/runtime/routes (admin) summarizes upstreams.

Planned observability (OTLP, not shipped)

Layer	Contract (design)
Emit	OTLP traces + logs; `cellp.project` / `cellp.version`; Gateway `traceparent`
Query	cellpd facade (`context`, `traces/{id}`, template `search`) — `ADMIN_TOKEN` only
Live	Process stream (SSE), not OTEL
Backend	Operator-chosen collector (Jaeger, Grafana stack, etc.)

Bring your own Grafana for boards. The Dashboard talks only to :8790.

Promote windows

Watch gateway 5xx around promote. Rollback path: Rollback.

Observability ​

Metrics ​

Logs (today) ​

Health ​

Planned observability (OTLP, not shipped) ​