Operations and deployment health

Editable Mermaid source · All architecture diagrams.

The shared Operations dashboard at /admin/operations serves current administrator-authorized observations. Portable Golf uses its process/telemetry records; native Golf uses D1 job/delivery/service observations and optional provider collection. Cloudflare needs no Grafana/collector deployment for this page.

Native observations

• Jobs, queues, retries, preview lifetimes, webhook/mail delivery receipts and maintenance stages.
• Optional Artifacts/R2 health probes using existing bindings.
• Optional Workers GraphQL analytics using a separate read-only token, never the app-management token.
• Optional Analytics Engine evidence metrics and configured provider observability.

Bound/configured and successfully collected are different states. The retained demo has Analytics Engine and operations-analytics bindings; hosted ingestion/SQL proof is incomplete and the Workers analytics feed has recorded failures. Provider-native logs/observability are enabled; private data and prompts still need explicit collection/retention policy.

Cron and alarms

Cron Triggers declare the schedule. The retained account has delivered no scheduled events in the recorded window; the cause is unresolved. The Workspace Durable Object arms a five-minute alarm and performs the shared maintenance whenever no scheduled run has reported for ten minutes, standing down after cron resumes. It first arms when the workspace receives a request.

Check the journal, service probes and actual cleanup state. A schedule definition does not prove execution. Purge, evidence/index reconciliation, CI retry and notification retry depend on maintenance.

Limits

Golf does not currently collect complete D1/R2/Artifacts capacity, storage billing, container invoices or estimated spend. Unknown values stay unavailable; gateway limits and demo budgets are not account-wide billing caps.

See the Operations contract and current status.