Production Hardening

Copy Markdown View Source

Use this checklist after the first successful operator session.

Checklist

  • Confirm the host browser pipeline around /ops/jobs matches your real auth boundary. The native shell assumes the same session/auth posture your operators already use elsewhere in the host application.
  • Treat auth_module as host-owned application logic, not a generated placeholder. The library depends on it for current-actor lookup and authorization decisions, but the policy remains your responsibility.
  • Treat actor/session lookup as a host-owned seam. Verify that the same operator identity reaches native pages, LiveView mounts, and any optional bridge request path.
  • Treat display_policy as a production redaction boundary. It should reflect the data your operators may inspect and the fields you need to hide.
  • Verify repo wiring and process supervision in the same environment where operators work so persistence-backed native pages can boot cleanly.
  • Decide whether the optional /ops/jobs/oban bridge belongs in production at all. If you do expose it, keep it aligned with the narrower read-only support posture.
  • Review reverse-proxy and WebSocket behavior before rollout. LiveView transport failures at the edge will make /ops/jobs feel broken even when the library is configured correctly.
  • Review telemetry consumers against the public low-cardinality telemetry contract and avoid coupling downstream dashboards to private payload details.

Telemetry

Powertools telemetry is public API. Keep consumers aligned to the published low-cardinality event families and do not depend on job args, preview tokens, or free-form reasons appearing in telemetry payloads.

Policy seams

The host owns authorization, actor identity, display-policy output, the outer router scope, and the browser pipeline in front of /ops/jobs. Do not ship production defaults until those seams reflect your real operator, redaction, and deployment rules.