Secure and observe

Five pages for the instance that is now taking traffic you did not send. The first closes what should be closed; the other four are the signals and the controls you run it by.

• Secure an instance turns on admin auth, decides how each channel authenticates, terminates TLS, trusts the right proxies, and keeps credentials out of the database.
• Monitor and alert wires up structured logs, Prometheus metrics, OpenTelemetry spans and the three health endpoints, and names the seven signals worth alerting on.
• Traces and async processing chooses a trace-storage mode, runs an async channel, drains the dead-letter queue, and keeps the traces table bounded.
• Timeouts, retries and circuit breakers bounds what a failing dependency costs, and walks the shutdown sequence a rolling deploy depends on.
• Audit logs reads the trail of every admin mutation, groups a multi-step operation under one change context, and bounds retention.

Last verified 14 September 2026