Narwhal Production Operations

A router pair shares one lease domain for each model fleet. The lease gives one router admission and placement authority while its peer remains on standby.

clients
  |
TLS, authentication, WAF, model routing
  |---------------- router pair A ---------------- fleet A, model A
  |---------------- router pair B ---------------- fleet B, model B
  `---------------- router pair C ---------------- fleet C, model C

Operator tasks

• Production boundary and router pair
• Router state and placement monitoring
• Engine restart and process replacement
• Upgrade, rollback, and release drills

Production startup checklist

For a new or replaced production deployment:

1. Assemble one release, fleet configuration, profile store, and evidence set.
2. Confirm router handoff compatibility with narwhal-check --print-contract-versions.
3. Configure private control interfaces and trusted ingress rewriting.
4. Confirm both router hosts share the required lease domain.
5. Start the intended primary.
6. Start its standby from the same deployment set.
7. Verify lease ownership through /ready.
8. Verify process and fleet state through /health.
9. Run the deployment workload through production ingress.
10. Verify dashboard collection and paging thresholds.
11. Run the engine lifecycle procedure permitted by the configured restart policy.
12. Run router failover through the production load balancer.
13. Open client admission.

Use Troubleshoot a fleet for failure procedures and Measure a fleet for production evidence and deployment validation.