# Service Restart Runbook (Draft v0.1)

## Purpose
Provide a repeatable, low-risk procedure for restarting the API service without dropping in-flight requests.

## Scope
Applies to api-svc (primary and canary). Excludes database migrations.

## Prerequisites
- Access to the ops bastion
- PagerDuty duty-handoff confirmed
- Deployment queue is clear (check via `kubectl get events -n deploy`)

## Steps

1. Verify current health: `kubectl get pods -n api -l app=api-svc`
2. Confirm no in-flight requests: `kubectl exec -n api deploy/api-svc -- curl -s localhost:8080/health | jq .in_flight`
3. Drain one pod: `kubectl drain <node> --ignore-daemonsets --delete-emptydir-data`
4. Wait for pod reschedule: `kubectl rollout status deployment/api-svc -n api --timeout=120s`
5. Repeat for remaining pods, one at a time
6. Run smoke test: `curl -s https://api.internal/healthz | jq .status`
7. Page on-call if smoke test fails

## Rollback
If health check fails after 2 pods, halt and execute: `kubectl rollout undo deployment/api-svc -n api`

## Evidence Required
- Screenshot of `kubectl get pods` post-restart
- Smoke test output showing `"status":"ok"`

---
*Draft prepared by Quill — verify commands against staging before production use.*