majiayu000 Read ../_house-style/house-style.md before starting.
Anchor phrases
- Wrong IAM is production breakage with paperwork.
- If the secret path is wrong, your health checks are fiction.
- A rollout without rollback is just optimism with YAML.
- Autoscaling hides mistakes until traffic arrives.
What to interrogate
1. Release surface
State what is shipping:
- Cloud Run
- GKE
- App Engine
- Cloud Functions
- load balancer or IAM change
- secrets, networking, or env config
2. Runtime and identity
Check:
- service account assumptions
- IAM bindings
- secret access
- environment parity
- network reachability and ingress expectations
3. Scale and startup risks
Review:
- cold start sensitivity
- request timeout assumptions
- concurrency and memory settings
- health probe realism
- autoscaling behavior under burst load
4. Rollout and rollback
Ask:
- is this canary, gradual, or instant?
- what happens if new revision is bad?
- can traffic shift back cleanly?
- do data or config changes make rollback fake?
Output format
GCP release surface
What actually changes in the platform.
Platform blockers
IAM, secret, network, startup, or scaling blockers.
Rollout risk
What fails during deploy, scale-up, or rollback.
Mitigation and rollback plan
Concrete safer rollout steps.
Verdict
- Safe to release
- Fix before release
- GCP release red flag