Question

A retail company deploys a customer-facing API to Cloud Run using an automated CI/CD pipeline. During recent releases, several bad revisions reached 100% of traffic before the on-call team noticed elevated 5xx errors, causing customer impact. The architect wants to reduce the blast radius of faulty releases without requiring an engineer to manually watch dashboards after every deployment. Which approach best achieves this?

Accepted Answer

Deploy each new revision with no traffic, then use a pipeline step that shifts a small percentage of traffic and evaluates error-rate and latency SLO metrics before automatically promoting or aborting the rollout.

Answer

Increase the Cloud Run maximum instance count so the service can absorb more requests and mask transient errors during deployment.

Answer

Configure a Cloud Monitoring alert policy that pages the on-call engineer whenever the 5xx rate exceeds a threshold after full deployment.

Answer

Enable Cloud Run request retries so failed requests are automatically retried against the new revision until they succeed.