Professional Cloud Developer

Rollback Execution and Health-Based Traffic Steering

Traffic splitting is a method of sending a specific percentage of incoming requests to different versions of a service. This technique lets developers test new features safely by showing them to only a small group of users first. By controlling how traffic flows, teams can catch problems before they affect everyone.

There are two main ways to split traffic. IP address splitting uses a mathematical formula on the user's internet address to decide which version they see. This method is easier to set up but can be inconsistent for mobile users whose addresses change as they move. Cookie splitting is more accurate because it uses a special identifier called GOOGAPPUID to keep a user on the same version throughout their entire visit.

When rolling out a new version, developers should use a gradual approach called the 500/50/5 pattern. This means starting with no more than 500 requests per second and increasing traffic by 50% every five minutes. This slow approach prevents the backend from becoming overwhelmed and gives the system time to scale up naturally.

Cloud Monitoring tracks how well the service is performing. If the new version shows errors or slow response times, developers can quickly shift all traffic back to a previous version that worked correctly. This is called a rollback, and it protects users from experiencing problems.

Caching can cause issues when traffic is split between versions. If a browser stores an old version of a file, it might not work with the new version. Developers use Cache-Control headers or change file names between versions to prevent this mismatch.

Cloud Run Revision and Traffic Tag Management

Cloud Run manages different versions of an application through revisions. Each time you deploy a new version, Cloud Run creates a new revision that can receive a portion of incoming traffic. This lets you test changes with real users before committing to them fully.

Traffic tags are labels you can assign to specific revisions. These tags let you route traffic to a particular version without using traffic percentages. Tags are useful for testing because you can send traffic to a specific revision using its tag name instead of calculating percentages.

When managing revisions, you should follow the same gradual rollout pattern used elsewhere. Start with a very small percentage of traffic, then increase it slowly while watching for problems. If something goes wrong, you can immediately send all traffic back to the previous revision.

Rate controls help prevent target overload, which happens when too many requests reach your service at once. By setting limits on how fast requests arrive, you give your service time to scale up and handle the load. Cloud Monitoring watches for these spikes and alerts you before they cause failures.

GKE Deployment Patterns and Service Mesh Integration

On GKE, developers use deployment patterns to release new versions safely. A blue-green deployment runs two identical environments at the same time. One runs the current version (blue) while the other runs the new version (green). When ready, all traffic switches from blue to green in one step. A canary deployment sends only a small amount of traffic to the new version first, then gradually increases it based on how well the new version performs.

Kubernetes Services and Ingress controllers manage how traffic reaches these different versions. The load balancer can be configured to send a percentage of requests to each version, allowing for controlled testing. This approach reduces risk because only a small group of users encounters any problems.

Cloud Service Mesh adds another layer of control by managing how services communicate with each other. It can route traffic based on headers, paths, or other request attributes without changing the application code. This makes it possible to implement complex routing rules for A/B testing or gradual rollouts.

Before switching all traffic to a new version, thorough testing ensures the release is stable. If the new version fails checks, an immediate rollback restores the previous stable version. Google Cloud Observability provides the metrics and logs needed to make these decisions quickly and confidently.