How do I implement a Blue-Green deployment strategy in K8s?

All answers to this question.

The simplest way to do Blue-Green in vanilla Kubernetes is by using Labels and Selectors within your Service object. You maintain two separate Deployments: one labeled 'version: blue' and one 'version: green'. Your Service initially has a selector pointing to 'blue'. When you’re ready to switch, you update the Service's selector to 'green' using kubectl patch. If things go wrong, you simply patch it back to 'blue'. For more advanced traffic splitting, you might want to look into an Ingress Controller like NGINX or a Service Mesh.

   Answered 2024-01-24 by Helen Robinson

Have you considered using a tool like Argo CD or Flux for this? Manual patching works, but it’s prone to human error. GitOps tools can automate the entire transition and even handle the health checks for you.

   Answered 2024-01-25 by Steven Hall

• Steven, I actually looked into Argo Rollouts after reading your comment. It seems much more robust than my manual patching idea! It supports "AnalysisTemplates" which can automatically roll back the traffic to the 'Blue' environment if Prometheus metrics show an increase in 5xx errors during the 'Green' phase. This is exactly the kind of automated safety net our DevOps team has been looking for.

     Commented 2024-01-26 by Elizabeth Taylor

Blue-Green is great, but remember it requires double the resources since you run two versions of the app at once. Make sure your cluster has enough headroom!

   Answered 2024-01-27 by Nancy Young

• Good point, Nancy. We always check our node capacity before starting a Blue-Green shift to avoid scheduling failures during the peak traffic hours.

     Commented 2024-01-28 by Helen Robinson