We have a bluish green implementation system that we are using for quite some time. We have two back-end services in the load balancer. One service is for testing and another is for production. There are 2 groups of different nodes (k8s) used behind these back-end services. To implement a new version in production, we simply change the group of instances in the production fund service. It worked for quite some time until June 2019. After changing the instance group for a short period of time (approximately 2-3 minutes), the back-end service is not available and LB responds with a 502 error.
I've also created a problem in the GCP bug tracker that includes screenshots and steps to play: https://issuetracker.google.com/issues/136020917