IssueTracker

Green/Blue deployments which use a standard Load Balancer, are able to be implemented without downtime [1]. Nevertheless, when this strategy is performed through deployments using an Ingress resource instead of the Load Balancer pointed above, service can be impacted by downtime.

The described downtime can last up to 20 seconds, and has been related on previous threats, to the node pool switch step [2]. Having said that, we would like to point, that another fact that triggers a downtime period is the change on the selector field from the service.

On the public Internet, diferent sources point to this issue, for example the following Github threat [3], and to avoid such behavior, usage of applications like HA Proxy are recommended [4].

It would be really helpful to support such deployment, using only GKE resources instead of appealing to 3rd party tools, as the available workarounds increase the complexity of an already not trivial deployment.

[1] https://cloud.google.com/solutions/implementing-deployment-and-testing-strategies-on-gke#perform_a_bluegreen_deployment

[2] https://issuetracker.google.com/136020917

[3] https://github.com/kubernetes/ingress-gce/issues/638#issuecomment-646499539

[4] https://www.haproxy.com/blog/rolling-updates-and-blue-green-deployments-with-kubernetes-and-haproxy/#blue-green-deployments