GKE upgrade -- how does it handle pod movement?

Google Cloud Serverless Applications
1

So basically, let’s say we have a cluster, and in that cluster a node pool with 3 nodes. Let’s say we have a deployment that has 1 replica, so 1/1.

What would happen to that deployment in detail during an GKE upgrade? From what I know, GKE picks a node to upgrade, so it condones it, then creates a new node with newer version, then after it’s done, it moves over the pods over to the new node by destroying/re-creating correct?

My question though is that the service that the pod provides (whatever that may be) will be disrupted correct? Due to there only being 1 replica, OR, does GKE force create a 2nd replica on the new node, so the deployment becomes 2/1, and then after that is done it deletes the pod in the old node putting the deployment back to 1/1 without any disruption?

I’m wondering whether I should force all deployments with 1/1 replicas to 2/2 during the upgrade via kubectl, and then put them back to 1/1 replicas after the upgrade has finished, or whether GKE kinda does this under the hood?

2

Ah ok, for future reference, my above thinking was true, you have to have at least 2 replicas to prevent the service from becoming unavailable, potentially more if it’s more critical, just to make sure.

Discruption budgets don’t seem to be necessary.