Hi!
I’m using GKE autopilot with a /28 subnet (gets me 12 nodes max).
So, It would be best for me to get the best of the nodes it creates.
What happens is, it’s always trying to create a new node when I change resources or add a new deployment, until it has no more IPs.
The problem: I see that it creates ek-standard-8 nodes, so it has around 8 CPU and 32Gi Mem. Sometimes my deployment only requests 2 CPU and 14Gi Mem, so it leaves a lot of resources to use. But, all nodes have a “balloon” pod that hogs the remaining resources. When I have a new deployment (1 CPU and 4Gi Mem), the scheduler “thinks” the nodes don’t have enough resources and tries to create a new node… why??? There’s clearly enough resources in the existing node.
If I have 12 nodes up, all my new deployments/ scale ups stay pending!
I need help optimizing this.