Gameserver's that are not assigned to a node are left behind even after the scale was lowered

[ilkercelikyilmaz]

Autosclaing on the Kubernetes cluster was turned off.

• kubectl create -f ./fleet.yaml

• kubectl get fleet
  NAME SCHEDULING DESIRED CURRENT ALLOCATED READY AGE
  fleet-example Packed 2 2 0 2 15s

• kubectl get pods
  NAME READY STATUS RESTARTS AGE
  fleet-example-7x6k4-bn49h-wqbgl 2/2 Running 0 19s
  fleet-example-7x6k4-vkpgv-98nc9 2/2 Running 0 19s

• kubectl get po -o=custom-columns=NODE:.spec.nodeName,NAME:.metadata.name | cut -f 1 -d " " | sort | uniq -c | sort
  1 NODE
  2 gke-my-cluster-1-default-c8dd396a-4fsl

• kubectl edit fleet fleet-example (Set the replicas to 600)

• Gameservers created but 493 were not assigned to a node
  -kubectl get po -o=custom-columns=NODE:.spec.nodeName,NAME:.metadata.name | cut -f 1 -d " " | sort | uniq -c | sort
  1 NODE
  26 gke-my-cluster-1-default-c8dd396a-6s57
  26 gke-my-cluster-1-default-c8dd396a-fgbk
  27 gke-my-cluster-1-default-c8dd396a-5zvq
  28 gke-my-cluster-1-default-c8dd396a-4fsl
  493

• then changed the replica size to 50 but the number of gameservers didn't go down:

• kubectl get fleet
  NAME SCHEDULING DESIRED CURRENT ALLOCATED READY AGE
  fleet-example Packed 50 600 0 107 3h

[markmandel]

This feels very similar to #540 - or maybe even duplicate.

[ilkercelikyilmaz]

I didn't see that issue before but you are right, they are duplicates.

[ilkercelikyilmaz]

I was talking to Jarek and he thinks that both problems might have surfaced from same activity/situation, each issue captures different problem. We probably should keep this open.

[ilkercelikyilmaz]

Recent performance improvements and bug fixes seems to fix this issue. I can't repro anymore.