Google Cloud Kubernetes集群缩容至0节点失败：命令执行后节点池无变化

Great question! The short answer is yes, GKE does support scaling a cluster down to 0 nodes—but there are a few common gotchas that might be preventing your command from working as expected. Let’s break this down:

Why your gcloud container clusters resize command didn’t change the node pool size #

The most likely culprit here is a minimum node count restriction on your node pool, especially if it’s the default node pool created with your cluster:

• GKE’s Standard mode sets a default minimum of 1 node for the initial default node pool. Even if you run --size=0, this minimum limit overrides your request and keeps the node pool at 1 node.
• If you have Cluster Autoscaler enabled on the node pool, its configured minNodeCount will also block any attempt to scale below that value.

You can verify this by checking your node pool’s configuration:

gcloud container node-pools describe default-pool --cluster $CLUSTER_NAME --zone $ZONE

Look for the autoscaling.minNodeCount field in the output—if it’s set to 1, that’s the issue.

How to properly scale down to 0 nodes #

For the default node pool #

First, you need to remove the minimum node count restriction, then resize the pool:

# Update the default node pool to allow 0 minimum nodes
gcloud container node-pools update default-pool --cluster $CLUSTER_NAME --zone $ZONE --min-nodes=0

# Now scale the node pool down to 0
gcloud container node-pools resize default-pool --cluster $CLUSTER_NAME --zone $ZONE --size=0

You can also combine these into a single command if you prefer:

gcloud container node-pools resize default-pool --cluster $CLUSTER_NAME --zone $ZONE --size=0 --min-nodes=0

For custom node pools #

If you created a custom node pool (not the default one), you can usually scale it directly to 0 as long as there’s no minimum node count set. Just target the specific node pool in your command:

gcloud container node-pools resize YOUR_CUSTOM_POOL_NAME --cluster $CLUSTER_NAME --zone $ZONE --size=0

Important notes #

• When your cluster is at 0 nodes, all pods (including system pods like kube-dns) will be in a Pending state until you scale the node pool back up.
• For GKE Autopilot clusters, you don’t need to manually manage node counts—Autopilot will automatically scale down to 0 nodes if there are no running pods in the cluster.
• Make sure you don’t have any critical workloads that require running nodes before scaling down to 0!

内容的提问来源于stack exchange，提问作者gkatzioura