Issue

• Both the Sandbox and SandboxClaim CRDs remain "Ready" indefinitely after a pod is deleted when a node is drained manually (kubectl drain command on the node) even when the pod is rescheduled.
• The bug is only observed when using a WarmPool, not when controller directly creates a new pod.

Steps to reproduce

1. Create a SandboxTemplate and SandboxWarmPool with replicas: 1
2. Create a SandboxClaim, make sure it adopts the pod from WarmPool
3. Get the adopted pod & node names

ADOPTED_POD_NAME=$(kubectl get sandbox <warmpool> -n default -o jsonpath='{.metadata.annotations.agents\.x-k8s\.io/pod-name}')
NODE_TO_DRAIN=$(kubectl get pod $ADOPTED_POD_NAME -n default -o jsonpath='{.spec.nodeName}')

5. Drain the node kubectl drain $NODE_TO_DRAIN --ignore-daemonsets --delete-emptydir-data --force

Environment

• GKE Autopilot cluster
• agent-sandbox controller: installed via kubectl apply from the repo, v0.1.0