PR Description

Summary

This PR upgrades minikube’s built-in dynamic storage-provisioner (hostpath provisioner) to support multi-node clusters properly.

The Problem:

Previously, the storage-provisioner ran as a single control-plane Pod. In a multi-node cluster:

1. If a PersistentVolume (PV) host directory was created on Node B, it was physically inaccessible to pods scheduled on Node A.
2. Because the StorageClass immediately provisioned volumes and lacked topology knowledge, pods could be scheduled on Node A, attempt to mount Node B's directory, and crash due to missing data.

The Solution:

1. DaemonSet Deployment: The storage-provisioner now runs as a leader-less DaemonSet on every node in the cluster.
2. Delayed Volume Binding (WaitForFirstConsumer): The standard StorageClass is updated to use volumeBindingMode: WaitForFirstConsumer. This delays volume provisioning until the Kubernetes Scheduler has assigned the consuming Pod to a specific target node.
3. Node Affinity Enforcement: The provisioner running on the target node detects the scheduling choice, provisions the folder locally on that node, and stamps the resulting PV with pv.Spec.NodeAffinity matching the node. This restricts the consuming pod (and all future pods using the PV) to only run on that specific node.
4. Deterministic Identity: Upgraded to use deterministic identities k8s.io/minikube-hostpath-<nodeName> so DaemonSet pod updates or crashes do not create orphaned PVs.

Detailed Changes

Core Provisioner Updates

• go.mod: Upgraded sigs.k8s.io/sig-storage-lib-external-provisioner/v6 to sigs.k8s.io/sig-storage-lib-external-provisioner/v13 (v13.0.0) to ensure compatibility with modern client-go v1.36 APIs.
• pkg/storage/storage_provisioner.go:
  • Initialized nodeName using NODE_NAME environment variable.
  • Generated deterministic, node-specific identities.
  • Implemented intercept checks in Provision() to ignore PVCs targeted at other nodes.
  • Programmed VolumeNodeAffinity generation on PV metadata specs.
  • Disabled leader election on the provisioner controller.

Manifest & Build Configuration Updates

• deploy/addons/storage-provisioner/storage-provisioner.yaml.tmpl: Converted the workload template from Pod to DaemonSet and added Coordination lease RBAC rules.
• deploy/addons/storageclass/storageclass.yaml: Configured volumeBindingMode: WaitForFirstConsumer.
• Makefile / deploy/storage-provisioner/Dockerfile: Bumped tag to v6.0.1 and migrated custom arch build parameter to standard Buildx TARGETARCH.

Test Suite Additions

• test/integration/multinode_test.go: Added validateStorageProvisionerNodeAffinity integration test. It deploys a pod targeted at a worker node (multinode-m02), verifies that the local PV is successfully created, and asserts that the correct NodeAffinity metadata points to multinode-m02.
• test/integration/testdata/: Added node-affinity-pvc.yaml and node-affinity-pod.yaml test manifests.

Verification Results

Tested successfully against a local 2-node Docker driver cluster with the integration test suite:

$ go test ./test/integration -tags="integration" -run="TestMultiNode/serial/FreshStart2Nodes|TestMultiNode/serial/StorageProvisionerNodeAffinity" -binary="../../out/minikube" -profile="multinode" -minikube-start-args="--driver=docker" -cleanup=false -v

Log Output:

=== RUN   TestMultiNode/serial/FreshStart2Nodes
    ...
    multinode_test.go:98: (dbg) Done: out/minikube start -p multinode ... (1m23.14s)
=== RUN   TestMultiNode/serial/StorageProvisionerNodeAffinity
    ...
    multinode_test.go:702: Found correct NodeAffinity: multinode-m02
    multinode_test.go:620: (dbg) Done: out/minikube kubectl -p multinode -- delete pod affinity-pod: (36.74s)
    ...
--- PASS: TestMultiNode (125.36s)
    --- PASS: TestMultiNode/serial (125.31s)
        --- PASS: TestMultiNode/serial/FreshStart2Nodes (84.73s)
        --- PASS: TestMultiNode/serial/StorageProvisionerNodeAffinity (40.58s)
PASS
ok      k8s.io/minikube/test/integration        125.819s