fix: find node by name

We will find the node by name more precisely.
Check the UUID and VM name to determine the VM ID.

Signed-off-by: Serge Logvinov <serge.logvinov@sinextra.dev>

.github/PULL_REQUEST_TEMPLATE.md

@@ -4,7 +4,7 @@
 ## Note to the Contributor
 
 We encourage contributors to go through a proposal process to discuss major changes.
-Before your PR is allowed to run through CI, the maintainers of Talos CCM will first have to approve the PR.
+Before your PR is allowed to run through CI, the maintainers of Proxmox CCM will first have to approve the PR.
 -->
 
 ## What? (description)

pkg/cluster/client.go

@@ -29,6 +29,7 @@ import (
 
 	pxapi "github.com/Telmate/proxmox-api-go/proxmox"
 
+	v1 "k8s.io/api/core/v1"
 	"k8s.io/klog/v2"
 )
 
@@ -106,6 +107,33 @@ func (c *Cluster) GetProxmoxCluster(region string) (*pxapi.Client, error) {
 	return nil, fmt.Errorf("proxmox cluster %s not found", region)
 }
 
+// FindVMByNode find a VM by kubernetes node resource in all Proxmox clusters.
+func (c *Cluster) FindVMByNode(ctx context.Context, node *v1.Node) (*pxapi.VmRef, string, error) {
+	for region, px := range c.proxmox {
+		vmrs, err := px.GetVmRefsByName(ctx, node.Name)
+		if err != nil {
+			if strings.Contains(err.Error(), "not found") {
+				continue
+			}
+
+			return nil, "", err
+		}
+
+		for _, vmr := range vmrs {
+			config, err := px.GetVmConfig(ctx, vmr)
+			if err != nil {
+				return nil, "", err
+			}
+
+			if c.GetVMUUID(config) == node.Status.NodeInfo.SystemUUID {
+				return vmr, region, nil
+			}
+		}
+	}
+
+	return nil, "", fmt.Errorf("vm '%s' not found", node.Name)
+}
+
 // FindVMByName find a VM by name in all Proxmox clusters.
 func (c *Cluster) FindVMByName(ctx context.Context, name string) (*pxapi.VmRef, string, error) {
 	for region, px := range c.proxmox {
@@ -162,6 +190,24 @@ func (c *Cluster) FindVMByUUID(ctx context.Context, uuid string) (*pxapi.VmRef,
 	return nil, "", fmt.Errorf("vm with uuid '%s' not found", uuid)
 }
 
+// GetVMName returns the VM name.
+func (c *Cluster) GetVMName(vmInfo map[string]interface{}) string {
+	if vmInfo["name"] != nil {
+		return vmInfo["name"].(string) //nolint:errcheck
+	}
+
+	return ""
+}
+
+// GetVMUUID returns the VM UUID.
+func (c *Cluster) GetVMUUID(vmInfo map[string]interface{}) string {
+	if vmInfo["smbios1"] != nil {
+		return c.getUUID(vmInfo["smbios1"].(string)) //nolint:errcheck
+	}
+
+	return ""
+}
+
 func (c *Cluster) getUUID(smbios string) string {
 	for _, l := range strings.Split(smbios, ",") {
 		if l == "" || l == "base64=1" {

pkg/proxmox/instances.go

@@ -142,7 +142,7 @@ func (i *instances) InstanceMetadata(ctx context.Context, node *v1.Node) (*cloud
 
 			mc := metrics.NewMetricContext("findVmByName")
 
-			vmRef, region, err = i.c.FindVMByName(ctx, node.Name)
+			vmRef, region, err = i.c.FindVMByNode(ctx, node)
 			if mc.ObserveRequest(err) != nil {
 				mc := metrics.NewMetricContext("findVmByUUID")
 
@@ -234,8 +234,10 @@ func (i *instances) getInstance(ctx context.Context, node *v1.Node) (*pxapi.VmRe
 		return nil, "", err
 	}
 
-	if vmInfo["name"] != nil && vmInfo["name"].(string) != node.Name { //nolint:errcheck
+	if i.c.GetVMName(vmInfo) != node.Name && i.c.GetVMUUID(vmInfo) != node.Status.NodeInfo.SystemUUID {
-		return nil, "", fmt.Errorf("instances.getInstance() vm.name(%s) != node.name(%s)", vmInfo["name"].(string), node.Name) //nolint:errcheck
+		klog.Errorf("instances.getInstance() vm.name(%s) != node.name(%s) with uuid=%s", i.c.GetVMName(vmInfo), node.Name, node.Status.NodeInfo.SystemUUID)
+
+		return nil, "", cloudprovider.InstanceNotFound
 	}
 
 	klog.V(5).Infof("instances.getInstance() vmInfo %+v", vmInfo)

pkg/proxmox/instances_test.go

@@ -66,20 +66,22 @@ clusters:
 			return httpmock.NewJsonResponse(200, map[string]interface{}{
 				"data": []interface{}{
 					map[string]interface{}{
-						"node":   "pve-1",
+						"node":    "pve-1",
-						"type":   "qemu",
+						"type":    "qemu",
-						"vmid":   100,
+						"vmid":    100,
-						"name":   "cluster-1-node-1",
+						"name":    "cluster-1-node-1",
-						"maxcpu": 4,
+						"maxcpu":  4,
-						"maxmem": 10 * 1024 * 1024 * 1024,
+						"maxmem":  10 * 1024 * 1024 * 1024,
+						"smbios1": "uuid=8af7110d-bfad-407a-a663-9527d10a6583",
 					},
 					map[string]interface{}{
-						"node":   "pve-2",
+						"node":    "pve-2",
-						"type":   "qemu",
+						"type":    "qemu",
-						"vmid":   101,
+						"vmid":    101,
-						"name":   "cluster-1-node-2",
+						"name":    "cluster-1-node-2",
-						"maxcpu": 2,
+						"maxcpu":  2,
-						"maxmem": 5 * 1024 * 1024 * 1024,
+						"maxmem":  5 * 1024 * 1024 * 1024,
+						"smbios1": "uuid=5d04cb23-ea78-40a3-af2e-dd54798dc887",
 					},
 				},
 			})
@@ -91,18 +93,58 @@ clusters:
 			return httpmock.NewJsonResponse(200, map[string]interface{}{
 				"data": []interface{}{
 					map[string]interface{}{
-						"node":   "pve-3",
+						"node":    "pve-3",
-						"type":   "qemu",
+						"type":    "qemu",
-						"vmid":   100,
+						"vmid":    100,
-						"name":   "cluster-2-node-1",
+						"name":    "cluster-2-node-1",
-						"maxcpu": 1,
+						"maxcpu":  1,
-						"maxmem": 2 * 1024 * 1024 * 1024,
+						"maxmem":  2 * 1024 * 1024 * 1024,
+						"smbios1": "uuid=3d3db687-89dd-473e-8463-6599f25b36a8",
 					},
 				},
 			})
 		},
 	)
 
+	httpmock.RegisterResponder("GET", "https://127.0.0.1:8006/api2/json/nodes/pve-1/qemu/100/config",
+		func(_ *http.Request) (*http.Response, error) {
+			return httpmock.NewJsonResponse(200, map[string]interface{}{
+				"data": map[string]interface{}{
+					"node":    "pve-1",
+					"type":    "qemu",
+					"vmid":    100,
+					"smbios1": "uuid=8af7110d-bfad-407a-a663-9527d10a6583",
+				},
+			})
+		},
+	)
+
+	httpmock.RegisterResponder("GET", "https://127.0.0.1:8006/api2/json/nodes/pve-2/qemu/101/config",
+		func(_ *http.Request) (*http.Response, error) {
+			return httpmock.NewJsonResponse(200, map[string]interface{}{
+				"data": map[string]interface{}{
+					"node":    "pve-2",
+					"type":    "qemu",
+					"vmid":    101,
+					"smbios1": "uuid=5d04cb23-ea78-40a3-af2e-dd54798dc887",
+				},
+			})
+		},
+	)
+
+	httpmock.RegisterResponder("GET", "https://127.0.0.2:8006/api2/json/nodes/pve-3/qemu/100/config",
+		func(_ *http.Request) (*http.Response, error) {
+			return httpmock.NewJsonResponse(200, map[string]interface{}{
+				"data": map[string]interface{}{
+					"node":    "pve-3",
+					"type":    "qemu",
+					"vmid":    100,
+					"smbios1": "uuid=3d3db687-89dd-473e-8463-6599f25b36a8",
+				},
+			})
+		},
+	)
+
 	httpmock.RegisterResponder("GET", "https://127.0.0.1:8006/api2/json/nodes/pve-1/qemu/100/status/current",
 		func(_ *http.Request) (*http.Response, error) {
 			return httpmock.NewJsonResponse(200, map[string]interface{}{
@@ -207,9 +249,30 @@ func (ts *ccmTestSuite) TestInstanceExists() {
 				Spec: v1.NodeSpec{
 					ProviderID: "proxmox://cluster-1/100",
 				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-bfad-407a-a663-9527d10a6583",
+					},
+				},
 			},
-			expectedError: "vm.name(cluster-1-node-1) != node.name(cluster-1-node-3)",
+			expected: true,
-			expected:      false,
+		},
+		{
+			msg: "NodeExistsWithDifferentNameAndUUID",
+			node: &v1.Node{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: "cluster-1-node-3",
+				},
+				Spec: v1.NodeSpec{
+					ProviderID: "proxmox://cluster-1/100",
+				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-0000-0000-0000-9527d10a6583",
+					},
+				},
+			},
+			expected: false,
 		},
 	}
 
@@ -288,6 +351,11 @@ func (ts *ccmTestSuite) TestInstanceShutdown() {
 				Spec: v1.NodeSpec{
 					ProviderID: "proxmox://cluster-1/100",
 				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-bfad-407a-a663-9527d10a6583",
+					},
+				},
 			},
 			expected: false,
 		},
@@ -303,6 +371,40 @@ func (ts *ccmTestSuite) TestInstanceShutdown() {
 			},
 			expected: true,
 		},
+		{
+			msg: "NodeExistsWithDifferentName",
+			node: &v1.Node{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: "cluster-1-node-3",
+				},
+				Spec: v1.NodeSpec{
+					ProviderID: "proxmox://cluster-1/100",
+				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-bfad-407a-a663-9527d10a6583",
+					},
+				},
+			},
+			expected: false,
+		},
+		{
+			msg: "NodeExistsWithDifferentUUID",
+			node: &v1.Node{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: "cluster-1-node-1",
+				},
+				Spec: v1.NodeSpec{
+					ProviderID: "proxmox://cluster-1/100",
+				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-0000-0000-0000-9527d10a6583",
+					},
+				},
+			},
+			expected: false,
+		},
 	}
 
 	for _, testCase := range tests {
@@ -398,6 +500,11 @@ func (ts *ccmTestSuite) TestInstanceMetadata() {
 						cloudproviderapi.AnnotationAlphaProvidedIPAddr: "1.2.3.4",
 					},
 				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-bfad-407a-a663-9527d10a6583",
+					},
+				},
 			},
 			expected: &cloudprovider.InstanceMetadata{
 				ProviderID: "proxmox://cluster-1/100",
@@ -425,6 +532,11 @@ func (ts *ccmTestSuite) TestInstanceMetadata() {
 						cloudproviderapi.AnnotationAlphaProvidedIPAddr: "1.2.3.4,2001::1",
 					},
 				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "8af7110d-bfad-407a-a663-9527d10a6583",
+					},
+				},
 			},
 			expected: &cloudprovider.InstanceMetadata{
 				ProviderID: "proxmox://cluster-1/100",
@@ -456,6 +568,11 @@ func (ts *ccmTestSuite) TestInstanceMetadata() {
 						cloudproviderapi.AnnotationAlphaProvidedIPAddr: "1.2.3.4",
 					},
 				},
+				Status: v1.NodeStatus{
+					NodeInfo: v1.NodeSystemInfo{
+						SystemUUID: "3d3db687-89dd-473e-8463-6599f25b36a8",
+					},
+				},
 			},
 			expected: &cloudprovider.InstanceMetadata{
 				ProviderID: "proxmox://cluster-2/100",