draft lvm ci

Signed-off-by: Vicente Cheng <[email protected]>

ci/scripts/get-debug-info.sh

@@ -11,4 +11,8 @@ kubectl logs $NDMPOD -n harvester-system |grep -v Skip
 
 # get blockdevices info
 echo "========== Dump blockdevices =========="
-kubectl get blockdevice -n longhorn-system -o yaml
+kubectl get blockdevice -n longhorn-system -o yaml
+
+# get lvmvgs info
+echo "========== Dump LVMVolumeGroups =========="
+kubectl get lvmvg -n harvester-system -o yaml

pkg/utils/virt.go

@@ -1,9 +1,12 @@
 package utils
 
 import (
+	"crypto/sha512"
+	"encoding/hex"
 	"encoding/xml"
 	"fmt"
 	"os"
+	"time"
 )
 
 type Disk struct {
@@ -67,3 +70,11 @@ func XMLWriter(targetFilePath string, xmlData any) error {
 
 	return nil
 }
+
+func GenHash() string {
+	timestamp := time.Now().Unix()
+	timestampStr := fmt.Sprintf("%d", timestamp)
+	hash := sha512.Sum512([]byte(timestampStr))
+	hashStr := hex.EncodeToString(hash[:])
+	return hashStr[:6]
+}

tests/integration/test_1_disk_hotplug_test.go

@@ -233,7 +233,7 @@ func (s *HotPlugTestSuite) Test_4_RemoveInactiveDisk() {
 	curBlockdevice, err = bdi.Get(context.TODO(), s.targetDiskName, v1.GetOptions{})
 	require.Equal(s.T(), err, nil, "Get BlockdevicesList should not get error before we want to check remove")
 	require.Equal(s.T(), curBlockdevice.Status.DeviceStatus.FileSystem.MountPoint, "", "Mountpoint should be empty after we remove disk!")
-	require.Equal(s.T(), diskv1.ProvisionPhaseUnprovisioned, curBlockdevice.Status.ProvisionPhase, "Block device provisionPhase should be Provisioned")
+	require.Equal(s.T(), diskv1.ProvisionPhaseUnprovisioned, curBlockdevice.Status.ProvisionPhase, "Block device provisionPhase should be Unprovisioned")
 }
 
 func doCommand(cmdString string) (string, string, error) {

tests/integration/test_2_lvm_test.go

@@ -0,0 +1,325 @@
+package integration
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/kevinburke/ssh_config"
+	"github.com/melbahja/goph"
+	"github.com/stretchr/testify/require"
+	"github.com/stretchr/testify/suite"
+	corev1 "k8s.io/api/core/v1"
+	v1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/tools/clientcmd"
+
+	diskv1 "github.com/harvester/node-disk-manager/pkg/apis/harvesterhci.io/v1beta1"
+	clientset "github.com/harvester/node-disk-manager/pkg/generated/clientset/versioned"
+	"github.com/harvester/node-disk-manager/pkg/utils"
+)
+
+type LVMSuite struct {
+	suite.Suite
+	SSHClient             *goph.Client
+	coreClientSet         *kubernetes.Clientset
+	clientSet             *clientset.Clientset
+	targetNodeName        string
+	hotplugTargetNodeName string
+	hotplugTargetBaseDir  string
+	targetDevFirst        *diskv1.BlockDevice
+	targetDevSecond       *diskv1.BlockDevice
+}
+
+func (s *LVMSuite) SetupSuite() {
+	nodeName := ""
+	f, err := os.Open(filepath.Join(os.Getenv("NDM_HOME"), "ssh-config"))
+	require.Equal(s.T(), err, nil, "Open ssh-config should not get error")
+	cfg, err := ssh_config.Decode(f)
+	require.Equal(s.T(), err, nil, "Decode ssh-config should not get error")
+	// consider wildcard, so length shoule be 2
+	require.Equal(s.T(), len(cfg.Hosts), 2, "number of Hosts on SSH-config should be 1")
+	for _, host := range cfg.Hosts {
+		if host.String() == "" {
+			// wildcard, continue
+			continue
+		}
+		nodeName = host.Patterns[0].String()
+		break
+	}
+	require.NotEqual(s.T(), nodeName, "", "nodeName should not be empty.")
+	s.targetNodeName = nodeName
+	targetHost, _ := cfg.Get(nodeName, "HostName")
+	targetUser, _ := cfg.Get(nodeName, "User")
+	targetPrivateKey, _ := cfg.Get(nodeName, "IdentityFile")
+	splitedResult := strings.Split(targetPrivateKey, "node-disk-manager/")
+	privateKey := filepath.Join(os.Getenv("NDM_HOME"), splitedResult[len(splitedResult)-1])
+	// Start new ssh connection with private key.
+	auth, err := goph.Key(privateKey, "")
+	require.Equal(s.T(), err, nil, "generate ssh auth key should not get error")
+
+	s.SSHClient, err = goph.NewUnknown(targetUser, targetHost, auth)
+	require.Equal(s.T(), err, nil, "New ssh connection should not get error")
+
+	kubeconfig := filepath.Join(os.Getenv("NDM_HOME"), "kubeconfig")
+	config, err := clientcmd.BuildConfigFromFlags("", kubeconfig)
+	require.Equal(s.T(), err, nil, "Generate kubeconfig should not get error")
+
+	s.coreClientSet, err = kubernetes.NewForConfig(config)
+	require.Equal(s.T(), err, nil, "New clientset(K8S) should not get error")
+
+	s.clientSet, err = clientset.NewForConfig(config)
+	require.Equal(s.T(), err, nil, "New clientset(NDM) should not get error")
+
+	cmd := fmt.Sprintf("ls %s |grep vagrant-k3s", os.Getenv("NDM_HOME"))
+	targetDirDomain, _, err := doCommand(cmd)
+	require.Equal(s.T(), err, nil, "Running command `%s` should not get error : %v", cmd, err)
+
+	s.hotplugTargetNodeName = fmt.Sprintf("%s_node1", strings.TrimSpace(targetDirDomain))
+	s.hotplugTargetBaseDir = fmt.Sprintf("/tmp/hotplug_disks/%s", strings.TrimSpace(targetDirDomain))
+
+	// cleanup the previous blockdevices
+	s.cleanupBlockDevices()
+
+	// we need to remove `NDM_AUTO_PROVISION_FILTER` to test other provisioner
+	s.patchDaemonSet()
+
+	// attach two disks to the target node
+	s.attachTwoDisks()
+}
+
+func (s *LVMSuite) AfterTest(_, _ string) {
+	if s.SSHClient != nil {
+		s.SSHClient.Close()
+	}
+	time.Sleep(5 * time.Second)
+}
+
+func TestLVMOperation(t *testing.T) {
+	suite.Run(t, new(LVMSuite))
+}
+
+func (s *LVMSuite) cleanupBlockDevices() {
+	bdi := s.clientSet.HarvesterhciV1beta1().BlockDevices("longhorn-system")
+	bdList, err := bdi.List(context.TODO(), v1.ListOptions{})
+	require.Equal(s.T(), err, nil, "List BlockDevices should not get error")
+
+	for _, blockdevice := range bdList.Items {
+		err := bdi.Delete(context.TODO(), blockdevice.Name, v1.DeleteOptions{})
+		require.Equal(s.T(), err, nil, "Delete BlockDevices should not get error")
+	}
+}
+
+func (s *LVMSuite) patchDaemonSet() {
+	currentDS, err := s.coreClientSet.AppsV1().DaemonSets("harvester-system").Get(context.TODO(), "harvester-node-disk-manager", v1.GetOptions{})
+	require.Equal(s.T(), err, nil, "Get DaemonSet should not get error")
+
+	newDS := currentDS.DeepCopy()
+	envs := newDS.Spec.Template.Spec.Containers[0].Env
+	newEnvs := []corev1.EnvVar{}
+	for _, item := range envs {
+		if item.Name == "NDM_AUTO_PROVISION_FILTER" {
+			continue
+		}
+		newEnvs = append(newEnvs, item)
+	}
+	newDS.Spec.Template.Spec.Containers[0].Env = newEnvs
+	_, err = s.coreClientSet.AppsV1().DaemonSets("harvester-system").Update(context.TODO(), newDS, v1.UpdateOptions{})
+	require.Equal(s.T(), err, nil, "Update DaemonSet should not get error")
+}
+
+func (s *LVMSuite) attachTwoDisks() {
+	// Create Target Disk.
+	// we can ignore the qcow2 file because we already have the disk files on test_1_disk_hotplug_test.go
+	firstDeviceRaw := fmt.Sprintf("%s/node1-sda.qcow2", s.hotplugTargetBaseDir)
+	firstDeviceXMLFile := fmt.Sprintf("%s/node1-sda.xml", s.hotplugTargetBaseDir)
+	secondDeviceRaw := fmt.Sprintf("%s/node1-sdb.qcow2", s.hotplugTargetBaseDir)
+	secondDeviceXMLFile := fmt.Sprintf("%s/node1-sdb.xml", s.hotplugTargetBaseDir)
+
+	disk, err := utils.DiskXMLReader(firstDeviceXMLFile)
+	require.Equal(s.T(), err, nil, "Read xml file should not get error")
+	disk.Source.File = firstDeviceRaw
+	disk.Target.Dev = "sda"
+	disk.VENDOR = "HAR_DEV1"
+	err = utils.XMLWriter(firstDeviceXMLFile, disk)
+	require.Equal(s.T(), err, nil, "Write xml file should not get error")
+
+	s.doAttachDisk(s.hotplugTargetNodeName, firstDeviceXMLFile)
+
+	disk, err = utils.DiskXMLReader(secondDeviceXMLFile)
+	require.Equal(s.T(), err, nil, "Read xml file should not get error")
+	newWWN := fmt.Sprintf("0x5000c50015%s", utils.GenHash())
+	disk.Source.File = secondDeviceRaw
+	disk.Target.Dev = "sdb"
+	disk.VENDOR = "HAR_DEV2"
+	disk.WWN = newWWN
+	err = utils.XMLWriter(secondDeviceXMLFile, disk)
+	require.Equal(s.T(), err, nil, "Write xml file should not get error")
+
+	s.doAttachDisk(s.hotplugTargetNodeName, secondDeviceXMLFile)
+
+}
+
+func (s *LVMSuite) Test_0_ProvisionLVMWithMultipleDisk() {
+	bdi := s.clientSet.HarvesterhciV1beta1().BlockDevices("longhorn-system")
+	bdList, err := bdi.List(context.TODO(), v1.ListOptions{})
+	require.Equal(s.T(), len(bdList.Items), 2, "BlockdevicesList should only have two devices")
+	require.Equal(s.T(), err, nil, "Get BlockdevicesList should not get error")
+	require.NotEqual(s.T(), len(bdList.Items), 0, "BlockdevicesList should not be empty")
+
+	for _, blockdevice := range bdList.Items {
+		if blockdevice.Spec.NodeName != s.targetNodeName {
+			// focus the target node
+			continue
+		}
+		bdStatus := blockdevice.Status
+		if bdStatus.DeviceStatus.Details.Vendor == "HAR_DEV1" {
+			s.targetDevFirst = blockdevice.DeepCopy()
+		} else if bdStatus.DeviceStatus.Details.Vendor == "HAR_DEV2" {
+			s.targetDevSecond = blockdevice.DeepCopy()
+		}
+	}
+	require.NotEqual(s.T(), nil, s.targetDevFirst, "targetDevFirst should not be empty")
+	require.NotEqual(s.T(), nil, s.targetDevSecond, "targetDevSecond should not be empty")
+
+	targetVGName := "test-vg01"
+	// provision first disks
+	s.targetDevFirst.Spec.Provision = true
+	s.targetDevFirst.Spec.Provisioner = &diskv1.ProvisionerInfo{
+		LVM: &diskv1.LVMProvisionerInfo{
+			VgName: targetVGName,
+		},
+	}
+	err = s.updateBlockdevice(s.targetDevFirst)
+	require.Equal(s.T(), err, nil, "Update BlockDevice(First) should not get error")
+
+	// provision second disks
+	s.targetDevSecond.Spec.Provision = true
+	s.targetDevSecond.Spec.Provisioner = &diskv1.ProvisionerInfo{
+		LVM: &diskv1.LVMProvisionerInfo{
+			VgName: targetVGName,
+		},
+	}
+	err = s.updateBlockdevice(s.targetDevSecond)
+	require.Equal(s.T(), err, nil, "Update BlockDevice(Second) should not get error")
+
+	// sleep 30 seconds to wait controller handle
+	time.Sleep(30 * time.Second)
+
+	targetLVMVG := s.getTargetLVMVG(targetVGName)
+	require.NotEqual(s.T(), nil, targetLVMVG, "targetLVM should not be empty")
+	require.Equal(s.T(), targetLVMVG.Status.Status, diskv1.VGStatusActive, "LVMVolumeGroup should be Active")
+	_, found := targetLVMVG.Spec.Devices[s.targetDevFirst.Name]
+	require.Equal(s.T(), found, true, "targetDevFirst should be in the LVMVolumeGroup")
+	_, found = targetLVMVG.Spec.Devices[s.targetDevSecond.Name]
+	require.Equal(s.T(), found, true, "targetDevSecond should be in the LVMVolumeGroup")
+}
+
+func (s *LVMSuite) Test_1_RemoveOneDiskOfLVMVolumeGroup() {
+	// remove second disk from the lvm volume group
+	targetVGName := s.targetDevSecond.Spec.Provisioner.LVM.VgName
+
+	s.targetDevSecond.Spec.Provision = false
+	s.targetDevSecond.Spec.Provisioner = nil
+	err := s.updateBlockdevice(s.targetDevSecond)
+	require.Equal(s.T(), err, nil, "Update BlockDevice should not get error")
+
+	// sleep 30 seconds to wait controller handle
+	time.Sleep(30 * time.Second)
+	targetLVMVG := s.getTargetLVMVG(targetVGName)
+
+	_, found := targetLVMVG.Spec.Devices[s.targetDevSecond.Name]
+	require.Equal(s.T(), found, false, "targetDevSecond should not be in the LVMVolumeGroup")
+}
+
+func (s *LVMSuite) Test_2_ProvisionAnotherLVMVG() {
+	// Provision another LVM VG
+	targetVGName := "test-vg02"
+	s.targetDevSecond.Spec.Provision = true
+	s.targetDevSecond.Spec.Provisioner = &diskv1.ProvisionerInfo{
+		LVM: &diskv1.LVMProvisionerInfo{
+			VgName: targetVGName,
+		},
+	}
+
+	err := s.updateBlockdevice(s.targetDevSecond)
+	require.Equal(s.T(), err, nil, "Update BlockDevice should not get error")
+
+	// sleep 30 seconds to wait controller handle
+	time.Sleep(30 * time.Second)
+
+	targetLVMVG := s.getTargetLVMVG(targetVGName)
+	require.NotEqual(s.T(), targetLVMVG, nil, "targetLVMVG should not be empty")
+	require.Equal(s.T(), targetLVMVG.Status.Status, diskv1.VGStatusActive, "LVMVolumeGroup should be Active")
+
+	_, found := targetLVMVG.Spec.Devices[s.targetDevSecond.Name]
+	require.Equal(s.T(), found, true, "targetDevSecond should be in the LVMVolumeGroup")
+}
+
+func (s *LVMSuite) Test_3_RemoveAllDisksOfLVMVolumeGroup() {
+	// remove all disks from the lvm volume group
+	targetVGNameFirstBD := s.targetDevSecond.Spec.Provisioner.LVM.VgName
+	s.targetDevFirst.Spec.Provision = false
+	s.targetDevFirst.Spec.Provisioner = nil
+	err := s.updateBlockdevice(s.targetDevFirst)
+	require.Equal(s.T(), err, nil, "Update BlockDevice(First) should not get error")
+
+	targetVGNameSecondBD := s.targetDevSecond.Spec.Provisioner.LVM.VgName
+	s.targetDevSecond.Spec.Provision = false
+	s.targetDevSecond.Spec.Provisioner = nil
+	err = s.updateBlockdevice(s.targetDevSecond)
+	require.Equal(s.T(), err, nil, "Update BlockDevice(Second) should not get error")
+
+	// sleep 30 seconds to wait controller handle
+	time.Sleep(30 * time.Second)
+
+	targetLVMVG := s.getTargetLVMVG(targetVGNameFirstBD)
+	require.Equal(s.T(), targetLVMVG, nil, "targetLVMVG should be empty")
+	targetLVMVG = s.getTargetLVMVG(targetVGNameSecondBD)
+	require.Equal(s.T(), targetLVMVG, nil, "targetLVMVG should be empty")
+}
+
+func (s *LVMSuite) doAttachDisk(nodeName, xmlFile string) {
+	cmd := fmt.Sprintf("virsh attach-device --domain %s --file %s --live", nodeName, xmlFile)
+	_, _, err := doCommand(cmd)
+	require.Equal(s.T(), err, nil, "Running command `%s` should not get error", cmd)
+
+	// wait for controller handling
+	time.Sleep(5 * time.Second)
+}
+
+func (s *LVMSuite) getTargetLVMVG(vgName string) *diskv1.LVMVolumeGroup {
+	lvmClient := s.clientSet.HarvesterhciV1beta1().LVMVolumeGroups("harvester-system")
+	lvmList, err := lvmClient.List(context.TODO(), v1.ListOptions{})
+	require.Equal(s.T(), err, nil, "Get LVMVolumeGroups should not get error")
+	require.NotEqual(s.T(), len(lvmList.Items), 0, "LVMVolumeGroups should not be empty")
+
+	targetLVM := &diskv1.LVMVolumeGroup{}
+	// find the LVM and check the status
+	for _, lvm := range lvmList.Items {
+		if lvm.Spec.NodeName == s.targetNodeName && lvm.Spec.VgName == vgName {
+			targetLVM = lvm.DeepCopy()
+		}
+	}
+	return targetLVM
+}
+
+func (s *LVMSuite) updateBlockdevice(bd *diskv1.BlockDevice) error {
+	bdi := s.clientSet.HarvesterhciV1beta1().BlockDevices("longhorn-system")
+	err := error(nil)
+	retry := 0
+	for retry < 3 {
+		_, err = bdi.Update(context.TODO(), bd, v1.UpdateOptions{})
+		if err != nil {
+			retry++
+			time.Sleep(10 * time.Second)
+			continue
+		}
+		return err
+	}
+	return err
+}