cluster: handle data PVC upsize

.github/workflows/e2e-tests.yml

@@ -79,7 +79,7 @@ jobs:
     strategy:
       max-parallel: 2
       matrix:
-        test-name: [operations, sidecars, scaling, multi-dcs, backup-restore, nodetool]
+        test-name: [operations, sidecars, scaling, multi-dcs, backup-restore, nodetool, storage-upsize]
     steps:
       - id: lower-repo
         shell: pwsh
@@ -105,13 +105,65 @@ jobs:
           ref: ${{ steps.get-branch.outputs.branch }}
 
       - uses: rinx/setup-k3d@v0.0.4
+        if: matrix.test-name != 'storage-upsize'
         with:
           version: v5.4.0
           options: --image rancher/k3s:v1.24.13-k3s1
 
-      - uses: azure/setup-helm@v1
+      - name: Create kind config with LVM mounts
+        if: matrix.test-name == 'storage-upsize'
+        run: |
+          mkdir -p kindAndLvm
+          cat <<EOF > kindAndLvm/kind-lvm-config.yaml
+          kind: Cluster
+          apiVersion: kind.x-k8s.io/v1alpha4
+          nodes:
+          - role: control-plane
+          - role: worker
+            extraMounts:
+            - hostPath: /tmp/kind-worker1-lvm
+              containerPath: /mnt/disks
+          - role: worker
+            extraMounts:
+            - hostPath: /tmp/kind-worker2-lvm
+              containerPath: /mnt/disks
+          EOF
+
+          mkdir -p /tmp/kind-worker1-lvm /tmp/kind-worker2-lvm
+
+      - name: Setup kind cluster
+        if: matrix.test-name == 'storage-upsize'
+        uses: helm/kind-action@v1.8.0
         with:
-          version: v3.8.1
+          cluster_name: lvm-test
+          config: kindAndLvm/kind-lvm-config.yaml
+          wait: 300s
+
+      - name: Setup LVM on worker nodes
+        if: matrix.test-name == 'storage-upsize'
+        run: |
+          WORKER_NODES=$(kubectl get nodes -o name | grep worker | sed 's|node/||')
+
+          for NODE in $WORKER_NODES; do
+            echo "Setting up LVM on $NODE..."
+            docker exec $NODE bash -c '
+              apt-get update -qq && apt-get install -y -qq lvm2 thin-provisioning-tools > /dev/null 2>&1
+              mkdir -p /mnt/disks
+              truncate -s 10G /mnt/disks/disk.img
+              LOOP_DEVICE=$(losetup -f)
+              losetup $LOOP_DEVICE /mnt/disks/disk.img
+              pvcreate $LOOP_DEVICE
+              vgcreate lvmvg $LOOP_DEVICE
+              vgs
+            '
+          done
+
+      - name: Install OpenEBS LVM LocalPV
+        if: matrix.test-name == 'storage-upsize'
+        run: |
+          kubectl apply -f https://openebs.github.io/charts/lvm-operator.yaml
+          kubectl wait --for=condition=ready pod -l app=openebs-lvm-controller -n kube-system --timeout=300s
+          kubectl rollout status daemonset/openebs-lvm-node -n kube-system --timeout=300s
 
       - name: Restore Kuttl
         id: cache-kuttl

api/v2/cassandracluster_types.go

@@ -70,6 +70,8 @@ var (
 
 	ActionCorrectCRDConfig = ClusterStateInfo{11, "CorrectCRDConfig"} //The Operator has correct a bad CRD configuration
 
+	ActionStorageUpsize = ClusterStateInfo{12, "StorageUpsize"}
+
 	regexDCRackName = regexp.MustCompile("^[a-z]([-a-z0-9]*[a-z0-9])?$")
 )
 
@@ -232,6 +234,10 @@ func (cc *CassandraCluster) GetDCName(dc int) string {
 	return cc.Spec.Topology.DC[dc].Name
 }
 
+func (cc *CassandraCluster) GetDCNameStrongType(dc int) DcName {
+	return DcName(cc.GetDCName(dc))
+}
+
 func (cc *CassandraCluster) getDCNodesPerRacksFromIndex(dc int) int32 {
 	if dc >= cc.GetDCSize() {
 		return cc.Spec.NodesPerRacks
@@ -462,6 +468,11 @@ func (cc *CassandraCluster) GetDataCapacityForDC(dcName string) string {
 	return cc.GetDataCapacityFromDCName(dcName)
 }
 
+// GetDataCapacityForDC sends back the data capacity of cassandra nodes for the given strongly-typed dcName
+func (cc *CassandraCluster) GetDataCapacityForDCName(dcName DcName) string {
+	return cc.GetDataCapacityFromDCName(dcName.String())
+}
+
 // GetDataCapacityFromDCName send DataCapacity used for the given dcName
 func (cc *CassandraCluster) GetDataCapacityFromDCName(dcName string) string {
 	dcIndex := cc.GetDCIndexFromDCName(dcName)
@@ -480,6 +491,11 @@ func (cc *CassandraCluster) GetDataStorageClassForDC(dcName string) string {
 	return cc.GetDataStorageClassFromDCName(dcName)
 }
 
+// GetDataStorageClassForDCName send DataStorageClass used for the given strongly-typed dcName
+func (cc *CassandraCluster) GetDataStorageClassForDCName(dcName DcName) string {
+	return cc.GetDataStorageClassFromDCName(dcName.String())
+}
+
 // GetDataCapacityFromDCName send DataStorageClass used for the given dcName
 func (cc *CassandraCluster) GetDataStorageClassFromDCName(dcName string) string {
 	dcIndex := cc.GetDCIndexFromDCName(dcName)
@@ -539,6 +555,11 @@ func (cc *CassandraCluster) GetNodesPerRacks(dcRackName string) int32 {
 	return nodesPerRacks
 }
 
+// GetNodesPerRacks sends back the number of cassandra nodes to uses for this strongly-typed dc-rack
+func (cc *CassandraCluster) GetNodesPerRacksStrongType(dcRackName DcRackName) int32 {
+	return cc.GetNodesPerRacks(dcRackName.String())
+}
+
 // GetDCNodesPerRacksFromDCRackName send NodesPerRack used for the given dcRackName
 func (cc *CassandraCluster) GetDCRackNames() []string {
 	dcsize := cc.GetDCSize()
@@ -931,6 +952,11 @@ type BackRestSidecar struct {
 	VolumeMounts []v1.VolumeMount         `json:"volumeMount,omitempty"`
 }
 
+// GetCassandraRackStatus returns CassandraRackStatus for a given strongly-typed dcRack
+func (in *CassandraClusterStatus) GetCassandraRackStatus(dcRackName DcRackName) *CassandraRackStatus {
+	return in.CassandraRackStatus[dcRackName.String()]
+}
+
 // CassandraRackStatus defines states of Cassandra for 1 rack (1 statefulset)
 type CassandraRackStatus struct {
 	// Phase indicates the state this Cassandra cluster jumps in.
@@ -943,6 +969,10 @@ type CassandraRackStatus struct {
 
 	// PodLastOperation manage status for Pod Operation (nodetool cleanup, upgradesstables..)
 	PodLastOperation PodLastOperation `json:"podLastOperation,omitempty"`
+
+	// StatefulSetSnapshotBeforeStorageResize is the StatefulSet snapshot taken before storage resize
+	// The purpose is to isolate the storage resize operation from other operations
+	StatefulSetSnapshotBeforeStorageResize string `json:"statefulSetSnapshotBeforeStorageResize,omitempty"`
 }
 
 // CassandraClusterStatus defines Global state of CassandraCluster

api/v2/names.go

@@ -0,0 +1,25 @@
+package v2
+
+type DcName string
+
+func (n DcName) String() string {
+	return string(n)
+}
+
+type RackName string
+
+func (n RackName) String() string {
+	return string(n)
+}
+
+type DcRackName string
+
+func (n DcRackName) String() string {
+	return string(n)
+}
+
+type CompleteRackName struct {
+	DcName     DcName
+	RackName   RackName
+	DcRackName DcRackName
+}

charts/casskop/crds/db.orange.com_cassandraclusters.yaml

@@ -2308,6 +2308,11 @@ spec:
                             format: date-time
                           status:
                             type: string
+                      statefulSetSnapshotBeforeStorageResize:
+                        description: |-
+                          StatefulSetSnapshotBeforeStorageResize is the StatefulSet snapshot taken before storage resize
+                          The purpose is to isolate the storage resize operation from other operations
+                        type: string
                 lastClusterAction:
                   description: Store last action at cluster level
                   type: string

charts/multi-casskop/crds/db.orange.com_cassandraclusters.yaml

@@ -2308,6 +2308,11 @@ spec:
                             format: date-time
                           status:
                             type: string
+                      statefulSetSnapshotBeforeStorageResize:
+                        description: |-
+                          StatefulSetSnapshotBeforeStorageResize is the StatefulSet snapshot taken before storage resize
+                          The purpose is to isolate the storage resize operation from other operations
+                        type: string
                 lastClusterAction:
                   description: Store last action at cluster level
                   type: string

config/crd/bases/db.orange.com_cassandraclusters.yaml

@@ -2308,6 +2308,11 @@ spec:
                             format: date-time
                           status:
                             type: string
+                      statefulSetSnapshotBeforeStorageResize:
+                        description: |-
+                          StatefulSetSnapshotBeforeStorageResize is the StatefulSet snapshot taken before storage resize
+                          The purpose is to isolate the storage resize operation from other operations
+                        type: string
                 lastClusterAction:
                   description: Store last action at cluster level
                   type: string