Fix pg_autoscaler won't scale due to overlapping roots

There was no CephBlockPool defined for the .mgr pool, therefore it was using the default crush rule (with no deviceClass specified). Also the CephObjectStore & CephFilesystem metadataPool does not define the deviceClass (it is only set in the dataPools). To fix this Create a CephBlockPool CR for the .mgr pool with the same poolSpec as the default CephBlockPool. Also set the deviceClass for the metadataPool in CephObjectStore and CephFilesystem. Signed-off-by: Malay Kumar Parida <[email protected]>
red-hat-storage · Jul 12, 2024 · bf8120c · bf8120c
1 parent 2939ab0
commit bf8120c
Show file tree

Hide file tree

Showing 4 changed files with 23 additions and 3 deletions.
diff --git a/controllers/storagecluster/cephblockpools.go b/controllers/storagecluster/cephblockpools.go
@@ -44,6 +44,8 @@ func (r *StorageClusterReconciler) addPeerSecretsToCephBlockPool(initData *ocsv1
 func (r *StorageClusterReconciler) newCephBlockPoolInstances(initData *ocsv1.StorageCluster) ([]*cephv1.CephBlockPool, error) {
 	var mirroringSpec cephv1.MirroringSpec
 	poolName := generateNameForCephBlockPool(initData)
+	// This name is used by UI to hide this pool from the list of CephBlockPools
+	builtinMgrPoolName := "builtin-mgr"
 	poolNamespace := initData.Namespace
 
 	if initData.Spec.Mirroring.Enabled {
@@ -69,6 +71,20 @@ func (r *StorageClusterReconciler) newCephBlockPoolInstances(initData *ocsv1.Sto
 				},
 			},
 		},
+		{
+			ObjectMeta: metav1.ObjectMeta{
+				Name:      builtinMgrPoolName,
+				Namespace: poolNamespace,
+			},
+			Spec: cephv1.NamedBlockPoolSpec{
+				Name: ".mgr",
+				PoolSpec: cephv1.PoolSpec{
+					DeviceClass:   initData.Status.DefaultCephDeviceClass,
+					FailureDomain: getFailureDomain(initData),
+					Replicated:    generateCephReplicatedSpec(initData, "metadata"),
+				},
+			},
+		},
 	}
 
 	// Create Non-Resilient CephBlockPools if enabled

diff --git a/controllers/storagecluster/cephblockpools_test.go b/controllers/storagecluster/cephblockpools_test.go
@@ -171,8 +171,8 @@ func assertCephNFSBlockPool(t *testing.T, reconciler StorageClusterReconciler, c
 	expectedAf, err := reconciler.newCephBlockPoolInstances(cr)
 	assert.NoError(t, err)
 
-	assert.Equal(t, len(expectedAf[1].OwnerReferences), 1)
+	assert.Equal(t, len(expectedAf[2].OwnerReferences), 1)
 
-	assert.Equal(t, expectedAf[1].ObjectMeta.Name, actualNFSBlockPool.ObjectMeta.Name)
-	assert.Equal(t, expectedAf[1].Spec, actualNFSBlockPool.Spec)
+	assert.Equal(t, expectedAf[2].ObjectMeta.Name, actualNFSBlockPool.ObjectMeta.Name)
+	assert.Equal(t, expectedAf[2].Spec, actualNFSBlockPool.Spec)
 }
diff --git a/controllers/storagecluster/cephfilesystem.go b/controllers/storagecluster/cephfilesystem.go
@@ -79,6 +79,9 @@ func (r *StorageClusterReconciler) newCephFilesystemInstances(initStorageCluster
 		}
 	}
 
+	// set device class for metadata pool from the default data pool
+	ret.Spec.MetadataPool.DeviceClass = ret.Spec.DataPools[0].PoolSpec.DeviceClass
+
 	err := controllerutil.SetControllerReference(initStorageCluster, ret, r.Scheme)
 	if err != nil {
 		r.Log.Error(err, "Unable to set Controller Reference for CephFileSystem.", "CephFileSystem", klog.KRef(ret.Namespace, ret.Name))

diff --git a/controllers/storagecluster/cephobjectstores.go b/controllers/storagecluster/cephobjectstores.go
@@ -173,6 +173,7 @@ func (r *StorageClusterReconciler) newCephObjectStoreInstances(initData *ocsv1.S
 					Replicated:    generateCephReplicatedSpec(initData, "data"),
 				},
 				MetadataPool: cephv1.PoolSpec{
+					DeviceClass:   initData.Status.DefaultCephDeviceClass,
 					FailureDomain: initData.Status.FailureDomain,
 					Replicated:    generateCephReplicatedSpec(initData, "metadata"),
 				},