make pvc count in group configurable

cmd/manager/main.go

@@ -101,6 +101,7 @@ func main() {
 	flag.BoolVar(&showVersion, "version", false, "Print Version details")
 	flag.StringVar(&cfg.SchedulePrecedence, "schedule-precedence", "", "The order of precedence in which schedule of reclaimspace and keyrotation is considered. Possible values are sc-only")
 	flag.BoolVar(&enableAuth, "enable-auth", true, "Enables TLS and adds bearer token to the headers (enabled by default)")
+	flag.IntVar(&cfg.MaxGroupPVC, "max-group-pvc", cfg.MaxGroupPVC, "Maximum number of PVCs allowed in a volume group")
 	opts := zap.Options{
 		Development: true,
 		TimeEncoder: zapcore.ISO8601TimeEncoder,
@@ -244,9 +245,10 @@ func main() {
 	}
 
 	if err = (&replicationController.VolumeGroupReplicationReconciler{
-		Client:   mgr.GetClient(),
-		Scheme:   mgr.GetScheme(),
-		Recorder: mgr.GetEventRecorderFor("volumegroupreplication-controller"),
+		Client:           mgr.GetClient(),
+		Scheme:           mgr.GetScheme(),
+		Recorder:         mgr.GetEventRecorderFor("volumegroupreplication-controller"),
+		MaxGroupPVCCount: cfg.MaxGroupPVC,
 	}).SetupWithManager(mgr); err != nil {
 		setupLog.Error(err, "unable to create controller", "controller", "VolumeGroupReplication")
 		os.Exit(1)

deploy/controller/csi-addons-config.yaml

@@ -8,3 +8,4 @@ metadata:
 data:
   "reclaim-space-timeout": "3m"
   "max-concurrent-reconciles": "100"
+  "max-group-pvcs": "100"

docs/csi-addons-config.md

@@ -4,11 +4,14 @@ CSI-Addons Operator can consume configuration from a ConfigMap named `csi-addons
 in the same namespace as the operator. This enables configuration of the operator to persist across
 upgrades. The ConfigMap can support the following configuration options:
 
-| Option                      | Default value | Description                             |
-| --------------------------- | ------------- | --------------------------------------- |
-| `reclaim-space-timeout`     | `"3m"`        | Timeout for reclaimspace operation      |
-| `max-concurrent-reconciles` | `"100"`       | Maximum number of concurrent reconciles |
+| Option                      | Default value | Description                                      |
+| --------------------------- | ------------- | ------------------------------------------------ |
+| `reclaim-space-timeout`     | `"3m"`        | Timeout for reclaimspace operation               |
+| `max-concurrent-reconciles` | `"100"`       | Maximum number of concurrent reconciles          |
+| `max-group-pvcs`            | `"100"`       | Maximum number of PVCs allowed in a volume group |
 
 [`csi-addons-config` ConfigMap](../deploy/controller/csi-addons-config.yaml) is provided as an example.
 
 > Note: The operator pod needs to be restarted for any change in configuration to take effect.
+>
+> Note: `max-group-pvcs` default value is set based on ceph's support/testing. User can tweak this value based on the supported count for their storage vendor.

internal/controller/replication.storage/volumegroupreplication_controller.go

@@ -58,10 +58,11 @@ const (
 // VolumeGroupReplicationReconciler reconciles a VolumeGroupReplication object
 type VolumeGroupReplicationReconciler struct {
 	client.Client
-	Scheme   *runtime.Scheme
-	ctx      context.Context
-	log      logr.Logger
-	Recorder record.EventRecorder
+	Scheme           *runtime.Scheme
+	ctx              context.Context
+	log              logr.Logger
+	Recorder         record.EventRecorder
+	MaxGroupPVCCount int
 }
 
 //+kubebuilder:rbac:groups="",resources=events,verbs=create;patch
@@ -162,9 +163,9 @@ func (r *VolumeGroupReplicationReconciler) Reconcile(ctx context.Context, req ct
 			_ = r.setGroupReplicationFailure(instance, err)
 			return reconcile.Result{}, err
 		}
-		if len(pvcList) > 100 {
-			err = fmt.Errorf("more than 100 PVCs match the given selector")
-			r.log.Error(err, "only 100 PVCs are allowed for volume group replication")
+		if len(pvcList) > r.MaxGroupPVCCount {
+			err = fmt.Errorf("more than %q PVCs match the given selector", r.MaxGroupPVCCount)
+			r.log.Error(err, "only %q PVCs are allowed for volume group replication", r.MaxGroupPVCCount)
 			_ = r.setGroupReplicationFailure(instance, err)
 			return reconcile.Result{}, err
 		}

internal/controller/replication.storage/volumegroupreplication_test.go

@@ -97,10 +97,11 @@ func createFakeVolumeGroupReplicationReconciler(t *testing.T, obj ...runtime.Obj
 	reconcilerCtx := context.TODO()
 
 	return VolumeGroupReplicationReconciler{
-		Client: client,
-		Scheme: scheme,
-		log:    logger,
-		ctx:    reconcilerCtx,
+		Client:           client,
+		Scheme:           scheme,
+		log:              logger,
+		ctx:              reconcilerCtx,
+		MaxGroupPVCCount: 100,
 	}
 }
 

internal/util/config.go

@@ -34,6 +34,7 @@ type Config struct {
 	ReclaimSpaceTimeout     time.Duration
 	MaxConcurrentReconciles int
 	SchedulePrecedence      string
+	MaxGroupPVC             int
 }
 
 const (
@@ -45,6 +46,8 @@ const (
 	defaultReclaimSpaceTimeout     = time.Minute * 3
 	SchedulePrecedenceKey          = "schedule-precedence"
 	ScheduleSCOnly                 = "sc-only"
+	MaxGroupPVCKey                 = "max-group-pvcs"
+	defaultMaxGroupPVC             = 100 // based on ceph's support/testing
 )
 
 // NewConfig returns a new Config object with default values.
@@ -54,6 +57,7 @@ func NewConfig() Config {
 		ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 		MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 		SchedulePrecedence:      "",
+		MaxGroupPVC:             defaultMaxGroupPVC,
 	}
 }
 
@@ -99,6 +103,17 @@ func (cfg *Config) readConfig(dataMap map[string]string) error {
 			}
 			cfg.SchedulePrecedence = val
 
+		case MaxGroupPVCKey:
+			maxGroupPVCs, err := strconv.Atoi(val)
+			if err != nil {
+				return fmt.Errorf("failed to parse key %q value %q as int: %w",
+					MaxGroupPVCKey, val, err)
+			}
+			if maxGroupPVCs <= 0 || maxGroupPVCs > 100 {
+				return fmt.Errorf("invalid value %q for key %q", val, MaxGroupPVCKey)
+			}
+			cfg.MaxGroupPVC = maxGroupPVCs
+
 		default:
 			return fmt.Errorf("unknown config key %q", key)
 		}

internal/util/config_test.go

@@ -38,6 +38,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -49,6 +50,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -62,6 +64,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     time.Minute * 10,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -75,6 +78,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: true,
 		},
@@ -88,6 +92,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: 1,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -101,6 +106,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: true,
 		},
@@ -115,6 +121,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     time.Minute * 10,
 				MaxConcurrentReconciles: 5,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -128,6 +135,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: true,
 		},
@@ -141,6 +149,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      ScheduleSCOnly,
+				MaxGroupPVC:             100,
 			},
 			wantErr: false,
 		},
@@ -154,6 +163,7 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: true,
 		},
@@ -167,6 +177,49 @@ func TestConfigReadConfigFile(t *testing.T) {
 				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
 				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
 				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
+			},
+			wantErr: true,
+		},
+		{
+			name: "config file has empty max-group-pvcs",
+			dataMap: map[string]string{
+				"max-group-pvcs": "",
+			},
+			newConfig: Config{
+				Namespace:               defaultNamespace,
+				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
+				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
+				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
+			},
+			wantErr: true,
+		},
+		{
+			name: "config file modifies max-group-pvcs",
+			dataMap: map[string]string{
+				"max-group-pvcs": "25",
+			},
+			newConfig: Config{
+				Namespace:               defaultNamespace,
+				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
+				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
+				SchedulePrecedence:      "",
+				MaxGroupPVC:             25,
+			},
+			wantErr: false,
+		},
+		{
+			name: "config file has invalid max-group-pvcs",
+			dataMap: map[string]string{
+				"max-group-pvcs": "200",
+			},
+			newConfig: Config{
+				Namespace:               defaultNamespace,
+				ReclaimSpaceTimeout:     defaultReclaimSpaceTimeout,
+				MaxConcurrentReconciles: defaultMaxConcurrentReconciles,
+				SchedulePrecedence:      "",
+				MaxGroupPVC:             100,
 			},
 			wantErr: true,
 		},