fix: do not allow scaling down controlplane to zero

Doing that is pretty pointless as it's going to destroy a cluster. Instead of doing that cluster should be deleted and recreated. And with the new approach without init nodes scaling cluster back up won't be able to succeed as bootstrap is done only a single time. Still allowing user to set replicas to zero, but block scaling in the controller with error messages when there's only a single node remaining. Signed-off-by: Artem Chernyshev <[email protected]>
siderolabs · Nov 26, 2021 · a24dad3 · a24dad3
1 parent 8a73e6a
commit a24dad3
Show file tree

Hide file tree

Showing 2 changed files with 49 additions and 0 deletions.
diff --git a/controllers/taloscontrolplane_controller.go b/controllers/taloscontrolplane_controller.go
@@ -273,6 +273,14 @@ func (r *TalosControlPlaneReconciler) Reconcile(ctx context.Context, req ctrl.Re
 			"Scaling down control plane to %d replicas (actual %d)",
 			desiredReplicas, numMachines)
 
+		if numMachines == 1 {
+			conditions.MarkFalse(tcp, controlplanev1.ResizedCondition, controlplanev1.ScalingDownReason, clusterv1.ConditionSeverityError,
+				"Cannot scale down control plane nodes to 0",
+				desiredReplicas, numMachines)
+
+			return res, nil
+		}
+
 		if err := r.ensureNodesBooted(ctx, cluster, ownedMachines); err != nil {
 			logger.Info("Waiting for all nodes to finish boot sequence", "error", err)
 

diff --git a/internal/integration/integration_test.go b/internal/integration/integration_test.go
@@ -23,10 +23,14 @@ import (
 	"gopkg.in/yaml.v3"
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 	"k8s.io/apimachinery/pkg/labels"
+	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/runtime/schema"
 	clusterv1 "sigs.k8s.io/cluster-api/cmd/clusterctl/api/v1alpha3"
 	logf "sigs.k8s.io/cluster-api/cmd/clusterctl/log"
+	"sigs.k8s.io/cluster-api/util/conditions"
 	runtimeclient "sigs.k8s.io/controller-runtime/pkg/client"
+
+	controlplanev1 "github.com/talos-systems/cluster-api-control-plane-provider-talos/api/v1alpha3"
 )
 
 var (
@@ -286,6 +290,43 @@ func (suite *IntegrationSuite) Test04ScaleControlPlaneNoWait() {
 	suite.Require().NoError(err)
 }
 
+// Test05ScaleControlPlaneToZero try to scale control plane to zero and check that it never does that.
+func (suite *IntegrationSuite) Test05ScaleControlPlaneToZero() {
+	ctx, cancel := context.WithCancel(suite.ctx)
+
+	go func() {
+		time.Sleep(time.Second * 5)
+
+		cancel()
+	}()
+
+	suite.cluster.Scale(ctx, 0, capi.ControlPlaneNodes) //nolint:errcheck
+
+	err := retry.Constant(time.Second*20, retry.WithUnits(time.Second)).Retry(func() error {
+		controlplane, err := suite.cluster.ControlPlanes(suite.ctx)
+		if err != nil {
+			return err
+		}
+
+		var tcp controlplanev1.TalosControlPlane
+
+		err = runtime.DefaultUnstructuredConverter.
+			FromUnstructured(controlplane.UnstructuredContent(), &tcp)
+		if err != nil {
+			return err
+		}
+
+		if !conditions.Has(&tcp, controlplanev1.ResizedCondition) &&
+			conditions.GetMessage(&tcp, controlplanev1.ResizedCondition) != "Cannot scale down control plane nodes to 0" {
+			return retry.ExpectedErrorf("node resized conditions error status hasn't updated")
+		}
+
+		return nil
+	})
+
+	suite.Require().NoError(err)
+}
+
 // TestIntegration runs integration tests.
 func TestIntegration(t *testing.T) {
 	version := os.Getenv("WORKLOAD_TALOS_VERSION")