Add systemd-restart annotation for firewall resource.

README.md

@@ -41,6 +41,23 @@ A user can initiate rolling the latest firewall set by annotating a monitor in t
 kubectl annotate fwmon <firewall-name> firewall.metal-stack.io/roll-set=true
 ```
 
+## Restarting a systemd-service on the Firewall through Annotation
+
+A user can initiate the restart of a systemd service through annotating the `FirewallMonitor`:
+
+```bash
+kubectl annotate fwmon <firewall-name> firewall.metal-stack.io/restart-systemd-services=tailscale
+```
+
+The firewall-controller imeplements a whitelist of allowed services to restart.
+
+In addition, operators can overwrite this whitelist and also initiate a service restart by annotating the `Firewall` resource instead of the `FirewallMonitor`:
+
+```bash
+kubectl annotate fw <firewall-name> firewall.metal-stack.io/restart-systemd-services-whitelist=tailscale,frr
+kubectl annotate fw <firewall-name> firewall.metal-stack.io/restart-systemd-services=frr
+```
+
 ## Development
 
 Most of the functionality is developed with the help of the [integration](integration) test suite.

api/v2/types_annotations.go

@@ -36,6 +36,13 @@ const (
 	// Defaults to 0 if no annotation is present. Negative values are allowed.
 	FirewallWeightAnnotation = "firewall.metal-stack.io/weight"
 
+	// FirewallRestartSystemdServicesAnnotation can be used to restart a whitelisted set of systemd services running on the firewall.
+	// Services must be passed comma-separated.
+	FirewallRestartSystemdServicesAnnotation = "firewall.metal-stack.io/restart-systemd-services"
+	// FirewallSystemdServicesWhitelistAnnotation can be used to overwrite the default systemd service whitelisted. This allows operators
+	// to temporarily allow restarts of services like FRR, which might not be desired to be allowed permanently for platform users.
+	FirewallRestartSystemdServicesWhitelistAnnotation = "firewall.metal-stack.io/restart-systemd-services-whitelist"
+
 	// FirewallControllerSetAnnotation is a tag added to the firewall entity indicating to which set a firewall belongs to.
 	FirewallControllerSetAnnotation = "firewall.metal.stack.io/set"
 )

controllers/firewall/reconcile.go

@@ -12,14 +12,42 @@ import (
 	"github.com/metal-stack/metal-go/api/client/machine"
 	"github.com/metal-stack/metal-go/api/models"
 	"github.com/metal-stack/metal-lib/pkg/pointer"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	corev1 "k8s.io/api/core/v1"
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/sets"
 )
 
 // Reconciler must always return either an error or requeue to ensure that it detects if a firewall get lost etc.
 func (c *controller) Reconcile(r *controllers.Ctx[*v2.Firewall]) error {
+	if services, ok := r.Target.GetAnnotations()[v2.FirewallRestartSystemdServicesAnnotation]; ok {
+		mon := &v2.FirewallMonitor{
+			ObjectMeta: metav1.ObjectMeta{
+				Name:      r.Target.Name,
+				Namespace: c.c.GetShootNamespace(),
+			},
+		}
+
+		err := c.c.GetShootClient().Get(r.Ctx, client.ObjectKeyFromObject(mon), mon)
+		if err != nil && !apierrors.IsNotFound(err) {
+			return fmt.Errorf("unable to get firewall monitor: %w", err)
+		}
+
+		if err == nil {
+			if err := v2.AddAnnotation(r.Ctx, c.c.GetShootClient(), mon, v2.FirewallRestartSystemdServicesAnnotation, services); err != nil {
+				return fmt.Errorf("unable to pass systemd service restart annotation to the firewall monitor: %w", err)
+			}
+		}
+
+		if err := v2.RemoveAnnotation(r.Ctx, c.c.GetSeedClient(), r.Target, v2.FirewallRestartSystemdServicesAnnotation); err != nil {
+			return fmt.Errorf("unable to remove systemd service restart annotation from firewall: %w", err)
+		}
+
+		return controllers.RequeueAfter(0*time.Second, "removed systemd service restart annotation, requeue for regular reconcile")
+	}
+
 	var f *models.V1FirewallResponse
 	defer func() {
 		if err := c.setStatus(r, f); err != nil {

integration/integration_test.go

@@ -441,6 +441,48 @@ var _ = Context("integration test", Ordered, func() {
 			})
 		})
 
+		When("the firewall gets annotated with a systemd service restart annotation", Ordered, func() {
+			var (
+				fw  *v2.Firewall
+				mon *v2.FirewallMonitor
+			)
+
+			BeforeEach(func() {
+				fw = testcommon.WaitForResourceAmount(k8sClient, ctx, namespaceName, 1, &v2.FirewallList{}, func(l *v2.FirewallList) []*v2.Firewall {
+					return l.GetItems()
+				}, 2*time.Second)
+				mon = testcommon.WaitForResourceAmount(k8sClient, ctx, namespaceName, 1, &v2.FirewallMonitorList{}, func(l *v2.FirewallMonitorList) []*v2.FirewallMonitor {
+					return l.GetItems()
+				}, 2*time.Second)
+			})
+
+			It("setting the annotation works", func() {
+				fw.Annotations = map[string]string{
+					v2.FirewallRestartSystemdServicesAnnotation: "droptailer",
+				}
+				Expect(k8sClient.Update(ctx, fw)).To(Succeed())
+			})
+
+			It("the annotation gets removed from the firewall", func() {
+				Eventually(func() map[string]string {
+					Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(fw), fw)).To(Succeed())
+					return fw.Annotations
+				}, 5*time.Second, interval).Should(Not(HaveKey(v2.FirewallRestartSystemdServicesAnnotation)), "systemd service restart annotation was not removed")
+			})
+
+			It("the annotation was added to the firewall monitor", func() {
+				Eventually(func() map[string]string {
+					Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(mon), mon)).To(Succeed())
+					return mon.Annotations
+				}, 5*time.Second, interval).Should(HaveKey(v2.FirewallRestartSystemdServicesAnnotation), "systemd service restart annotation was not added to the firewall monitor")
+			})
+
+			It("removing the annotation from the monitor works", func() {
+				mon.Annotations = nil
+				Expect(k8sClient.Update(ctx, mon)).To(Succeed())
+			})
+		})
+
 		When("a significant change occurs", Ordered, func() {
 			var (
 				installingFirewall = firewall2("Installing", "is installing")