skip remedidiation for excluded nodes

Signed-off-by: Michael Shitrit <[email protected]>

Author: mshitrit

controllers/nodehealthcheck_controller.go

@@ -28,6 +28,7 @@ import (
 	commonconditions "github.com/medik8s/common/pkg/conditions"
 	"github.com/medik8s/common/pkg/etcd"
 	commonevents "github.com/medik8s/common/pkg/events"
+	commonlabels "github.com/medik8s/common/pkg/labels"
 	"github.com/medik8s/common/pkg/lease"
 	"github.com/medik8s/common/pkg/nodes"
 	"github.com/pkg/errors"
@@ -353,6 +354,13 @@ func (r *NodeHealthCheckReconciler) Reconcile(ctx context.Context, req ctrl.Requ
 			continue
 		}
 
+		if r.isNodeRemediationExcluded(&node) {
+			msg := fmt.Sprintf("Skipped remediation because node %s is marked to exclude remediations", node.GetName())
+			log.Info(msg)
+			commonevents.WarningEvent(r.Recorder, nhc, utils.EventReasonRemediationSkipped, msg)
+			continue
+		}
+
 		log.Info("handling unhealthy node", "node", node.GetName())
 		requeueAfter, err := r.remediate(ctx, &node, nhc, resourceManager)
 		if err != nil {
@@ -872,6 +880,15 @@ func (r *NodeHealthCheckReconciler) addRemediationTemplateCRWatch(templateCR *un
 	return nil
 }
 
+func (r *NodeHealthCheckReconciler) isNodeRemediationExcluded(node *v1.Node) bool {
+	if nodeLabels := node.GetLabels(); nodeLabels == nil {
+		return false
+	} else {
+		_, isNodeExcluded := nodeLabels[commonlabels.ExcludeFromRemediation]
+		return isNodeExcluded
+	}
+}
+
 func getTimeoutAt(remediation *remediationv1alpha1.Remediation, configuredTimeout *time.Duration) time.Time {
 	return remediation.Started.Add(*configuredTimeout)
 }

controllers/nodehealthcheck_controller_test.go

@@ -815,6 +815,20 @@ var _ = Describe("Node Health Check CR", func() {
 
 		})
 
+		Context("with Node marked for excluding remediation", func() {
+			BeforeEach(func() {
+				objects = newNodes(1, 2, false, true)
+				objects = append(objects, newNodeHealthCheck())
+				node := objects[0].(*v1.Node)
+				node.GetLabels()[commonLabels.ExcludeFromRemediation] = "true"
+
+			})
+			It("remediation shouldn't be created", func() {
+				Expect(underTest.Status.UnhealthyNodes).To(HaveLen(1))
+				Expect(underTest.Status.UnhealthyNodes[0].Remediations).To(HaveLen(0))
+			})
+		})
+
 		Context("with a single escalating remediation", func() {
 
 			BeforeEach(func() {