systemdunitcheck: Do not report unhealthy status for a disabled unit that is in failed state

Signed-off-by: cerealsnow <13483299+cerealsnow@users.noreply.github.com>

This commit can be dropped when upgrading to a gardener version that contains this bugfix.

Author: cerealsnow

pkg/nodeagent/controller/systemdunitcheck/reconciler.go

@@ -168,7 +168,10 @@ func (r *Reconciler) checkUnits(ctx context.Context, units []unitInfo) (unhealth
 			}
 
 		case "failed":
-			unhealthyMessages = append(unhealthyMessages, fmt.Sprintf("%s: failed", unit.name))
+			// Only enabled units should be marked as erroring in case they are in a failed state.
+			if unit.enabled {
+				unhealthyMessages = append(unhealthyMessages, fmt.Sprintf("%s: failed", unit.name))
+			}
 
 		case "activating", "deactivating":
 			// Services configured with Restart=always/on-success cycle through "activating (auto-restart)"

test/integration/nodeagent/systemdunitcheck/systemdunitcheck_test.go

@@ -302,6 +302,28 @@ var _ = Describe("SystemdUnitCheck controller tests", func() {
 		})))
 	})
 
+	It("should not report unhealthy for a disabled unit that is in a failed state", func() {
+		writeOSC(&extensionsv1alpha1.OperatingSystemConfig{
+			Spec: extensionsv1alpha1.OperatingSystemConfigSpec{
+				Units: []extensionsv1alpha1.Unit{
+					{Name: "optional.service", Enable: ptr.To(false)},
+				},
+			},
+		})
+
+		fakeDBus.AddUnitsToList(
+			systemddbus.UnitStatus{Name: "optional.service", ActiveState: "failed"},
+		)
+
+		Eventually(func(g Gomega) *corev1.NodeCondition {
+			return getNodeCondition(g)
+		}).Should(PointTo(MatchFields(IgnoreExtras, Fields{
+			"Type":   Equal(nodeagentconfigv1alpha1.ConditionTypeSystemdUnitsReady),
+			"Status": Equal(corev1.ConditionTrue),
+			"Reason": Equal("AllUnitsHealthy"),
+		})))
+	})
+
 	It("should also monitor gardener-node-agent's own units", func() {
 		writeOSC(&extensionsv1alpha1.OperatingSystemConfig{
 			Spec: extensionsv1alpha1.OperatingSystemConfigSpec{