✨ Add support for checking Machine conditions in MachineHealthCheck

[furkatgofurov7]

What this PR does / why we need it:
MachineHealthCheck currently only allows checking Node conditions to validate if a machine is healthy. However, machine conditions capture conditions that do not exist on nodes, for example, control plane node conditions such as EtcdPodHealthy, SchedulerPodHealthy that can indicate if a controlplane machine has been created correctly.

Adding support for Machine conditions enables us to perform remediation during control plane upgrades.

This PR introduces a new field as part of the MachineHealthCheckChecks:

• UnhealthyMachineConditions

This will mirror the behavior of UnhealthyNodeConditions but the MachineHealthCheck controller will instead check the machine conditions.

This reimplements and extends the work originally proposed by @justinmir in PR #12275.

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):
Fixes: #5450
Part of #12291

Label(s) to be applied
/kind feature
/area machinehealthcheck

Notes for Reviewers
Additional notes on what has changed in this PR to get a general idea of what this change is trying to achieve.

MHC related tests:
We updated the tests to validate the new MachineHealthCheck code paths for UnhealthyMachineConditions:

• internal/controllers/machinehealthcheck/machinehealthcheck_controller_test.go: Added a test case with UnhealthyMachineConditions to verify machine condition evaluation
• internal/controllers/machinehealthcheck/machinehealthcheck_targets_test.go: Added unit tests verifying machines need remediation based on machine conditions
• Added test coverage for scenarios where both UnhealthyNodeConditions and UnhealthyMachineConditions are configured to ensure they work together correctly

Core Logic Refactor:
Modified needsRemediation() in internal/controllers/machinehealthcheck/machinehealthcheck_targets.go to:

• Always evaluate machine conditions first, regardless of node state
• Ensure machine conditions are checked in ALL scenarios (node missing, startup timeout, node exists)
• Consistently merge machine and node condition messages across all failure scenarios
• Maintain backward compatibility with existing condition message formats

CEL Validation for UnhealthyMachineConditions:

• Added CEL validation rule to UnhealthyMachineCondition.Type field
• Disallowed condition types: Ready, Available, HealthCheckSucceeded, OwnerRemediated, ExternallyRemediated
• Added envtest-based integration tests in internal/webhooks/test/machinehealthcheck_test.go to verify CEL validation enforces the restriction.

[furkatgofurov7]

/test pull-cluster-api-e2e-main

[sbueringer]

/test pull-cluster-api-e2e-main

[furkatgofurov7]

Quick update on failing main tests:

1. they are failing now, and probably the reason is that the condition matcher seems to be too strict (order/length and dynamic fields like ObservedGeneration/LastTransitionTime) example failure from local run:

--- FAIL: TestMachineHealthCheck_Reconcile (441.16s)

machinehealthcheck_controller_test.go:232:  
Timed out after 5.001s.  
expected  
    &v1beta2.MachineHealthCheckStatus{Conditions:[]v1.Condition{v1.Condition{Type:"RemediationAllowed", Status:"True", ObservedGeneration:1, LastTransitionTime:time.Date(2025, time.October, 14, 0, 40, 34, 0, time.Local), Reason:"RemediationAllowed", Message:""}, v1.Condition{Type:"Paused", Status:"False", ObservedGeneration:1, LastTransitionTime:time.Date(2025, time.October, 14, 0, 40, 34, 0, time.Local), Reason:"NotPaused", Message:""}}, ExpectedMachines:(*int32)(0x14000c9498c), CurrentHealthy:(*int32)(0x14000c94990), RemediationsAllowed:(*int32)(0x14000c94994), ObservedGeneration:1, Targets:[]string{"test-mhc-machine-pm82p", "test-mhc-machine-v74dz"}, Deprecated:(*v1beta2.MachineHealthCheckDeprecatedStatus)(0x140008227d0)}  
to match  
    &v1beta2.MachineHealthCheckStatus{Conditions:[]v1.Condition{v1.Condition{Type:"RemediationAllowed", Status:"True", ObservedGeneration:0, LastTransitionTime:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), Reason:"RemediationAllowed", Message:""}, v1.Condition{Type:"Paused", Status:"False", ObservedGeneration:0, LastTransitionTime:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), Reason:"NotPaused", Message:""}}, ExpectedMachines:(*int32)(0x14001fa10ac), CurrentHealthy:(*int32)(0x14001fa10b0), RemediationsAllowed:(*int32)(0x14001fa10b4), ObservedGeneration:1, Targets:[]string{"test-mhc-machine-pm82p", "test-mhc-machine-v74dz"}, Deprecated:(*v1beta2.MachineHealthCheckDeprecatedStatus)(0x14001d9e6d0)}  

machinehealthcheck_controller_test.go:366:  
Timed out after 30.001s.  
expected  
    &v1beta2.MachineHealthCheckStatus{Conditions:[]v1.Condition{v1.Condition{Type:"RemediationAllowed", Status:"True", ObservedGeneration:1, LastTransitionTime:time.Date(2025, time.October, 14, 0, 40, 40, 0, time.Local), Reason:"RemediationAllowed", Message:""}, v1.Condition{Type:"Paused", Status:"False", ObservedGeneration:1, LastTransitionTime:time.Date(2025, time.October, 14, 0, 40, 40, 0, time.Local), Reason:"NotPaused", Message:""}}, ExpectedMachines:(*int32)(0x14001ebd99c), CurrentHealthy:(*int32)(0x14001ebd9a0), RemediationsAllowed:(*int32)(0x14001ebd9a4), ObservedGeneration:1, Targets:[]string{"test-mhc-machine-9mctz", "test-mhc-machine-jdf2x"}, Deprecated:(*v1beta2.MachineHealthCheckDeprecatedStatus)(0x1400061eb08)}  
to match  
    &v1beta2.MachineHealthCheckStatus{Conditions:[]v1.Condition{v1.Condition{Type:"RemediationAllowed", Status:"True", ObservedGeneration:0, LastTransitionTime:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), Reason:"RemediationAllowed", Message:""}}, ExpectedMachines:(*int32)(0x14001ebc1cc), CurrentHealthy:(*int32)(0x14001ebc1d0), RemediationsAllowed:(*int32)(0x14001ebc1d4), ObservedGeneration:1, Targets:[]string{"test-mhc-machine-9mctz", "test-mhc-machine-jdf2x"}, Deprecated:(*v1beta2.MachineHealthCheckDeprecatedStatus)(0x1400061e240)}

2. aggregated unhealthy messages didn’t consistently include the “Health check failed:” prefix:

--- FAIL: TestHealthCheckTargets (0.00s)

machinehealthcheck_targets_test.go:636:  
Expected  
<[]v1.Condition | len:1, cap:1>:  
[
  {
    Type: "HealthCheckSucceeded",
    Status: "False",
    ObservedGeneration: 0,
    LastTransitionTime: { Time: 0001-01-01T00:00:00Z },
    Reason: "UnhealthyNode",
    Message: "Condition Ready on Node is reporting status Unknown for more than 5m0s",
  },
]  

to contain elements  
<[]v1.Condition | len:1, cap:1>:  
[
  {
    Type: "HealthCheckSucceeded",
    Status: "False",
    ObservedGeneration: 0,
    LastTransitionTime: { Time: 0001-01-01T00:00:00Z },
    Reason: "UnhealthyNode",
    Message: "Health check failed: Condition Ready on Node is reporting status Unknown for more than 5m0s",
  },
]  

the missing elements were  
<[]v1.Condition | len:1, cap:1>:  
[
  {
    Type: "HealthCheckSucceeded",
    Status: "False",
    ObservedGeneration: 0,
    LastTransitionTime: { Time: 0001-01-01T00:00:00Z },
    Reason: "UnhealthyNode",
    Message: "Health check failed: Condition Ready on Node is reporting status Unknown for more than 5m0s",
  },
]

Currently, I am trying to fix this by relaxing the custom matcher to be order-insensitive and subset-based, and standardizing the combined unhealthy message to include the prefix. However, if you have any other suggestions, I am all open for suggestions, thank you.

[sbueringer]

Currently, I am trying to fix this by relaxing the custom matcher to be order-insensitive and subset-based, and standardizing the combined unhealthy message to include the prefix. However, if you have any other suggestions, I am all open for suggestions, thank you.

I think gomega should already not care about the order if we use the right matcher

I also thought we have some matcher that allows ignoring timestamps for condition comparisons (grep for HaveSameStateOf, maybe it's useful)

[fabriziopandini]

I have found the root cause of failures in TestMachineHealthCheck_Reconcile and provided a few suggestion for the computation of the condition (see comments).

Unfortunately I have also found another issue in the needsRemediation func, that probably needs a bigger refactor.

the current logic in needsRemediation sort of assumes that check were only applied to nodes, so e.g. it returns immediately if the node is not showing up at startup, or if the node has been deleted at a later stage

cluster-api/internal/controllers/machinehealthcheck/machinehealthcheck_targets.go

Lines 134 to 183 in f96b742

	if t.Node == nil {
	if timeoutForMachineToHaveNode == disabledNodeStartupTimeout {
	// Startup timeout is disabled so no need to go any further.
	// No node yet to check conditions, can return early here.
	return false, 0
	}
	controlPlaneInitialized := conditions.GetLastTransitionTime(t.Cluster, clusterv1.ClusterControlPlaneInitializedCondition)
	clusterInfraReady := conditions.GetLastTransitionTime(t.Cluster, clusterv1.ClusterInfrastructureReadyCondition)
	machineInfraReady := conditions.GetLastTransitionTime(t.Machine, clusterv1.MachineInfrastructureReadyCondition)
	machineCreationTime := t.Machine.CreationTimestamp.Time
	// Use the latest of the following timestamps.
	comparisonTime := machineCreationTime
	logger.V(5).Info("Determining comparison time",
	"machineCreationTime", machineCreationTime,
	"clusterInfraReadyTime", clusterInfraReady,
	"controlPlaneInitializedTime", controlPlaneInitialized,
	"machineInfraReadyTime", machineInfraReady,
	)
	if conditions.IsTrue(t.Cluster, clusterv1.ClusterControlPlaneInitializedCondition) && controlPlaneInitialized != nil && controlPlaneInitialized.After(comparisonTime) {
	comparisonTime = controlPlaneInitialized.Time
	}
	if conditions.IsTrue(t.Cluster, clusterv1.ClusterInfrastructureReadyCondition) && clusterInfraReady != nil && clusterInfraReady.After(comparisonTime) {
	comparisonTime = clusterInfraReady.Time
	}
	if conditions.IsTrue(t.Machine, clusterv1.MachineInfrastructureReadyCondition) && machineInfraReady != nil && machineInfraReady.After(comparisonTime) {
	comparisonTime = machineInfraReady.Time
	}
	logger.V(5).Info("Using comparison time", "time", comparisonTime)
	timeoutDuration := timeoutForMachineToHaveNode.Duration
	if comparisonTime.Add(timeoutForMachineToHaveNode.Duration).Before(now) {
	v1beta1conditions.MarkFalse(t.Machine, clusterv1.MachineHealthCheckSucceededV1Beta1Condition, clusterv1.NodeStartupTimeoutV1Beta1Reason, clusterv1.ConditionSeverityWarning, "Node failed to report startup in %s", timeoutDuration)
	logger.V(3).Info("Target is unhealthy: machine has no node", "duration", timeoutDuration)
	conditions.Set(t.Machine, metav1.Condition{
	Type: clusterv1.MachineHealthCheckSucceededCondition,
	Status: metav1.ConditionFalse,
	Reason: clusterv1.MachineHealthCheckNodeStartupTimeoutReason,
	Message: fmt.Sprintf("Health check failed: Node failed to report startup in %s", timeoutDuration),
	})
	return true, time.Duration(0)
	}
	durationUnhealthy := now.Sub(comparisonTime)
	nextCheck := timeoutDuration - durationUnhealthy + time.Second
	return false, nextCheck
	}

cluster-api/internal/controllers/machinehealthcheck/machinehealthcheck_targets.go

Lines 105 to 116 in f96b742

	if t.nodeMissing {
	logger.V(3).Info("Target is unhealthy: node is missing")
	v1beta1conditions.MarkFalse(t.Machine, clusterv1.MachineHealthCheckSucceededV1Beta1Condition, clusterv1.NodeNotFoundV1Beta1Reason, clusterv1.ConditionSeverityWarning, "")
	conditions.Set(t.Machine, metav1.Condition{
	Type: clusterv1.MachineHealthCheckSucceededCondition,
	Status: metav1.ConditionFalse,
	Reason: clusterv1.MachineHealthCheckNodeDeletedReason,
	Message: fmt.Sprintf("Health check failed: Node %s has been deleted", t.Machine.Status.NodeRef.Name),
	})
	return true, time.Duration(0)
	}

While this code structure worked well when checking only node conditions at the end of the func, it does not work well with the addition of the check for machine conditions at the end of the func.

More specifically, I think that we should find a way to always check machine conditions, not only in case the node is existing / the func doesn't hit the two if branches highlighted above, as in the current implementation.

Additionally, we should always consider that we merge messages from machine conditions and from node conditions in all the possible scenarios:

• when node is not showing up at startup
• when the node has been deleted at a later stage
• when the node exists (which is the only scenario covered in the current change set)

I will try to come up with a some ideas to solve this problem, but of course suggestions are more than welcome

[furkatgofurov7]

/test pull-cluster-api-test-main

[furkatgofurov7]

I have found the root cause of failures in TestMachineHealthCheck_Reconcile and provided a few suggestion for the computation of the condition (see comments).

Unfortunately I have also found another issue in the needsRemediation func, that probably needs a bigger refactor.

the current logic in needsRemediation sort of assumes that check were only applied to nodes, so e.g. it returns immediately if the node is not showing up at startup, or if the node has been deleted at a later stage

cluster-api/internal/controllers/machinehealthcheck/machinehealthcheck_targets.go

Lines 134 to 183 in f96b742

	if t.Node == nil {
	if timeoutForMachineToHaveNode == disabledNodeStartupTimeout {
	// Startup timeout is disabled so no need to go any further.
	// No node yet to check conditions, can return early here.
	return false, 0
	}
	controlPlaneInitialized := conditions.GetLastTransitionTime(t.Cluster, clusterv1.ClusterControlPlaneInitializedCondition)
	clusterInfraReady := conditions.GetLastTransitionTime(t.Cluster, clusterv1.ClusterInfrastructureReadyCondition)
	machineInfraReady := conditions.GetLastTransitionTime(t.Machine, clusterv1.MachineInfrastructureReadyCondition)
	machineCreationTime := t.Machine.CreationTimestamp.Time
	// Use the latest of the following timestamps.
	comparisonTime := machineCreationTime
	logger.V(5).Info("Determining comparison time",
	"machineCreationTime", machineCreationTime,
	"clusterInfraReadyTime", clusterInfraReady,
	"controlPlaneInitializedTime", controlPlaneInitialized,
	"machineInfraReadyTime", machineInfraReady,
	)
	if conditions.IsTrue(t.Cluster, clusterv1.ClusterControlPlaneInitializedCondition) && controlPlaneInitialized != nil && controlPlaneInitialized.After(comparisonTime) {
	comparisonTime = controlPlaneInitialized.Time
	}
	if conditions.IsTrue(t.Cluster, clusterv1.ClusterInfrastructureReadyCondition) && clusterInfraReady != nil && clusterInfraReady.After(comparisonTime) {
	comparisonTime = clusterInfraReady.Time
	}
	if conditions.IsTrue(t.Machine, clusterv1.MachineInfrastructureReadyCondition) && machineInfraReady != nil && machineInfraReady.After(comparisonTime) {
	comparisonTime = machineInfraReady.Time
	}
	logger.V(5).Info("Using comparison time", "time", comparisonTime)
	timeoutDuration := timeoutForMachineToHaveNode.Duration
	if comparisonTime.Add(timeoutForMachineToHaveNode.Duration).Before(now) {
	v1beta1conditions.MarkFalse(t.Machine, clusterv1.MachineHealthCheckSucceededV1Beta1Condition, clusterv1.NodeStartupTimeoutV1Beta1Reason, clusterv1.ConditionSeverityWarning, "Node failed to report startup in %s", timeoutDuration)
	logger.V(3).Info("Target is unhealthy: machine has no node", "duration", timeoutDuration)
	conditions.Set(t.Machine, metav1.Condition{
	Type: clusterv1.MachineHealthCheckSucceededCondition,
	Status: metav1.ConditionFalse,
	Reason: clusterv1.MachineHealthCheckNodeStartupTimeoutReason,
	Message: fmt.Sprintf("Health check failed: Node failed to report startup in %s", timeoutDuration),
	})
	return true, time.Duration(0)
	}
	durationUnhealthy := now.Sub(comparisonTime)
	nextCheck := timeoutDuration - durationUnhealthy + time.Second
	return false, nextCheck
	}

cluster-api/internal/controllers/machinehealthcheck/machinehealthcheck_targets.go

Lines 105 to 116 in f96b742

	if t.nodeMissing {
	logger.V(3).Info("Target is unhealthy: node is missing")
	v1beta1conditions.MarkFalse(t.Machine, clusterv1.MachineHealthCheckSucceededV1Beta1Condition, clusterv1.NodeNotFoundV1Beta1Reason, clusterv1.ConditionSeverityWarning, "")
	conditions.Set(t.Machine, metav1.Condition{
	Type: clusterv1.MachineHealthCheckSucceededCondition,
	Status: metav1.ConditionFalse,
	Reason: clusterv1.MachineHealthCheckNodeDeletedReason,
	Message: fmt.Sprintf("Health check failed: Node %s has been deleted", t.Machine.Status.NodeRef.Name),
	})
	return true, time.Duration(0)
	}

While this code structure worked well when checking only node conditions at the end of the func, it does not work well with the addition of the check for machine conditions at the end of the func.

More specifically, I think that we should find a way to always check machine conditions, not only in case the node is existing / the func doesn't hit the two if branches highlighted above, as in the current implementation.

Additionally, we should always consider that we merge messages from machine conditions and from node conditions in all the possible scenarios:

• when node is not showing up at startup
• when the node has been deleted at a later stage
• when the node exists (which is the only scenario covered in the current change set)

I will try to come up with a some ideas to solve this problem, but of course suggestions are more than welcome

@fabriziopandini Thanks for the detailed feedback! You're absolutely right about the inconsistent behavior. I've refactored now the needsRemediation function to address all the concerns you raised.

Changes I made:

1. We now always evaluate Machine Conditions first
   Machine conditions are now evaluated before any node-related logic, to ensure they're checked in ALL scenarios:

• When node is missing t.nodeMissing
• When node hasn't appeared yet t.Node == nil
• When node exists t.Node != nil

2. We should have a consistent message merging in ALL scenarios
   Error messages from both machine and node conditions are now properly merged in every failure scenario:

• Node Missing: "Node X has been deleted; Condition Y on Machine is reporting status Z"
• Node Startup Timeout: "Node failed to report startup in 5m; Condition Y on Machine is reporting status Z"
• Node Exists: "Condition A on Node is reporting status B; Condition Y on Machine is reporting status Z"

3. No More Early Returns Bypassing Machine Evaluation
   The problematic early return paths now occur after machine conditions are evaluated, not before.

Let me know what you think about the refactor.

[furkatgofurov7]

@sbueringer, thanks for another round of feedback on conversion; hopefully all your suggestions should be incorporated now.

[furkatgofurov7]

/test pull-cluster-api-e2e-main

[fabriziopandini]

Thanks @furkatgofurov7 for this iteration!

I'm wondering if we can further simplify the code/improve readability by using two sub functions, one for machineChecks and the other for nodeChecks.

The resulting needsRemediation will look like:

func (t *healthCheckTarget) needsRemediation(logger logr.Logger, timeoutForMachineToHaveNode metav1.Duration) (bool, time.Duration) {
	// checks for HasRemediateMachineAnnotation, ClusterControlPlaneInitializedCondition, ClusterInfrastructureReadyCondition
        ...

	// Check machine conditions
	unhealthyMachineMessages, nextMachineCheck := t.machineChecks(logger)

	// Check node conditions
	nodeConditionReason, nodeV1beta1ConditionReason, unhealthyNodeMessages, nextNodeCheck := t.nodeChecks(logger, timeoutForMachineToHaveNode)

	// Combine results and set conditions
	...
}

Another benefit of this code struct, is that condition management is implemented only in one place.

In case it can help, this is a commit where I experimented a little bit about this idea

wdyt?

[furkatgofurov7]

/test pull-cluster-api-e2e-main

[furkatgofurov7]

Looks like I am hitting #12334 in e2e tests here?

capi-e2e: [It] When following the Cluster API quick-start with v1beta1 ClusterClass [ClusterClass] Should create a workload cluster [ClusterClass] expand_less | 3m54s
-- | --
{Failed after 13.722s. resourceVersions didn't stay stable The function passed to Consistently failed at /home/prow/go/src/sigs.k8s.io/cluster-api/test/framework/resourceversion_helpers.go:59 with:  Detected objects with changed resourceVersion  Object with changed resourceVersion KubeadmControlPlane/quick-start-i8nppc/quick-start-v5eonr-cp-gr9xz:   strings.Join({   	... // 6037 identical bytes   	"Replicas: {}\n        f:version: {}\n    manager: manager\n    oper",   	"ation: Update\n    subresource: status\n    time: \"2025-10-27T11:1", - 	"5:24", + 	"7:17",   	"Z\"\n  name: quick-start-v5eonr-cp-gr9xz\n  namespace: quick-start-",   	"i8nppc\n  ownerReferences:\n  - apiVersion: cluster.x-k8s.io/v1bet",   	"a2\n    blockOwnerDeletion: true\n    controller: true\n    kind: C",   	"luster\n    name: quick-start-v5eonr\n    uid: dc1be33f-0f0d-494a-",   	"a05c-49d99e2e956b\n  resourceVersion: \"2", - 	"6533", + 	"7270",   	"\"\n  uid: 366545c5-f6a2-441e-87ad-fd56f479464e\nspec:\n  kubeadmCon",   	"figSpec:\n    clusterConfiguration:\n      apiServer:\n        cert",   	... // 1156 identical bytes   	"    type: RollingUpdate\n  version: v1.34.0\nstatus:\n  availableRe",   	"plicas: 1\n  conditions:\n  - lastTransitionTime: \"2025-10-27T11:1", - 	"5:24", + 	"7:16",   	"Z\"\n    message: \"\"\n    observedGeneration: 1\n    reason: Availab",   	"le\n    status: \"True\"\n    type: Available\n  - lastTransitionTime",   	... // 206 identical bytes   	" observedGeneration: 1\n    reason: Initialized\n    status: \"True",   	"\"\n    type: Initialized\n  - lastTransitionTime: \"2025-10-27T11:1", - 	"5:24", + 	"7:16",   	"Z\"\n    message: \"\"\n    observedGeneration: 1\n    reason: Healthy",   	"\n    status: \"True\"\n    type: EtcdClusterHealthy\n  - lastTransit",   	... // 691 identical bytes   	" observedGeneration: 1\n    reason: NotScalingUp\n    status: \"Fal",   	"se\"\n    type: ScalingUp\n  - lastTransitionTime: \"2025-10-27T11:1", - 	"5:24", + 	"7:16",   	"Z\"\n    message: \"\"\n    observedGeneration: 1\n    reason: Ready\n ",   	"   status: \"True\"\n    type: MachinesReady\n  - lastTransitionTime",   	... // 815 identical bytes   	"10-27T11:15:08Z\"\n        status: \"True\"\n        type: ControlPla",   	"neComponentsHealthy\n      - lastTransitionTime: \"2025-10-27T11:1", - 	"5:24", + 	"7:16",   	"Z\"\n        status: \"True\"\n        type: EtcdClusterHealthy\n     ",   	" - lastTransitionTime: \"2025-10-27T11:13:56Z\"\n        status: \"T",   	... // 552 identical bytes   }, "")

......

[furkatgofurov7]

/test pull-cluster-api-e2e-main

[furkatgofurov7]

/test pull-cluster-api-e2e-main

[sbueringer]

Just a few minor findings

[sbueringer]

Thank you very much!!!

/lgtm
/assign @fabriziopandini

[k8s-ci-robot]

LGTM label has been added.

Git tree hash: 4ead6b65f629a47eeda3ee75246b77679a6ddc5a

[sbueringer]

/test pull-cluster-api-e2e-main

[fabriziopandini]

Nice!
/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: fabriziopandini

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [fabriziopandini]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment