[release 4.6] Bug 1953582: GatherClusterOperators and GatherClusterOp…

…eratorsPodAndEvents (#413)

* Bug 1951034: Split up the GatherClusterOperators into smaller parts (#397)

* Split operators gather

This splits the cluster operators gather into two gathers, one for operator resources and other for unhealty operators.

* Adds precommit target

This adds the precommit target to the makefile, so we can execute it to test the stashed changes before actually commit.

* Move CompactedEvent

This move the structure that was previously defined on operators to operators_unhealthy.

* Change operators name

This changes the previous operator_unhealty to operator_pods_and_events.

* Update gathered documentation

This updates the gatherer documentation for operators and operators_pods_and_events

* Skip Test_UnhealtyOperators_FetchPodContainerLog

This removes and skips with note the Test_UnhealtyOperators_FetchPodContainerLog test

docs/gathered-data.md

@@ -117,10 +117,25 @@ Location in archive: config/oauth/
 See: docs/insights-archive-sample/config/oauth
 
 
+## ClusterOperatorPodsAndEvents
+
+collects all the ClusterOperators degraded Pods
+for degraded cluster operators or that lives at the Cluster Operator's namespace, to collect:
+
+- Pod definitions
+- Previous and current Pod Container logs (when available)
+- Namespace Events
+
+* Location of pods in archive: config/pod/
+* Location of events in archive: events/
+* Id in config: operators_pods_and_events
+
+
 ## ClusterOperators
 
 collects all ClusterOperators and their resource.
 It finds unhealthy Pods for unhealthy operators
+GatherClusterOperators collects all the ClusterOperators definitions and their resources.
 
 The Kubernetes api https://github.com/openshift/client-go/blob/master/config/clientset/versioned/typed/config/v1/clusteroperator.go#L62
 Response see https://docs.openshift.com/container-platform/4.3/rest_api/index.html#clusteroperatorlist-v1config-openshift-io

pkg/gather/clusterconfig/0_gatherer.go

@@ -37,6 +37,7 @@ func (g *Gatherer) Gather(ctx context.Context, recorder record.Interface) error
 		GatherPodDisruptionBudgets(g),
 		GatherMostRecentMetrics(g),
 		GatherClusterOperators(g),
+		GatherClusterOperatorPodsAndEvents(g),
 		GatherContainerImages(g),
 		GatherNodes(g),
 		GatherConfigMaps(g),

pkg/gather/clusterconfig/0_utils.go

@@ -15,7 +15,6 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/runtime/serializer"
-	"k8s.io/apimachinery/pkg/util/json"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
 	kubescheme "k8s.io/client-go/kubernetes/scheme"
 
@@ -168,32 +167,6 @@ func (a Anonymizer) GetExtension() string {
 	return "json"
 }
 
-// CompactedEvent holds one Namespace Event
-type CompactedEvent struct {
-	Namespace     string    `json:"namespace"`
-	LastTimestamp time.Time `json:"lastTimestamp"`
-	Reason        string    `json:"reason"`
-	Message       string    `json:"message"`
-}
-
-// CompactedEventList is collection of events
-type CompactedEventList struct {
-	Items []CompactedEvent `json:"items"`
-}
-
-// EventAnonymizer implements serializaion of Events with anonymization
-type EventAnonymizer struct{ *CompactedEventList }
-
-// Marshal serializes Events with anonymization
-func (a EventAnonymizer) Marshal(_ context.Context) ([]byte, error) {
-	return json.Marshal(a.CompactedEventList)
-}
-
-// GetExtension returns extension for anonymized event objects
-func (a EventAnonymizer) GetExtension() string {
-	return "json"
-}
-
 func anonymizeURLCSV(s string) string {
 	strs := strings.Split(s, ",")
 	outSlice := anonymizeURLSlice(strs)
@@ -243,40 +216,6 @@ func anonymizePod(pod *corev1.Pod) *corev1.Pod {
 	return pod
 }
 
-func isHealthyPod(pod *corev1.Pod, now time.Time) bool {
-	// pending pods may be unable to schedule or start due to failures, and the info they provide in status is important
-	// for identifying why scheduling has not happened
-	if pod.Status.Phase == corev1.PodPending {
-		if now.Sub(pod.CreationTimestamp.Time) > 2*time.Minute {
-			return false
-		}
-	}
-	// pods that have containers that have terminated with non-zero exit codes are considered failure
-	for _, status := range pod.Status.InitContainerStatuses {
-		if status.LastTerminationState.Terminated != nil && status.LastTerminationState.Terminated.ExitCode != 0 {
-			return false
-		}
-		if status.State.Terminated != nil && status.State.Terminated.ExitCode != 0 {
-			return false
-		}
-		if status.RestartCount > 0 {
-			return false
-		}
-	}
-	for _, status := range pod.Status.ContainerStatuses {
-		if status.LastTerminationState.Terminated != nil && status.LastTerminationState.Terminated.ExitCode != 0 {
-			return false
-		}
-		if status.State.Terminated != nil && status.State.Terminated.ExitCode != 0 {
-			return false
-		}
-		if status.RestartCount > 0 {
-			return false
-		}
-	}
-	return true
-}
-
 // MinimalNodeInfo contains the most essential information about a node
 type MinimalNodeInfo struct {
 	ProviderID string `json:"providerID"`