Merge branch 'master' into deployment-phase

Author: ewoutp

pkg/apis/deployment/v1alpha/conditions.go

@@ -37,6 +37,8 @@ const (
 	ConditionTypeTerminated ConditionType = "Terminated"
 	// ConditionTypeAutoUpgrade indicates that the member has to be started with `--database.auto-upgrade` once.
 	ConditionTypeAutoUpgrade ConditionType = "AutoUpgrade"
+	// ConditionTypePodSchedulingFailure indicates that one or more pods belonging to the deployment cannot be schedule.
+	ConditionTypePodSchedulingFailure ConditionType = "PodSchedulingFailure"
 	// ConditionTypeSecretsChanged indicates that the value of one of more secrets used by
 	// the deployment have changed. Once that is the case, the operator will no longer
 	// touch the deployment, until the original secrets have been restored.

pkg/deployment/resources/pod_inspector.go

@@ -23,6 +23,9 @@
 package resources
 
 import (
+	"fmt"
+	"time"
+
 	"github.com/arangodb/kube-arangodb/pkg/util/k8sutil"
 	"k8s.io/api/core/v1"
 
@@ -34,6 +37,10 @@ var (
 	inspectedPodCounter = metrics.MustRegisterCounter("deployment", "inspected_pods", "Number of pod inspections")
 )
 
+const (
+	podScheduleTimeout = time.Minute // How long we allow the schedule to take scheduling a pod.
+)
+
 // InspectPods lists all pods that belong to the given deployment and updates
 // the member status of the deployment accordingly.
 func (r *Resources) InspectPods() error {
@@ -49,6 +56,8 @@ func (r *Resources) InspectPods() error {
 	// Update member status from all pods found
 	status := r.context.GetStatus()
 	apiObject := r.context.GetAPIObject()
+	var podNamesWithScheduleTimeout []string
+	var unscheduledPodNames []string
 	for _, p := range pods {
 		if k8sutil.IsArangoDBImageIDAndVersionPod(p) {
 			// Image ID pods are not relevant to inspect here
@@ -93,6 +102,13 @@ func (r *Resources) InspectPods() error {
 				updateMemberStatusNeeded = true
 			}
 		}
+		if k8sutil.IsPodNotScheduledFor(&p, podScheduleTimeout) {
+			// Pod cannot be scheduled for to long
+			log.Debug().Str("pod-name", p.GetName()).Msg("Pod scheduling timeout")
+			podNamesWithScheduleTimeout = append(podNamesWithScheduleTimeout, p.GetName())
+		} else if !k8sutil.IsPodScheduled(&p) {
+			unscheduledPodNames = append(unscheduledPodNames, p.GetName())
+		}
 		if updateMemberStatusNeeded {
 			if err := status.Members.UpdateMemberStatus(memberStatus, group); err != nil {
 				return maskAny(err)
@@ -146,6 +162,22 @@ func (r *Resources) InspectPods() error {
 	allMembersReady := status.Members.AllMembersReady()
 	status.Conditions.Update(api.ConditionTypeReady, allMembersReady, "", "")
 
+	// Update conditions
+	if len(podNamesWithScheduleTimeout) > 0 {
+		if status.Conditions.Update(api.ConditionTypePodSchedulingFailure, true,
+			"Pods Scheduling Timeout",
+			fmt.Sprintf("The following pods cannot be scheduled: %v", podNamesWithScheduleTimeout)) {
+			r.context.CreateEvent(k8sutil.NewPodsSchedulingFailureEvent(podNamesWithScheduleTimeout, r.context.GetAPIObject()))
+		}
+	} else if status.Conditions.IsTrue(api.ConditionTypePodSchedulingFailure) &&
+		len(unscheduledPodNames) == 0 {
+		if status.Conditions.Update(api.ConditionTypePodSchedulingFailure, false,
+			"Pods Scheduling Resolved",
+			"No pod reports a scheduling timeout") {
+			r.context.CreateEvent(k8sutil.NewPodsSchedulingResolvedEvent(r.context.GetAPIObject()))
+		}
+	}
+
 	// Save status
 	if err := r.context.UpdateStatus(status); err != nil {
 		return maskAny(err)

pkg/util/k8sutil/events.go

@@ -78,6 +78,25 @@ func NewImmutableFieldEvent(fieldName string, apiObject APIObject) *v1.Event {
 	return event
 }
 
+// NewPodsSchedulingFailureEvent creates an event indicating that one of more cannot be scheduled.
+func NewPodsSchedulingFailureEvent(unscheduledPodNames []string, apiObject APIObject) *v1.Event {
+	event := newDeploymentEvent(apiObject)
+	event.Type = v1.EventTypeNormal
+	event.Reason = "Pods Scheduling Failure"
+	event.Message = fmt.Sprintf("One or more pods are not scheduled in time. Pods: %v", unscheduledPodNames)
+	return event
+}
+
+// NewPodsSchedulingResolvedEvent creates an event indicating that an earlier problem with
+// pod scheduling has been resolved.
+func NewPodsSchedulingResolvedEvent(apiObject APIObject) *v1.Event {
+	event := newDeploymentEvent(apiObject)
+	event.Type = v1.EventTypeNormal
+	event.Reason = "Pods Scheduling Resolved"
+	event.Message = "All pods have been scheduled"
+	return event
+}
+
 // NewSecretsChangedEvent creates an event indicating that one of more secrets have changed.
 func NewSecretsChangedEvent(changedSecretNames []string, apiObject APIObject) *v1.Event {
 	event := newDeploymentEvent(apiObject)

pkg/util/k8sutil/pods.go

@@ -25,6 +25,7 @@ package k8sutil
 import (
 	"fmt"
 	"path/filepath"
+	"time"
 
 	"k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -87,6 +88,21 @@ func IsPodFailed(pod *v1.Pod) bool {
 	return pod.Status.Phase == v1.PodFailed
 }
 
+// IsPodScheduled returns true if the pod has been scheduled.
+func IsPodScheduled(pod *v1.Pod) bool {
+	condition := getPodCondition(&pod.Status, v1.PodScheduled)
+	return condition != nil && condition.Status == v1.ConditionTrue
+}
+
+// IsPodNotScheduledFor returns true if the pod has not been scheduled
+// for longer than the given duration.
+func IsPodNotScheduledFor(pod *v1.Pod, timeout time.Duration) bool {
+	condition := getPodCondition(&pod.Status, v1.PodScheduled)
+	return condition != nil &&
+		condition.Status == v1.ConditionFalse &&
+		condition.LastTransitionTime.Time.Add(timeout).Before(time.Now())
+}
+
 // IsArangoDBImageIDAndVersionPod returns true if the given pod is used for fetching image ID and ArangoDB version of an image
 func IsArangoDBImageIDAndVersionPod(p v1.Pod) bool {
 	role, found := p.GetLabels()[LabelKeyRole]