Schedule only one CleanOutServer job. Introduced new Drain phase.

Author: lamai93

pkg/apis/deployment/v1alpha/member_phase.go

@@ -34,6 +34,8 @@ const (
 	MemberPhaseFailed MemberPhase = "Failed"
 	// MemberPhaseCleanOut indicates that a dbserver is in the process of being cleaned out
 	MemberPhaseCleanOut MemberPhase = "CleanOut"
+	// MemberPhaseDrain indicates that a dbserver is n the process of being cleaned out as result of draining a node
+	MemberPhaseDrain MemberPhase = "Drain"
 	// MemberPhaseShuttingDown indicates that a member is shutting down
 	MemberPhaseShuttingDown MemberPhase = "ShuttingDown"
 	// MemberPhaseRotating indicates that a member is being rotated

pkg/deployment/reconcile/plan_builder.go

@@ -120,7 +120,7 @@ func createPlan(log zerolog.Logger, apiObject k8sutil.APIObject,
 
 	// Check for cleaned out dbserver in created state
 	for _, m := range status.Members.DBServers {
-		if len(plan) == 0 && m.Phase == api.MemberPhaseCreated && m.Conditions.IsTrue(api.ConditionTypeCleanedOut) {
+		if len(plan) == 0 && (m.Phase == api.MemberPhaseCreated || m.Phase == api.MemberPhaseDrain) && m.Conditions.IsTrue(api.ConditionTypeCleanedOut) {
 			log.Debug().
 				Str("id", m.ID).
 				Str("role", api.ServerGroupDBServers.AsRole()).

pkg/deployment/resources/context.go

@@ -26,6 +26,7 @@ import (
 	"context"
 
 	driver "github.com/arangodb/go-driver"
+	"github.com/arangodb/go-driver/agency"
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
 	"github.com/arangodb/kube-arangodb/pkg/util/k8sutil"
 	"k8s.io/api/core/v1"
@@ -85,4 +86,6 @@ type Context interface {
 	// GetDatabaseClient returns a cached client for the entire database (cluster coordinators or single server),
 	// creating one if needed.
 	GetDatabaseClient(ctx context.Context) (driver.Client, error)
+	// GetAgency returns a connection to the entire agency.
+	GetAgency(ctx context.Context) (agency.Agency, error)
 }

pkg/deployment/resources/pod_inspector.go

@@ -182,9 +182,11 @@ func (r *Resources) InspectPods(ctx context.Context) (util.Interval, error) {
 		for _, m := range members {
 			if podName := m.PodName; podName != "" {
 				if !podExists(podName) {
+					log.Debug().Str("pod-name", podName).Msg("Does not exist")
 					switch m.Phase {
 					case api.MemberPhaseNone:
 						// Do nothing
+						log.Debug().Str("pod-name", podName).Msg("PodPhase is None, waiting for the pod to be recreated")
 					case api.MemberPhaseShuttingDown, api.MemberPhaseRotating, api.MemberPhaseUpgrading, api.MemberPhaseFailed:
 						// Shutdown was intended, so not need to do anything here.
 						// Just mark terminated

pkg/deployment/resources/pod_termination.go

@@ -31,8 +31,10 @@ import (
 	"k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
+	driver "github.com/arangodb/go-driver"
 	"github.com/arangodb/go-driver/agency"
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1alpha"
+	"github.com/arangodb/kube-arangodb/pkg/util/arangod"
 	"github.com/arangodb/kube-arangodb/pkg/util/k8sutil"
 )
 
@@ -185,15 +187,49 @@ func (r *Resources) prepareDBServerPodTermination(ctx context.Context, log zerol
 		return nil
 	}
 	// Not cleaned out yet, check member status
-	if memberStatus.Conditions.IsTrue(api.ConditionTypeTerminated) {
+	if memberStatus.Conditions.IsTrue(api.ConditionTypeTerminated) || (memberStatus.Phase == api.MemberPhaseCreated && !k8sutil.IsPodReady(p)) {
 		log.Warn().Msg("Member is already terminated before it could be cleaned out. Not good, but removing dbserver pod because we cannot do anything further")
 		return nil
 	}
-	// Ensure the cleanout is triggered
-	log.Debug().Msg("Server is not yet clean out. Triggering a clean out now")
-	if err := cluster.CleanOutServer(ctx, memberStatus.ID); err != nil {
-		log.Debug().Err(err).Msg("Failed to clean out server")
-		return maskAny(err)
+	if memberStatus.Phase == api.MemberPhaseCreated {
+		// No cleanout job triggered
+		var jobID string
+		ctx = driver.WithJobIDResponse(ctx, &jobID)
+		// Ensure the cleanout is triggered
+		log.Debug().Msg("Server is not yet clean out. Triggering a clean out now")
+		if err := cluster.CleanOutServer(ctx, memberStatus.ID); err != nil {
+			log.Debug().Err(err).Msg("Failed to clean out server")
+			return maskAny(err)
+		}
+		memberStatus.CleanoutJobID = jobID
+		memberStatus.Phase = api.MemberPhaseDrain
+		if err := updateMember(memberStatus); err != nil {
+			return maskAny(err)
+		}
+	} else if memberStatus.Phase == api.MemberPhaseDrain {
+		// Check the job progress
+		agency, err := r.context.GetAgency(ctx)
+		if err != nil {
+			log.Debug().Err(err).Msg("Failed to create agency client")
+			return maskAny(err)
+		}
+		jobStatus, err := arangod.CleanoutServerJobStatus(ctx, memberStatus.CleanoutJobID, c, agency)
+		if err != nil {
+			log.Debug().Err(err).Msg("Failed to fetch cleanout job status")
+			return maskAny(err)
+		}
+		if jobStatus.IsFailed() {
+			log.Warn().Str("reason", jobStatus.Reason()).Msg("Cleanout Job failed. Aborting plan")
+			// Revert cleanout state
+			memberStatus.Phase = api.MemberPhaseCreated
+			memberStatus.CleanoutJobID = ""
+			return maskAny(fmt.Errorf("Clean out server job failed"))
+		}
+	} else {
+		log.Warn().Msgf("Unexpected MemberPhase %s, allow removal", memberStatus.Phase)
+		return nil
 	}
+
 	return maskAny(fmt.Errorf("Server is not yet cleaned out"))
+
 }