Merge pull request #342 from arangodb/bug-fix/remove-server

Try harder to remove server from cluster.

Author: neunhoef

pkg/deployment/reconcile/action_remove_member.go

@@ -70,6 +70,22 @@ func (a *actionRemoveMember) Start(ctx context.Context) (bool, error) {
 		if err := arangod.RemoveServerFromCluster(ctx, client.Connection(), driver.ServerID(m.ID)); err != nil {
 			if !driver.IsNotFound(err) && !driver.IsPreconditionFailed(err) {
 				return false, maskAny(errors.Wrapf(err, "Failed to remove server from cluster: %#v", err))
+			} else if driver.IsPreconditionFailed(err) {
+				cluster, err := client.Cluster(ctx)
+				if err != nil {
+					return false, maskAny(errors.Wrapf(err, "Failed to obtain cluster: %#v", err))
+				}
+				health, err := cluster.Health(ctx)
+				if err != nil {
+					return false, maskAny(errors.Wrapf(err, "Failed to obtain cluster health: %#v", err))
+				}
+				// We don't care if not found
+				if record, ok := health.Health[driver.ServerID(m.ID)]; ok {
+					if record.Status != driver.ServerStatusFailed {
+						return false, maskAny(fmt.Errorf("can not remove server from cluster. Not yet terminated. Retry later"))
+					}
+					a.log.Warn().Msg("dbserver is failed but still in use")
+				}
 			} else {
 				a.log.Warn().Msgf("ignoring error: %s", err.Error())
 			}

pkg/deployment/resources/pod_termination.go

@@ -123,6 +123,12 @@ func (r *Resources) prepareDBServerPodTermination(ctx context.Context, log zerol
 		log.Debug().Msg("Pod is already failed, safe to remove dbserver pod")
 		return nil
 	}
+	// If pod is not member of cluster, do nothing
+	if !memberStatus.Conditions.IsTrue(api.ConditionTypeMemberOfCluster) {
+		log.Debug().Msg("Pod is not member of cluster")
+		return nil
+	}
+
 	// Inspect deployment deletion state
 	apiObject := r.context.GetAPIObject()
 	if apiObject.GetDeletionTimestamp() != nil {
@@ -154,6 +160,11 @@ func (r *Resources) prepareDBServerPodTermination(ctx context.Context, log zerol
 		dbserverDataWillBeGone = true
 	}
 
+	// Once decided to drain the member, never go back
+	if memberStatus.Phase == api.MemberPhaseCreated {
+		dbserverDataWillBeGone = true
+	}
+
 	// Is this a simple pod restart?
 	if !dbserverDataWillBeGone {
 		log.Debug().Msg("Pod is just being restarted, safe to remove dbserver pod")
@@ -230,11 +241,15 @@ func (r *Resources) prepareDBServerPodTermination(ctx context.Context, log zerol
 			return maskAny(err)
 		}
 		if jobStatus.IsFailed() {
-			log.Warn().Str("reason", jobStatus.Reason()).Msg("Cleanout Job failed. Aborting plan")
+			log.Warn().Str("reason", jobStatus.Reason()).Msg("Cleanout Job failed")
 			// Revert cleanout state
 			memberStatus.Phase = api.MemberPhaseCreated
 			memberStatus.CleanoutJobID = ""
-			return maskAny(fmt.Errorf("Clean out server job failed"))
+			if err := updateMember(memberStatus); err != nil {
+				return maskAny(err)
+			}
+			log.Error().Msg("Cleanout server job failed, continue anyway")
+			return nil
 		}
 	}