From b285e27ffa630caa091302335c923ededa4e0e7f Mon Sep 17 00:00:00 2001
From: Maciej Pytel <maciekpytel@google.com>
Date: Fri, 24 Mar 2017 11:45:25 +0100
Subject: [PATCH] Fix flaky cluster-autoscaler e2e

---
 test/e2e/cluster_size_autoscaling.go | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/test/e2e/cluster_size_autoscaling.go b/test/e2e/cluster_size_autoscaling.go
index 23bea8f21d0..8b34a80b0ae 100644
--- a/test/e2e/cluster_size_autoscaling.go
+++ b/test/e2e/cluster_size_autoscaling.go
@@ -26,6 +26,7 @@ import (
 	"strings"
 	"time"
 
+	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/fields"
 	"k8s.io/apimachinery/pkg/util/intstr"
@@ -740,11 +741,25 @@ func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, id stri
 		if err != nil {
 			return err
 		}
-		*rc.Spec.Replicas = int32(i + 1)
-		rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Update(rc)
-		if err != nil {
-			return err
+
+		// Update replicas count, to create new pods that will be allocated on node
+		// (we retry 409 errors in case rc reference got out of sync)
+		for j := 0; j < 3; j++ {
+			*rc.Spec.Replicas = int32(i + 1)
+			rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Update(rc)
+			if err == nil {
+				break
+			}
+			if !errors.IsConflict(err) {
+				return err
+			}
+			glog.Warningf("Got 409 conflict when trying to scale RC, retries left: %v", 3-j)
+			rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Get(id, metav1.GetOptions{})
+			if err != nil {
+				return err
+			}
 		}
+
 		err = wait.PollImmediate(5*time.Second, podTimeout, func() (bool, error) {
 			rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Get(id, metav1.GetOptions{})
 			if err != nil || rc.Status.ReadyReplicas < int32(i+1) {