From 84fae81322c189e90e5e4fb4e78479cf05147c0b Mon Sep 17 00:00:00 2001
From: Alexander Apalikov <alexander.apalikov@globant.com>
Date: Sun, 22 Sep 2019 20:11:53 +0300
Subject: [PATCH] Marking Evicted Pods as Unhealthy (#1056)

Add aditional health check for Gameservers Controller.
---
 pkg/gameservers/health.go   |  8 +++++++-
 test/e2e/gameserver_test.go | 19 +++++++++++++++++++
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/pkg/gameservers/health.go b/pkg/gameservers/health.go
index 45af430e4e..ef8aef5947 100644
--- a/pkg/gameservers/health.go
+++ b/pkg/gameservers/health.go
@@ -104,7 +104,7 @@ func NewHealthController(health healthcheck.Handler,
 // isUnhealthy returns if the Pod event is going
 // to cause the GameServer to become Unhealthy
 func (hc *HealthController) isUnhealthy(pod *corev1.Pod) bool {
-	return hc.unschedulableWithNoFreePorts(pod) || hc.failedContainer(pod)
+	return hc.evictedPod(pod) || hc.unschedulableWithNoFreePorts(pod) || hc.failedContainer(pod)
 }
 
 // unschedulableWithNoFreePorts checks if the reason the Pod couldn't be scheduled
@@ -120,6 +120,12 @@ func (hc *HealthController) unschedulableWithNoFreePorts(pod *corev1.Pod) bool {
 	return false
 }
 
+// evictedPod checks if the Pod was Evicted
+// could be caused by reaching limit on Ephemeral storage
+func (hc *HealthController) evictedPod(pod *corev1.Pod) bool {
+	return pod.Status.Reason == "Evicted"
+}
+
 // failedContainer checks each container, and determines if there was a failed
 // container
 func (hc *HealthController) failedContainer(pod *corev1.Pod) bool {
diff --git a/test/e2e/gameserver_test.go b/test/e2e/gameserver_test.go
index 15033aa94e..a9f77ee506 100644
--- a/test/e2e/gameserver_test.go
+++ b/test/e2e/gameserver_test.go
@@ -15,6 +15,7 @@
 package e2e
 
 import (
+	"fmt"
 	"testing"
 	"time"
 
@@ -368,6 +369,24 @@ func TestGameServerShutdown(t *testing.T) {
 	assert.NoError(t, err)
 }
 
+// TestGameServerEvicted test that if Gameserver would be evicted than it becomes Unhealthy
+// Ephemeral Storage limit set to 0Mi
+func TestGameServerEvicted(t *testing.T) {
+	t.Parallel()
+	gs := defaultGameServer()
+	gs.Spec.Template.Spec.Containers[0].Resources.Limits[corev1.ResourceEphemeralStorage] = resource.MustParse("0Mi")
+	newGs, err := framework.AgonesClient.AgonesV1().GameServers(defaultNs).Create(gs)
+
+	assert.Nil(t, err, fmt.Sprintf("creating %v GameServer instances failed (%v): %v", gs.Spec, gs.Name, err))
+
+	logrus.WithField("name", newGs.ObjectMeta.Name).Info("GameServer created, waiting for being Evicted and Unhealthy")
+
+	_, err = framework.WaitForGameServerState(newGs, agonesv1.GameServerStateUnhealthy, 5*time.Minute)
+
+	assert.Nil(t, err, fmt.Sprintf("waiting for %v GameServer Unhealthy state timed out (%v): %v",
+		gs.Spec, gs.Name, err))
+}
+
 func TestGameServerPassthroughPort(t *testing.T) {
 	t.Parallel()
 	gs := defaultGameServer()