agones-{extensions,allocator}: Be more defensive about draining

I looked at the flakes for TestGameServerCreationRightAfterDeletingOneExtensionsPod and TestAllocatorAfterDeleteReplica, and looked at kubelet logs. We definitely see the Pod go un-ready, but the timing is fairly tight and I'm not sure how long it takes the Service to react. Let's be a little more defensive here and lame-duck for a little longer.
googleforgames · May 23, 2024 · 68b04ee · 68b04ee
1 parent bff72b1
commit 68b04ee
Show file tree

Hide file tree

Showing 5 changed files with 10 additions and 10 deletions.
diff --git a/install/helm/agones/templates/extensions-deployment.yaml b/install/helm/agones/templates/extensions-deployment.yaml
@@ -92,7 +92,7 @@ spec:
       priorityClassName: {{ .Values.agones.priorityClassName }}
       {{- end }}
       serviceAccountName: {{ .Values.agones.serviceaccount.controller.name }}
-      terminationGracePeriodSeconds: {{ mul .Values.agones.extensions.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 3 }}
+      terminationGracePeriodSeconds: {{ mul .Values.agones.extensions.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 5 }}
       containers:
       - name: agones-extensions
         image: "{{ .Values.agones.image.registry }}/{{ .Values.agones.image.extensions.name}}:{{ default .Values.agones.image.tag .Values.agones.image.extensions.tag }}"
@@ -137,7 +137,7 @@ spec:
         - name: CONTAINER_NAME
           value: "agones-extensions"
         - name: READINESS_SHUTDOWN_DURATION
-          value: {{ mul .Values.agones.extensions.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 2 }}s
+          value: {{ mul .Values.agones.extensions.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 4 }}s
         ports:
         - name: webhooks
           containerPort: 8081

diff --git a/install/helm/agones/templates/service/allocation.yaml b/install/helm/agones/templates/service/allocation.yaml
@@ -213,7 +213,7 @@ spec:
 {{ toYaml .Values.agones.allocator.tolerations | indent 8 }}
       {{- end }}
       serviceAccountName: {{ $.Values.agones.serviceaccount.allocator.name }}
-      terminationGracePeriodSeconds: {{ mul .Values.agones.allocator.readiness.periodSeconds .Values.agones.allocator.readiness.failureThreshold 3 }}
+      terminationGracePeriodSeconds: {{ mul .Values.agones.allocator.readiness.periodSeconds .Values.agones.allocator.readiness.failureThreshold 5 }}
       {{- if eq .Values.agones.allocator.disableTLS false }}
       volumes:
       - name: tls
@@ -292,7 +292,7 @@ spec:
         - name: ALLOCATION_BATCH_WAIT_TIME
           value: {{ .Values.agones.allocator.allocationBatchWaitTime | quote }}
         - name: READINESS_SHUTDOWN_DURATION
-          value: {{ mul .Values.agones.allocator.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 2 }}s
+          value: {{ mul .Values.agones.allocator.readiness.periodSeconds .Values.agones.extensions.readiness.failureThreshold 4 }}s
         ports:
         {{- if .Values.agones.allocator.service.http.enabled }}
         - name: {{ .Values.agones.allocator.service.http.portName }}

diff --git a/install/yaml/install.yaml b/install/yaml/install.yaml
@@ -17216,7 +17216,7 @@ spec:
           value: "true"
       priorityClassName: agones-system
       serviceAccountName: agones-controller
-      terminationGracePeriodSeconds: 27
+      terminationGracePeriodSeconds: 45
       containers:
       - name: agones-extensions
         image: "us-docker.pkg.dev/agones-images/release/agones-extensions:1.41.0-dev"
@@ -17259,7 +17259,7 @@ spec:
         - name: CONTAINER_NAME
           value: "agones-extensions"
         - name: READINESS_SHUTDOWN_DURATION
-          value: 18s
+          value: 36s
         ports:
         - name: webhooks
           containerPort: 8081
@@ -17420,7 +17420,7 @@ spec:
           operator: Equal
           value: "true"
       serviceAccountName: agones-allocator
-      terminationGracePeriodSeconds: 27
+      terminationGracePeriodSeconds: 45
       volumes:
       - name: tls
         secret:
@@ -17491,7 +17491,7 @@ spec:
         - name: ALLOCATION_BATCH_WAIT_TIME
           value: "500ms"
         - name: READINESS_SHUTDOWN_DURATION
-          value: 18s
+          value: 36s
         ports:
         - name: https
           containerPort: 8443

diff --git a/test/e2e/allocator/pod_termination_test.go b/test/e2e/allocator/pod_termination_test.go
@@ -34,7 +34,7 @@ import (
 
 const (
 	retryInterval = 5 * time.Second
-	retryTimeout  = 45 * time.Second
+	retryTimeout  = 60 * time.Second
 )
 
 func TestAllocatorAfterDeleteReplica(t *testing.T) {

diff --git a/test/e2e/extensions/high_availability_test.go b/test/e2e/extensions/high_availability_test.go
@@ -84,7 +84,7 @@ func TestGameServerCreationRightAfterDeletingOneExtensionsPod(t *testing.T) {
 	logger.Infof("Removing one of the Extensions Pods: %v", list.Items[1].ObjectMeta.Name)
 	deleteAgonesExtensionsPod(ctx, t, false)
 
-	endTime := time.Now().Add(30 * time.Second)
+	endTime := time.Now().Add(60 * time.Second)
 	for time.Now().Before(endTime) {
 		gs := framework.DefaultGameServer(defaultNs)
 		logger.Infof("Creating game-server %s...", gs.Name)