hacks to avoid dockerhub rate limits during CI (#42)

Author: dgrove-oss

hack/e2e-util.sh

@@ -20,14 +20,16 @@ export KA_BIN=_output/bin
 export WAIT_TIME="20s"
 export KUTTL_VERSION=0.15.0
 export CERTMANAGER_VERSION=v1.13.3
-export KUBEFLOW_VERSION=v1.7.0
 DUMP_LOGS="true"
 
+# These must be kept in synch -- we prepull and load these to mitigate dockerhub rate limits
+export KUBEFLOW_VERSION=v1.7.0
+export IMAGE_KUBEFLOW_OPERATOR="docker.io/kubeflow/training-operator:v1-855e096"
+
 # These are small images used by the e2e tests.
 # Pull and kind load to avoid long delays during testing
 export IMAGE_ECHOSERVER="quay.io/project-codeflare/echo-server:1.0"
 export IMAGE_BUSY_BOX_LATEST="quay.io/project-codeflare/busybox:latest"
-export IMAGE_ALPINE_310="docker.io/alpine:3.10"
 
 function update_test_host {
 
@@ -115,7 +117,7 @@ function check_prerequisites {
 }
 
 function pull_images {
-  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_ALPINE_310}
+  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_KUBEFLOW_OPERATOR}
   do
       docker pull $image
       if [ $? -ne 0 ]
@@ -138,7 +140,7 @@ function kind_up_cluster {
   fi
   CLUSTER_STARTED="true"
 
-  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_ALPINE_310}
+  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_KUBEFLOW_OPERATOR}
   do
     kind load docker-image ${image} ${CLUSTER_CONTEXT}
     if [ $? -ne 0 ]

test/e2e/appwrapper_test.go

@@ -68,7 +68,7 @@ var _ = Describe("AppWrapper E2E Test", func() {
 
 	Describe("Creation of Kubeflow Training Operator GVKs", func() {
 		It("PyTorch Jobs", func() {
-			aw := createAppWrapper(ctx, pytorchjob(1, 100, 2, 250))
+			aw := createAppWrapper(ctx, pytorchjob(2, 250))
 			appwrappers = append(appwrappers, aw)
 			Expect(waitAWPodsReady(ctx, aw)).Should(Succeed())
 		})

test/e2e/fixtures_test.go

@@ -209,29 +209,16 @@ func batchjob(milliCPU int64) workloadv1beta2.AppWrapperComponent {
 	}
 }
 
+// This is not a useful PyTorchJob:
+// 1. Using a dummy busybox image to avoid pulling a large & rate-limited image from dockerhub
+// 2. We avoid needing the injected sidecar (alpine:3.10 from dockerhub) by not specifying a Master
 const pytorchYAML = `
 apiVersion: "kubeflow.org/v1"
 kind: PyTorchJob
 metadata:
   name: %v
 spec:
   pytorchReplicaSpecs:
-    Master:
-      replicas: %v
-      restartPolicy: OnFailure
-      template:
-        spec:
-          terminationGracePeriodSeconds: 0
-          containers:
-          - name: pytorch
-            image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v1beta1-fc858d1
-            command:
-            - "python3"
-            - "/opt/pytorch-mnist/mnist.py"
-            - "--epochs=1"
-            resources:
-              requests:
-                cpu: %v
     Worker:
       replicas: %v
       restartPolicy: OnFailure
@@ -240,31 +227,24 @@ spec:
           terminationGracePeriodSeconds: 0
           containers:
           - name: pytorch
-            image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v1beta1-fc858d1
-            command:
-            - "python3"
-            - "/opt/pytorch-mnist/mnist.py"
-            - "--epochs=1"
+            image: quay.io/project-codeflare/busybox:1.36
+            command: ["sh", "-c", "sleep 10"]
             resources:
               requests:
                 cpu: %v
 `
 
-func pytorchjob(replicasMaster int, milliCPUMaster int64, replicasWorker int, milliCPUWorker int64) workloadv1beta2.AppWrapperComponent {
+func pytorchjob(replicasWorker int, milliCPUWorker int64) workloadv1beta2.AppWrapperComponent {
 	yamlString := fmt.Sprintf(pytorchYAML,
 		randName("pytorchjob"),
-		replicasMaster,
-		resource.NewMilliQuantity(milliCPUMaster, resource.DecimalSI),
 		replicasWorker,
 		resource.NewMilliQuantity(milliCPUWorker, resource.DecimalSI),
 	)
 	jsonBytes, err := yaml.YAMLToJSON([]byte(yamlString))
 	Expect(err).NotTo(HaveOccurred())
-	masters := int32(replicasMaster)
 	workers := int32(replicasWorker)
 	return workloadv1beta2.AppWrapperComponent{
 		PodSets: []workloadv1beta2.AppWrapperPodSet{
-			{Replicas: &masters, Path: "template.spec.pytorchReplicaSpecs.Master.template"},
 			{Replicas: &workers, Path: "template.spec.pytorchReplicaSpecs.Worker.template"},
 		},
 		Template: runtime.RawExtension{Raw: jsonBytes},