hack around diskspace problem in CI when using real pytorch jobs

Author: dgrove-oss

hack/e2e-util.sh

@@ -27,7 +27,6 @@ DUMP_LOGS="true"
 # 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_PYTORCH_SAMPLE="docker.io/kubeflowkatib/pytorch-mnist:v1beta1-45c5727"
 
 function update_test_host {
 
@@ -115,7 +114,7 @@ function check_prerequisites {
 }
 
 function pull_images {
-  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_PYTORCH_SAMPLE}
+  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST}
   do
       docker pull $image
       if [ $? -ne 0 ]
@@ -138,7 +137,7 @@ function kind_up_cluster {
   fi
   CLUSTER_STARTED="true"
 
-  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST} ${IMAGE_PYTORCH_SAMPLE}
+  for image in ${IMAGE_ECHOSERVER} ${IMAGE_BUSY_BOX_LATEST}
   do
     kind load docker-image ${image} ${CLUSTER_CONTEXT}
     if [ $? -ne 0 ]

test/e2e/fixtures_test.go

@@ -209,6 +209,8 @@ func batchjob(milliCPU int64) workloadv1beta2.AppWrapperComponent {
 	}
 }
 
+// Using busybox is a hack to avoid pulling a massive pytorch image onto the GitHub runner.
+// To test the AppWrapper side of things, we don't actually care what the containers are doing
 const pytorchYAML = `
 apiVersion: "kubeflow.org/v1"
 kind: PyTorchJob
@@ -223,31 +225,25 @@ spec:
         spec:
           terminationGracePeriodSeconds: 0
           containers:
-          - name: pytorch
-            image: docker.io/kubeflowkatib/pytorch-mnist:v1beta1-45c5727
-            command:
-            - "python3"
-            - "/opt/pytorch-mnist/mnist.py"
-            - "--epochs=1"
-            resources:
-              requests:
-                cpu: %v
+          - name: busybox
+          image: quay.io/project-codeflare/busybox:1.36
+          command: ["sh", "-c", "sleep 120"]
+          resources:
+            requests:
+              cpu: %v
     Worker:
       replicas: %v
       restartPolicy: OnFailure
       template:
         spec:
           terminationGracePeriodSeconds: 0
           containers:
-          - name: pytorch
-            image: docker.io/kubeflowkatib/pytorch-mnist:v1beta1-45c5727
-            command:
-            - "python3"
-            - "/opt/pytorch-mnist/mnist.py"
-            - "--epochs=1"
-            resources:
-              requests:
-                cpu: %v
+          - name: busybox
+          image: quay.io/project-codeflare/busybox:1.36
+          command: ["sh", "-c", "sleep 120"]
+          resources:
+            requests:
+              cpu: %v
 `
 
 func pytorchjob(replicasMaster int, milliCPUMaster int64, replicasWorker int, milliCPUWorker int64) workloadv1beta2.AppWrapperComponent {