Revamp how Tekton pipelines to run notebooks work.

Makefile

@@ -13,3 +13,29 @@ hydrate:
 	kustomize build -o $(REPO_DIRS)/kf-ci-v1/namespaces/auto-deploy $(TEKTON_INSTALLS)/auto-deploy
 	kustomize build -o $(REPO_DIRS)/kf-ci-v1/namespaces/kf-ci $(TEKTON_INSTALLS)/kf-ci
 
+build-worker-image:
+	cd images && skaffold build -p testing --kube-context=kubeflow-testing -v info --file-output=latest_image.json
+
+set-worker-image:
+	kpt cfg set ./tekton test-image $(shell yq r ./images/latest_image.json builds[0].tag)
+
+update-worker-image: build-worker-image set-worker-image	
+
+# This is a debug rule providing some sugar to hydrate and push the manifests and then wait for the
+# sync
+debug-push-and-run:
+	make hydrate && git add . && git commit -m "Latest" && git push jlewi
+	cd ./go/cmd/nomos-wait && go run .
+	kubectl --context=kf-ci-v1 create -f ./tekton/runs/nb-test-run.yaml 
+
+# This is a debug rule providing some sugar for fast iteration during development
+# It might need to be customized for your usage.
+# make-update-worker-image builds and sets a new worker image.
+# make hydrate ... rehydrates and pushes the Tekton resources
+# nomos-wait waits for the latest nomos changes to be sync'd
+# and then we submit a run of the pipeline.
+debug-rebuild-and-run:
+	make update-worker-image
+	make hydrate && git add . && git commit -m "Latest" && git push jlewi
+	cd ./go/cmd/nomos-wait && go run .
+	kubectl --context=kf-ci-v1 create -f ./tekton/runs/nb-test-run.yaml 

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_pipeline_notebook-test.yaml

@@ -9,6 +9,16 @@ spec:
     description: Test targe name, used to group test results in JUNIT.
     name: test-target-name
     type: string
+  - description: This should be the bucket that the rendered notebook will be written
+      to. This should be a GCS path that is accessible from the KF cluster where the
+      notebook runs. It will be copied to artifacts-gcs
+    name: notebook-output
+    type: string
+  - description: Path to the notebook to run. This should be the relative path relative
+      to the root of the repository where the notebook lives. Do not include a leading
+      "/"
+    name: notebook-path
+    type: string
   - description: GCS bucket and directory artifacts will be uploaded to. Should be
       in the form of 'gs://'
     name: artifacts-gcs
@@ -24,36 +34,41 @@ spec:
   - description: Location to search for test clusters.
     name: testing-cluster-location
     type: string
-  - description: Directory to write outputs to in local FS.
-    name: output-workspace
-    type: string
   resources:
-  - name: examples-repo
+  - name: notebook-repo
     type: git
   - name: testing-repo
     type: git
+  - name: image
+    type: image
   tasks:
-  - name: mnist-gcp
+  - name: build-image
+    params:
+    - name: artifacts-gcs
+      value: $(params.artifacts-gcs)
+    resources:
+      inputs:
+      - name: notebook-repo
+        resource: notebook-repo
+      - name: image
+        resource: image
+    taskRef:
+      name: notebook-test-builder
+  - name: run-notebook
     params:
     - name: notebook-path
-      value: kubeflow/examples/mnist/mnist_gcp.ipynb
-    - name: junit-path
-      value: $(params.output-workspace)/$(params.junit-path)/junit_mnist-gcp.xml
+      value: $(params.notebook-path)
     - name: test-target-name
       value: $(params.test-target-name)
-    - name: output-workspace
-      value: $(params.output-workspace)
+    - name: notebook-output
+      value: $(params.notebook-output)
     - name: artifacts-gcs
       value: $(params.artifacts-gcs)
     - name: testing-cluster-pattern
       value: $(params.testing-cluster-pattern)
     - name: testing-cluster-location
       value: $(params.testing-cluster-location)
-    resources:
-      inputs:
-      - name: examples-repo
-        resource: examples-repo
-      - name: kf-testing-repo
-        resource: testing-repo
+    runAfter:
+    - build-image
     taskRef:
       name: nb-tests

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_cleanup-kubeflow-ci.yaml

@@ -48,7 +48,7 @@ spec:
       value: /workspace/kubeconfig
     - name: PYTHONPATH
       value: /workspace/$(inputs.resources.testing-repo.name)/py
-    image: gcr.io/kubeflow-ci/test-worker-py3:6f0d932-dirty@sha256:06ebe5412d638e3e51bdd792aecbafdc4ee1e7146ff367a7be346cd726738cbb
+    image: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
     name: create-context
   - command:
     - python

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_deploy-gcp-blueprint.yaml

@@ -39,7 +39,7 @@ spec:
     env:
     - name: KUBECONFIG
       value: /workspace/kubeconfig
-    image: gcr.io/kubeflow-ci/test-worker-py3:6f0d932-dirty@sha256:06ebe5412d638e3e51bdd792aecbafdc4ee1e7146ff367a7be346cd726738cbb
+    image: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
   steps:
   - command:
     - /workspace/$(inputs.resources.blueprint-repo.name)/kubeflow/hack/create_context.sh

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_golang-test.yaml

@@ -23,7 +23,7 @@ spec:
         be in the form of 'gs://'
       name: artifacts-gcs
       type: string
-    - default: gcr.io/kubeflow-ci/test-worker-py3:6f0d932-dirty@sha256:06ebe5412d638e3e51bdd792aecbafdc4ee1e7146ff367a7be346cd726738cbb
+    - default: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
       description: The docker image to run the tests in
       name: test-image
       type: string

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_iap-ready.yaml

@@ -23,7 +23,7 @@ spec:
         be in the form of 'gs://'
       name: artifacts-gcs
       type: string
-    - default: gcr.io/kubeflow-ci/test-worker-py3:6f0d932-dirty@sha256:06ebe5412d638e3e51bdd792aecbafdc4ee1e7146ff367a7be346cd726738cbb
+    - default: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
       description: The docker image to run the tests in
       name: test-image
       type: string

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_kf-ready.yaml

@@ -23,7 +23,7 @@ spec:
         be in the form of 'gs://'
       name: artifacts-gcs
       type: string
-    - default: gcr.io/kubeflow-ci/test-worker-py3:6f0d932-dirty@sha256:06ebe5412d638e3e51bdd792aecbafdc4ee1e7146ff367a7be346cd726738cbb
+    - default: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
       description: The docker image to run the tests in. Should contain a version
         of kubeflow/testing/py in /srcCache that we want to use.
       name: test-image

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_nb-tests.yaml

@@ -8,7 +8,8 @@ metadata:
 spec:
   inputs:
     params:
-    - description: Testing notebook location. Should be in the form of {REPO_OWNER}/{REPO}/path/to/notebook.ipynb
+    - description: Path to the notebook to run. This should be the relative path relative
+        to the root of the repository where the notebook lives.
       name: notebook-path
       type: string
     - description: Cluster pattern to run the notebook test. Default to be from master
@@ -18,32 +19,27 @@ spec:
     - description: Location to search for test clusters e.g. us-central1 or us-central1-f
       name: testing-cluster-location
       type: string
+    - description: This should be the bucket that the rendered notebook will be written
+        to. This should be a GCS path that is accessible from the KF cluster where
+        the notebook runs. It will be copied to artifacts-gcs
+      name: notebook-output
+      type: string
     - description: GCS bucket and directory artifacts will be uploaded to. Should
         be in the form of 'gs://'
       name: artifacts-gcs
       type: string
-    - description: Relative path to the GCS artifacts will be uploaded to. Base path
-        is artifacts-gcs so the actual GCS blob will be artifacts-gcs/junit-path
-      name: junit-path
-      type: string
     - default: manual-testing
       description: Test targe name, used to group test results in JUNIT.
       name: test-target-name
       type: string
-    - description: Directory to write outputs to in local FS.
-      name: output-workspace
-      type: string
     - default: default-profile
       description: The namespace to run the notebook in
       name: nb-namespace
       type: string
-    resources:
-    - name: examples-repo
-      targetPath: src/kubeflow/examples
-      type: git
-    - name: kf-testing-repo
-      targetPath: src/kubeflow/testing
-      type: git
+    - default: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
+      description: The docker image to run the tests in
+      name: test-image
+      type: string
   steps:
   - args:
     - -m
@@ -52,62 +48,49 @@ spec:
     - --location=$(inputs.params.testing-cluster-location)
     - get-credentials
     command:
-    - python3
+    - python
     env:
     - name: PYTHONPATH
-      value: /workspace/src/kubeflow/examples/py:/workspace/src/kubeflow/testing/py
-    - name: GOOGLE_APPLICATION_CREDENTIALS
-      value: /secret/gcp-credentials/key.json
-    image: gcr.io/kubeflow-ci/test-worker:latest
+      value: /srcCache/kubeflow/testing/py
+    image: $(inputs.params.test-image)
     name: get-credential
-    volumeMounts:
-    - mountPath: /secret/gcp-credentials
-      name: gcp-credentials
-      readOnly: true
   - env:
     - name: PYTHONPATH
-      value: /workspace/src/kubeflow/examples/py:/workspace/src/kubeflow/testing/py
-    - name: GOOGLE_APPLICATION_CREDENTIALS
-      value: /secret/gcp-credentials/key.json
-    image: gcr.io/kubeflow-ci/test-worker:latest
+      value: /srcCache/kubeflow/testing/py
+    image: $(inputs.params.test-image)
     name: run-notebook
     script: |
       #!/usr/bin/env bash
+      set -x
+      mkdir -p /workspace/artifacts
       pytest run_notebook_test.py \
         --log-cli-level=info \
         --log-cli-format='%(levelname)s|%(asctime)s|%(pathname)s|%(lineno)d| %(message)s' \
         --timeout=1800 \
-        --junitxml=$(inputs.params.junit-path) \
-        --notebook_path=$(inputs.params.notebook-path) \
+        --junitxml=/workspace/artifacts/junit_notebook.xml \
+        --notebook_path=/src/notebook-repo/$(inputs.params.notebook-path) \
         --test-target-name=$(inputs.params.test-target-name) \
-        --artifacts-gcs=$(inputs.params.artifacts-gcs) \
+        --artifacts-gcs=$(inputs.params.notebook-output) \
+        --image_file=$(inputs.params.artifacts-gcs)/image.yaml
         --namespace=$(inputs.params.nb-namespace)
-      || echo test finished.
-    volumeMounts:
-    - mountPath: /secret/gcp-credentials
-      name: gcp-credentials
-      readOnly: true
-    workingDir: /workspace/src/kubeflow/examples/py/kubeflow/examples/notebook_tests
+      echo test finished
+    workingDir: /srcCache/kubeflow/testing/py/kubeflow/testing/notebook_tests
+  - image: $(inputs.params.test-image)
+    name: copy-buckets
+    script: |
+      #!/usr/bin/env bash
+      set -x
+      gsutil cp -r $(inputs.params.output)/ $(inputs.params.artifacts-gcs)
   - args:
     - -m
     - kubeflow.testing.tekton_client
     - junit_parse_and_upload
-    - --artifacts_dir=$(inputs.params.output-workspace)
+    - --artifacts_dir=/workspace/artifacts
     - --output_gcs=$(inputs.params.artifacts-gcs)
     command:
     - python
     env:
     - name: PYTHONPATH
-      value: /workspace/src/kubeflow/examples/py:/workspace/src/kubeflow/testing/py
-    - name: GOOGLE_APPLICATION_CREDENTIALS
-      value: /secret/gcp-credentials/key.json
-    image: gcr.io/kubeflow-ci/test-worker:latest
+      value: /src/kubeflow/testing/py
+    image: $(inputs.params.test-image)
     name: copy-artifacts
-    volumeMounts:
-    - mountPath: /secret/gcp-credentials
-      name: gcp-credentials
-      readOnly: true
-  volumes:
-  - name: gcp-credentials
-    secret:
-      secretName: gcp-credentials

acm-repos/kf-ci-v1/namespaces/auto-deploy/tekton.dev_v1alpha1_task_notebook-test-builder.yaml

@@ -0,0 +1,66 @@
+apiVersion: tekton.dev/v1alpha1
+kind: Task
+metadata:
+  annotations:
+    sidecar.istio.io/inject: "false"
+  name: notebook-test-builder
+  namespace: auto-deploy
+spec:
+  inputs:
+    params:
+    - default: gcr.io/kubeflow-ci/test-worker-py3:3780b5d-dirty@sha256:4a766d6f5cc6cbcb00dbc96205f7a5b2816bc5f2b6d516fd67124d4a3e6508ea
+      description: The docker image to run the tests in
+      name: test-image
+      type: string
+    - description: GCS bucket and directory artifacts will be uploaded to. Should
+        be in the form of 'gs://'
+      name: artifacts-gcs
+      type: string
+    resources:
+    - name: notebook-repo
+      type: git
+    - name: image
+      type: image
+  steps:
+  - image: $(inputs.params.test-image)
+    name: setup
+    script: |
+      #!/usr/bin/env bash
+      set -x
+      mkdir -p /workspace/build
+      cd /workspace/build
+      # Copy the source code
+      cp -r /workspace/$(inputs.resources.notebook-repo.name) .
+      cp -r /srcCache/kubeflow/testing/notebook_testing/Dockerfile.notebook_runner ./Dockerfile.notebook_runner
+      mkdir -p kubeflow/
+      # Copy over the kubeflow/testing directory because we need it to run the
+      # notebooks; note that the copy is coming from the worker test image.
+      cp -r /srcCache/kubeflow/testing ./kubeflow/testing
+      # Create the artifacts directory
+      mkdir -p /workspace/artifacts
+  - command:
+    - /kaniko/executor
+    - --dockerfile=/workspace/build/Dockerfile.notebook_runner
+    - --destination=$(inputs.resources.image.url)
+    - --context=/workspace/build
+    - --digest-file=/workspace/artifacts/image-digest
+    image: gcr.io/kaniko-project/executor:v0.23.0
+    name: build-push
+    resources:
+      requests:
+        cpu: 7
+        memory: 16Gi
+  - args:
+    - -m
+    - kubeflow.testing.tekton_client
+    - create-image-file
+    - --image-name=$(inputs.resources.image.url)
+    - --digest-file=/workspace/artifacts/image-digest
+    - --output=$(inputs.params.artifacts-gcs)/image.yaml
+    command:
+    - python
+    env:
+    - name: PYTHONPATH
+      value: /srcCache/kubeflow/testing/py
+    image: $(inputs.params.test-image)
+    name: create-image-file