Skip to content

Latest commit

 

History

History
955 lines (665 loc) · 40.1 KB

README.md

File metadata and controls

955 lines (665 loc) · 40.1 KB

Table of Contents generated with DocToc

Test Infrastructure

This directory contains the Kubeflow test Infrastructure.

We use Prow, K8s' continuous integration tool.

  • Prow is a set of binaries that run on Kubernetes and respond to GitHub events.

We use Prow to run:

  • Presubmit jobs
  • Postsubmit jobs
  • Periodic tests

Here's how it works

  • Prow is used to trigger E2E tests
  • The E2E test will launch an Argo workflow that describes the tests to run
  • Each step in the Argo workflow will be a binary invoked inside a container
  • The Argo workflow will use an NFS volume to attach a shared POSIX compliant filesystem to each step in the workflow.
  • Each step in the pipeline can write outputs and junit.xml files to a test directory in the volume
  • A final step in the Argo pipeline will upload the outputs to GCS so they are available in gubernator

Quick Links

Anatomy of our Tests

  • Our prow jobs are defined here
  • Each prow job defines a K8s PodSpec indicating a command to run
  • Our prow jobs use run_e2e_workflow.py to trigger an Argo workflow that checks out our code and runs our tests.
  • Our tests are structured as Argo workflows so that we can easily perform steps in parallel.
  • The Argo workflow is defined in the repository being tested
    • We always use the worfklow at the commit being tested
  • checkout.sh is used to checkout the code being tested
    • This also checks out kubeflow/testing so that all repositories can rely on it for shared tools.

Accessing The Argo UI

The UI is publicly available at http://testing-argo.kubeflow.org/

Working with the test infrastructure

The tests store the results of tests in a shared NFS filesystem. To inspect the results you can mount the NFS volume.

To make this easy, We run a stateful set in our test cluster that mounts the same volumes as our Argo workers. Furthermore, this stateful set is using an environment (GCP credentials, docker image, etc...) that mimics our Argo workers. You can ssh into this stateful set in order to get access to the NFS volume.

kubectl exec -it debug-worker-0 /bin/bash

This can be very useful for reproducing test failures.

Logs

Logs from the E2E tests are available in a number of places and can be used to troubleshoot test failures.

Prow

These should be publicly accessible.

The logs from each step are copied to GCS and made available through gubernator. The K8s-ci robot should post a link to the gubernator UI in the PR. You can also find them as follows

  1. Open up the prow jobs dashboard e.g. for kubeflow/kubeflow
  2. Find your job
  3. Click on the link under job; this goes to the Gubernator dashboard
  4. Click on artifacts
  5. Navigate to artifacts/logs

If these logs aren't available it could indicate a problem running the step that uploads the artifacts to GCS for gubernator. In this case you can use one of the alternative methods listed below.

Argo UI

The argo UI is publicly accessible at http://testing-argo.kubeflow.org/timeline.

  1. Find and click on the workflow corresponding to your pre/post/periodic job
  2. Select the workflow tab
  3. From here you can select a specific step and then see the logs for that step

Stackdriver logs

Since we run our E2E tests on GKE, all logs are persisted in Stackdriver logging.

Viewer access to Stackdriver logs is available by joining one of the following groups

We use the new stackdriver Kubernetes logging which means we use the k8s_pod and k8s_container resource types.

Below are some relevant filters:

Get container logs for a specific pod

resource.type="k8s_container"
resource.labels.cluster_name="kubeflow-testing"
resource.labels.pod_name="${POD_NAME}"

Get logs using pod label

resource.type="k8s_container"
resource.labels.cluster_name="kubeflow-testing"
metadata.userLabels.${LABEL_KEY}="${LABEL_VALUE}"

Get events for a pod

resource.type="k8s_pod"
resource.labels.cluster_name="${CLUSTER}"
resource.labels.pod_name="${POD_NAME}"

The Kubeflow docs have some useful gcloud one liners for fetching logs.

Debugging Failed Tests

No results show up in Gubernator

If no results show up in Gubernator this means the prow job didn't get far enough to upload any results/logs to GCS.

To debug this you need the pod logs. You can access the pod logs via the build log link for your job in the prow jobs UI

  • Pod logs are ephmeral so you need to check shortly after your job runs.

The pod logs are available in StackDriver but only the Google Kubeflow Team has access

  • Prow runs on a cluster owned by the K8s team not Kubeflow
  • This policy is determined by K8s not Kubeflow
  • This could potentially be fixed by using our own prow build cluster issue#32

To access the stackdriver logs

resource.type="container"
resource.labels.pod_id=${POD_ID}
  • For example, if the TF serving workflow failed, filter the logs using
resource.type="container"
resource.labels.cluster_name="kubeflow-testing"
labels."container.googleapi.com/namespace_name"=WORKFLOW_NAME
resource.labels.container_name="mnist-cpu"

No Logs in Argo UI For Step or Pod Id missing in Argo Logs

An Argo workflow fails and you click on the failed step in the Argo UI to get the logs and you see the error

failed to get container status {"docker" "b84b751b0102b5658080a520c9a5c2655855960c4695cf557c0c1e45999f7429"}: 
rpc error: code = Unknown desc = Error: No such container: b84b751b0102b5658080a520c9a5c2655855960c4695cf557c0c1e45999f7429

This error is a red herring; it means the pod is probably gone so Argo couldn't get the logs.

The logs should be in StackDriver but to get them we need to identify the pod.

  1. Get the workflow spec:

  2. Search the YAML spec for the pod information for the failed step

    • We need to find information that can be used to fetch logs for the pod from stackdriver

      1. Using Pod labels

        • In the workflow spec look at the step metadata to see if it contains labels

          metadata:
            labels:
              BUILD_ID: "1405"
              BUILD_NUMBER: "1405"
              JOB_NAME: kubeflow-examples-presubmit
              JOB_TYPE: presubmit
              PULL_BASE_SHA: 8a26b23e3d35d5993d93e8b9ecae52371598d1cc
              PULL_NUMBER: "522"
              PULL_PULL_SHA: 9aecf80f1c41059cd8ff13d1ca8b9e821dc462bf
              REPO_NAME: examples
              REPO_OWNER: kubeflow
              step_name: tfjob-test
              workflow: kubeflow-examples-presubmit-gis-522-9aecf80-1405-9055
              workflow_template: gis
        • Follow the stackdriver instructions to query for the logs

          • Use labels BUILD_ID and step_name to identify the pod
      2. If no labels are specified for the step you can use displayName to match the text in the UI to step status

        kubeflow-presubmit-kfctl-1810-70210d5-3900-218a-2243590372:
        boundaryID: kubeflow-presubmit-kfctl-1810-70210d5-3900-218a
        displayName: kfctl-apply-gcp
        finishedAt: 2018-10-17T05:07:58Z
        id: kubeflow-presubmit-kfctl-1810-70210d5-3900-218a-2243590372
        message: failed with exit code 1
        name: kubeflow-presubmit-kfctl-1810-70210d5-3900-218a.kfctl-apply-gcp
        phase: Failed
        startedAt: 2018-10-17T05:04:20Z
        templateName: kfctl-apply-gcp
        type: Pod
        • id will be the name of the pod.

        • Follow the instructions to get the stackdriver logs for the pod or use the following gcloud command

            gcloud --project=kubeflow-ci logging read --format="table(timestamp, resource.labels.container_name, textPayload)" \
            --freshness=24h \
            --order asc  \
            "resource.type=\"k8s_container\" resource.labels.pod_name=\"${POD}\"  "

Debugging Failed Deployments

If an E2E test fails because one of the Kubeflow applications (e.g. the Jupyter web app) isn't reported as deploying successfully we can follow these instructions to debug it.

To debug it we want to look at the K8s events indicating why the K8s deployment failed. In most cases the cluster will already be torn down so we need to look at the kubernetes events associated with that deployment.

  1. Get the cluster used for Kubeflow

    1. In prow look at artifacts and find the YAML spec for the Argo workflow that ran your e2e test

    2. Identify the step that deployed Kubeflow

    3. Open up stack driver logging

    4. Use a filter (advanced) like the following to find the log entry getting the credentials for your deployment

      resource.type="k8s_container"
      resource.labels.pod_name=`<POD NAME>`
      resource.labels.container_name="main"
      get-credentials
      
    5. The log output should look like the following

      get-credentials kfctl-6742 --zone=us-east1-d --project=kubeflow-ci-deployment 
      
      • The argument kfctl-6742 is the name of the cluster
  2. Use a filter like the following to get the events associated with the deployment or statefulset

     resource.labels.cluster_name="kfctl-6742"
     logName="projects/kubeflow-ci-deployment/logs/events" 
     jsonPayload.involvedObject.name="jupyter-web-app"
    
    • Change the name of the involvedObject and cluster name to match your deployment.

    • If a pod was created the name of the pod should be present e.g.

      Scaled up replica set jupyter-web-app-5fcddbf75c to 1"
      
    • You can continue to look at event logs for the replica set to eventually get to the name of a pod and potentially the pod.

Testing Changes to the ProwJobs

Changes to our ProwJob configs in config.yaml should be relatively infrequent since most of the code invoked as part of our tests lives in the repository.

However, in the event we need to make changes here are some instructions for testing them.

Follow Prow's getting started guide to create your own prow cluster.

  • Note The only part you really need is the ProwJob CRD and controller.

Checkout kubernetes/test-infra.

git clone https://github.com/kubernetes/test-infra git_k8s-test-infra

Build the mkpj binary

bazel build build prow/cmd/mkpj

Generate the ProwJob Config

./bazel-bin/prow/cmd/mkpj/mkpj --job=$JOB_NAME --config-path=$CONFIG_PATH
  • This binary will prompt for needed information like the sha #
  • The output will be a ProwJob spec which can be instantiated using kubectl

Create the ProwJob

kubectl create -f ${PROW_JOB_YAML_FILE}
  • To rerun the job bump metadata.name and status.startTime

To monitor the job open Prow's UI by navigating to the external IP associated with the ingress for your Prow cluster or using kubectl proxy.

Cleaning up leaked resources

Test failures sometimes leave resources (GCP deployments, VMs, GKE clusters) still running. The following scripts can be used to garbage collect leaked resources.

py/testing/kubeflow/testing/cleanup_ci.py --delete_script=${DELETE_SCRIPT}

There's a second script cleanup_kubeflow_ci in the kubeflow repository to cleanup resources left by ingresses.

Integration with K8s Prow Infrastructure.

We rely on K8s instance of Prow to actually run our jobs.

Here's a dashboard with the results.

Our jobs should be added to K8s config

Setting up Kubeflow Test Infrastructure

Our tests require:

  • a K8s cluster
  • Argo installed on the cluster
  • A shared NFS filesystem

Our prow jobs execute Argo worflows in project/clusters owned by Kubeflow. We don't use the shared Kubernetes test clusters for this.

  • This gives us more control of the resources we want to use e.g. GPUs

This section provides the instructions for setting this up.

Create a GKE cluster

PROJECT=kubeflow-ci
ZONE=us-east1-d
CLUSTER=kubeflow-testing
NAMESPACE=kubeflow-test-infra

gcloud --project=${PROJECT} container clusters create \
	--zone=${ZONE} \
	--machine-type=n1-standard-8 \
	${CLUSTER}

Create a static ip for the Argo UI

gcloud compute --project=${PROJECT} addresses create argo-ui --global

Enable GCP APIs

gcloud services --project=${PROJECT} enable cloudbuild.googleapis.com
gcloud services --project=${PROJECT} enable containerregistry.googleapis.com
gcloud services --project=${PROJECT} enable container.googleapis.com

Create a GCP service account

  • The tests need a GCP service account to upload data to GCS for Gubernator
SERVICE_ACCOUNT=kubeflow-testing
gcloud iam service-accounts --project=${PROJECT} create ${SERVICE_ACCOUNT} --display-name "Kubeflow testing account"
gcloud projects add-iam-policy-binding ${PROJECT} \
    	--member serviceAccount:${SERVICE_ACCOUNT}@${PROJECT}.iam.gserviceaccount.com --role roles/container.admin \
      --role=roles/viewer \
      --role=roles/cloudbuild.builds.editor \
      --role=roles/logging.viewer \
      --role=roles/storage.admin \
      --role=roles/compute.instanceAdmin.v1
  • Our tests create K8s resources (e.g. namespaces) which is why we grant it developer permissions.
  • Project Viewer (because GCB requires this with gcloud)
  • Kubernetes Engine Admin (some tests create GKE clusters)
  • Logs viewer (for GCB)
  • Compute Instance Admin to create VMs for minikube
  • Storage Admin (For GCR)
GCE_DEFAULT=${PROJECT_NUMBER}[email protected]
FULL_SERVICE=${SERVICE_ACCOUNT}@${PROJECT}.iam.gserviceaccount.com
gcloud --project=${PROJECT} iam service-accounts add-iam-policy-binding \
   ${GCE_DEFAULT} --member="serviceAccount:${FULL_SERVICE}" \
   --role=roles/iam.serviceAccountUser
  • Service Account User of the Compute Engine Default Service account (to avoid this error)

Create a secret key containing a GCP private key for the service account

KEY_FILE=<path to key>
SECRET_NAME=gcp-credentials
gcloud iam service-accounts keys create ${KEY_FILE} \
    	--iam-account ${SERVICE_ACCOUNT}@${PROJECT}.iam.gserviceaccount.com
kubectl create secret generic ${SECRET_NAME} \
    --namespace=${NAMESPACE} --from-file=key.json=${KEY_FILE}

Make the service account a cluster admin

kubectl create clusterrolebinding  ${SERVICE_ACCOUNT}-admin --clusterrole=cluster-admin  \
		--user=${SERVICE_ACCOUNT}@${PROJECT}.iam.gserviceaccount.com
  • The service account is used to deploye Kubeflow which entails creating various roles; so it needs sufficient RBAC permission to do so.

Add a clusterrolebinding that uses the numeric id of the service account as a work around for ksonnet/ksonnet#396

NUMERIC_ID=`gcloud --project=kubeflow-ci iam service-accounts describe ${SERVICE_ACCOUNT}@${PROJECT}.iam.gserviceaccount.com --format="value(oauth2ClientId)"`
kubectl create clusterrolebinding  ${SERVICE_ACCOUNT}-numeric-id-admin --clusterrole=cluster-admin  \
    --user=${NUMERIC_ID}

Create a GitHub Token

You need to use a GitHub token with ksonnet otherwise the test quickly runs into GitHub API limits.

TODO(jlewi): We should create a GitHub bot account to use with our tests and then create API tokens for that bot.

You can use the GitHub API to create a token

  • The token doesn't need any scopes because its only accessing public data and is needed only for API metering.

To create the secret run

kubectl create secret generic github-token --namespace=${NAMESPACE} --from-literal=github_token=${GITHUB_TOKEN}

Deploy NFS

We use GCP Cloud FileStore to create an NFS filesystem.

There is a deployment manager config in the directory test-infra/gcp_configs

Create K8s Resources for Testing

The ksonnet app test-infra contains ksonnet configs to deploy the test infrastructure.

First, install the kubeflow package

ks pkg install kubeflow/core

Then change the server ip in test-infra/environments/prow/spec.json to point to your cluster.

You can deploy argo as follows (you don't need to use argo's CLI)

Set up the environment

NFS_SERVER=<Internal GCE IP address of the NFS Server>
ks env add ${ENV}
ks param set --env=${ENV} argo namespace ${NAMESPACE}
ks param set --env=${ENV} debug-worker namespace ${NAMESPACE}
ks param set --env=${ENV} nfs-external namespace ${NAMESPACE}
ks param set --env=${ENV} nfs-external nfsServer ${NFS_SERVER}

In the testing environment (but not release) we also expose the UI

ks param set --env=${ENV} argo exposeUi true
ks apply ${ENV} -c argo

Create the PVs corresponding to external NFS

ks apply ${ENV} -c nfs-external

Creating secret for deployapp test

The e2e test that runs click-to-deploy app will test deploying kubeflow to a cluter under project kubeflow-ci-deployment. So it needs to know a clientID and secret of that project. Check out this page and look for client ID called deployapp-test-client.

kubectl create secret generic --namespace=${NAMESPACE} kubeflow-oauth --from-literal=client_id=${CLIENT_ID} --from-literal=client_secret=${CLIENT_SECRET}

Troubleshooting

User or service account deploying the test infrastructure needs sufficient permissions to create the roles that are created as part deploying the test infrastructure. So you may need to run the following command before using ksonnet to deploy the test infrastructure.

kubectl create clusterrolebinding default-admin --clusterrole=cluster-admin [email protected]

Setting up Kubeflow Release Clusters For Testing

We maintain a pool of Kubeflow clusters corresponding to different releases of Kubeflow. These can be used for

  • Running continuous integration of our examples against a particular release
  • Manual testing of features in each release

The configs for each deployment are stored in the test-infra directory

The deployments should be named using one of the following patterns

  • kf-vX.Y-n?? - For clusters corresponding to a particular release
  • kf-vmaster-n?? - For clusters corresponding to master

This naming scheme is chosen to allow us to cycle through a fixed set of names e.g.

kf-v0.4-n00
...
kf-v0.4-n04

The reason we want to cycle through names is because the endpoint name for the deployment needs to be manually set in the OAuth credential used for IAP. By cycling through a fixed set of names we can automate redeployment without having to manually configure the OAuth credential.

  1. Get kfctl for the desired release

  2. Run the following command

    python -m kubeflow.testing.create_kf_instance --base_name=<kf-vX.Y|kf-vmaster>
    
  3. Create a PR with the resulting config.

Setting up a Kubeflow Repository to Use Prow

  1. Define ProwJobs see pull/4951

  2. Add the ci-bots team to the repository with write access

    • Write access will allow bots in the team to update status
  3. Follow instructions for adding a repository to the PR dashboard.

  4. Add an OWNERS to your Kubeflow repository. The OWNERS file, like this one, will specify who can review and approve on this repo.

Webhooks for prow should already be configured according to these instructions for the org so you shouldn't need to set hooks per repository. * Use https://prow.k8s.io/hook as the target * Get HMAC token from k8s test team

Writing An Argo Workflow For An E2E Test

This section provides guidelines for writing Argo workflows to use as E2E tests

This guide is complementary to the E2E testing guide for TFJob operator which describes how to author tests to performed as individual steps in the workflow.

Some examples to look at

Adding an E2E test to a repository

Follow these steps to add a new test to a repository.

  1. Create a ksonnet App in that repository and define an Argo workflow if one doesn't already exist

    • We use ksonnet apps defined in each repository to define the Argo workflows corresponding to E2E tests

    • If a ksonnet app already exists you can just define a new component in that app

      1. Create a .jsonnet file (e.g by copying an existing .jsonnet file)

      2. Update the params.libsonnet to add a stanza to define params for the new component

    • You can look at prow_config.yaml (see below) to see which ksonnet apps are already defined in a repository.

  2. Modify the prow_config.yaml at the root of the repo to trigger your new test.

    • If prow_config.yaml doesn't exist (e.g. the repository is new) copy one from an existing repository (example).

    • prow_config.yaml contains an array of workflows where each workflow defines an E2E test to run; example

      workflows:
       - app_dir: kubeflow/testing/workflows
         component: workflows
         name: unittests
         job_types:
           - presubmit
         include_dirs:
           - foo/*
           - bar/*
             params:
         params:
           platform: gke
           gkeApiVersion: v1beta1
      
      • app_dir: Is the path to the ksonnet directory within the repository. This should be of the form ${GITHUB_ORG}/${GITHUB_REPO_NAME}/${PATH_WITHIN_REPO_TO_KS_APP}

      • component: This is the name of the ksonnet component to use for the Argo workflow

      • name: This is the base name to use for the submitted Argo workflow.

        • The test infrastructure appends a suffix of 22 characters (see here)

        • The result is passed to your ksonnet component via the name parameter

        • Your ksonnet component should truncate the name if necessary to satisfy K8s naming constraints.

          • e.g. Argo workflow names should be less than 63 characters because they are used as pod labels
      • job_types: This is an array specifying for which types of prow jobs this workflow should be triggered on.

        • Currently allowed values are presubmit, postsubmit, and periodic.
      • include_dirs: If specified, the pre and postsubmit jobs will only trigger this test if the PR changed at least one file matching at least one of the listed directories.

        • Python's fnmatch function is used to compare the listed patterns against the full path of modified files (see here)

        • This functionality should be used to ensure that expensive tests are only run when test impacting changes are made; particularly if its an expensive or flaky presubmit

        • periodic runs ignore include_dirs; a periodic run will trigger all workflows that include job_type periodic

      • A given ksonnet component can have multiple workflow entries to allow different triggering conditions on pre/postsubmit

        • For example, on presubmit we might run a test on a single platform (GKE) but on postsubmit that same test might run on GKE and minikube
        • this can be accomplished with different entries pointing at the same ksonnet component but with different job_types and params.
      • params: A dictionary of parameters to set on the ksonnet component e.g. by running ks param set ${COMPONENT} ${PARAM_NAME} ${PARAM_VALUE}

Prow Variables

  • For each test run PROW defines several variables that pass useful information to your job.

  • The list of variables is defined in the prow docs.

  • These variables are often used to assign unique names to each test run to ensure isolation (e.g. by appending the BUILD_NUMBER)

  • The prow variables are passed via ksonnet parameter prow_env to your workflows

    • You can copy the macros defined in util.libsonnet to parse the ksonnet parameter into a jsonnet map that can be used in your workflow.

    • Important Always define defaults for the prow variables in the dict e.g. like

      local prowDict = {
        BUILD_ID: "notset",
        BUILD_NUMBER: "notset",
        REPO_OWNER: "notset",
        REPO_NAME: "notset",
        JOB_NAME: "notset",
        JOB_TYPE: "notset",
        PULL_NUMBER: "notset",  
       } + util.listOfDictToMap(prowEnv);
      
      • This prevents jsonnet from failing in a hard to debug way in the event that you try to access a key which is not in the map.

Argo Spec

  • Guard against long names by truncating the name and using the BUILD_ID to ensure the name remains unique e.g

    local name = std.substr(params.name, 0, std.min(58, std.lenght(params.name))) + "-" + prowDict["BUILD_ID"];            
    
    • Argo workflow names need to be less than 63 characters because they are used as pod labels

    • BUILD_ID are unique for each run per repo; we suggest reserving 5 characters for the BUILD_ID.

  • Argo workflows should have standard labels corresponding to prow variables; for example

    labels: prowDict + {    
      workflow_template: "code_search",    
    },
    
    • This makes it easy to query for Argo workflows based on prow job info.

    • In addition the convention is to use the following labels

      • workflow_template: The name of the ksonnet component from which the workflow is created.
  • The templates for the individual steps in the argo workflow should also have standard labels

    labels: prowDict + {
      step_name: stepName,
      workflow_template: "code_search",
      workflow: workflowName,
    },
    
    • step_name: Name of the step (e.g. what shows up in the Argo graph)
    • workflow_template: The name of the ksonnet component from which the workflow is created.
    • workflow: The name of the Argo workflow that owns this pod.
  • Following the above conventions make it very easy to get logs for specific steps

    kubectl logs -l step_name=checkout,REPO_OWNER=kubeflow,REPO_NAME=examples,BUILD_ID=0104-064201 -c main
    
    

Creating K8s resources in tests.

Tests often need a K8s/Kubeflow deployment on which to create resources and run various tests.

Depending on the change being tested

  • The test might need exclusive access to a Kubeflow/Kubernetes cluster

    • e.g. Testing a change to a custom resource usually requires exclusive access to a K8s cluster because only one CRD and controller can be installed per cluster. So trying to test two different changes to an operator (e.g. tf-operator) on the same cluster is not good.
  • The test might need a Kubeflow/K8s deployment but doesn't need exclusive access

    • e.g. When running tests for Kubeflow examples we can isolate each test using namespaces or other mechanisms.
  • If the test needs exclusive access to the Kubernetes cluster then there should be a step in the workflow that creates a KubeConfig file to talk to the cluster.

    • e.g. E2E tests for most operators should probably spin up a new Kubeflow cluster
  • If the test just needs a known version of Kubeflow (e.g. master or v0.4) then it should use one of the test clusters in project kubeflow-ci for this

To connect to the cluster:

  • The Argo workflow should have a step that configures the KUBE_CONFIG file to talk to the cluster

    • e.g. by running gcloud container clusters get-credentials
  • The Kubeconfig file should be stored in the NFS test directory so it can be used in subsequent steps

  • Set the environment variable KUBE_CONFIG on your steps to use the KubeConfig file

NFS Directory

An NFS volume is used to create a shared filesystem between steps in the workflow.

  • Your Argo workflows should use a PVC claim to mount the NFS filesystem into each step

    • The current PVC name is nfs-external
    • This should be a parameter to allow different PVC names in different environments.
  • Use the following directory structure

    ${MOUNT_POINT}/${WORKFLOW_NAME}
                                   /src
                                       /${REPO_ORG}/${REPO_NAME}
                                   /outputs
                                   /outputs/artifacts
    
    • MOUNT_PATH: Location inside the pod where the NFS volume is mounted
    • WORKFLOW_NAME: The name of the Argo workflow
      • Each Argo workflow job has a unique name (enforced by APIServer)
      • So using WORKFLOW_NAME as root for all results associated with a particular job ensures there are no conflicts
    • /src: Any repositories that are checked out should be checked out here
      • Each repo should be checked out to the sub-directory ${REPO_ORG}/${REPO_NAME}
    • /outputs: Any files that should be sync'd to GCS for Gubernator should be written here

Step Image

  • The Docker image used by the Argo steps should be a ksonnet parameter stepImage

  • The Docker image should use an immutable image tag e.g gcr.io/kubeflow-ci/test-worker:v20181017-bfeaaf5-dirty-4adcd0

    • This ensures tests don't break if someone pushes a new test image
  • The ksonnet parameter stepImage should be set in the prow_config.yaml file defining the E2E tests

    • This makes it easy to update all the workflows to use some new image.
  • A common runtime is defined here and published to gcr.io/kubeflow-ci/test-worker

Checking out code

  • The first step in the Argo workflow should checkout out the source repos to the NFS directory

  • Use checkout.sh to checkout the repos

  • checkout.sh environment variable EXTRA_REPOS allows checking out additional repositories in addition to the repository that triggered the pre/post submit test

    • This allows your test to use source code located in a different repository
    • You can specify whether to checkout the repository at HEAD or pin to a specific commit
  • Most E2E tests will want to checkout kubeflow/testing in order to use various test utilities

Building Docker Images

There are lots of different ways to build Docker images (e.g. GCB, Docker in Docker). Current recommendation is

  • Define a Makefile to provide a convenient way to invoke Docker builds

  • Using Google Container Builder (GCB) to run builds in Kubeflow's CI system generally works better than alternatives (e.g. Docker in Docker, Kaniko)

    • Your Makefile can have alternative rules to support building locally via Docker for developers
  • Use jsonnet if needed to define GCB workflows

  • Makefile should expose variables for the following

    • Registry where image is pushed
    • TAG used for the images
  • Argo workflow should define the image paths and tag so that subsequent steps can use the newly built images