examples/github_issue_summarization/demo

Demo

This folder contains the resources needed by the Kubeflow DevRel team to setup a public demo of the GitHub Issue Summarization Example.

Public gh-demo.kubeflow.org

We currently run a public instance of the ui at gh-demo.kubeflow.org

The current setup is as follows

PROJECT=kubecon-gh-demo-1
CLUSTER=gh-demo-1003
ZONE=us-east1-d

Directory contents

• gh-app - This contains the ksonnet for deploying the public instance of the model and ui.
• gh-demo-1003 - This is the app created by kfctl

Setting up the demo

Here are the instructions for setting up the demo.

1. Follow the GKE instructions for deploying Kubeflow

   • If you are using PROJECT kubecon-gh-demo-1 you can reuse the existing OAuth client

     • Use the Cloud console to lookup Client ID and secret and set the corresponding environment variables

     • You will also need to add an authorized redirect URI for the new Kubeflow deployment

2. Follow the instructions to Setup an NFS share

   • This is needed to do distributed training with the TF estimator example

3. Create static IP for serving gh-demo.kubeflow.org

   gcloud --project=${PROJECT}  deployment-manager deployments create  --config=gh-demo-dm-config.yaml gh-public-ui
   

4. Update the Cloud DNS record gh-demo.kubeflow.org in project kubeflow-dns to use the new static ip.

5. Create a namespace for serving the UI and model

   kubectl create namespace gh-public
   

6. Deploy Seldon controller in the namespace that will serve the public model

   • This is a work around for kubeflow/kubeflow#1712

   cd gh-demo-1003/ks_app
   ks env add gh-public --namespace=gh-public
   ks generate seldon seldon
   ks apply gh-public -c seldon
   

7. Create a secret with a GitHub token

   • Follow GitHub's instructions to create a token

   • Then run the following command to create the secret

     kubectl -n gh-public create secret generic github-token --from-literal=github-token=${GITHUB_TOKEN}
     

8. Deploy the public UI and model

   cd gh-app
   ks env add gh-public --namespace=gh-public
   ks apply gh-public
   

Training and Deploying the model.

We use the ksonnet app in github/kubeflow/examples/github_issue_summarization/ks-kubeflow

The current environment is

export ENV=gh-demo-1003

Set a bucket for the job output

DAY=$(date +%Y%m%d)
ks param set --env=${ENV} tfjob-v1alpha2 output_model_gcs_bucket kubecon-gh-demo
ks param set --env=${ENV} tfjob-v1alpha2 output_model_gcs_path gh-demo/${DAY}/output

Run the job

ks apply ${ENV} -c tfjob-v1alpha2

Using TF Estimator with Keras

1. Copy the data to the GCFS mount by launching a notebook and then running the following commands

   !mkdir -p /mnt/kubeflow-gcfs/gh-demo/data
   !gcloud auth activate-service-account --key-file=${GOOGLE_APPLICATION_CREDENTIALS}
   !gsutil cp gs://kubeflow-examples/github-issue-summarization-data/github-issues.zip /mnt/kubeflow-gcfs/gh-demo/data
   !unzip /mnt/kubeflow-gcfs/gh-demo/data/github-issues.zip
   !cp github_issues.csv /mnt/kubeflow-gcfs/gh-demo/data/
   

   • TODO(jlewi): Can we modify the existing job that downloads data to a PVC to do this?

2. Run the estimator job

   ks apply ${ENV} -c tfjob-estimator
   

3. Run TensorBoard

   ks apply ${ENV} -c tensorboard-pvc-tb