Deploying TensorFlask on Kubernetes Engine

• A Dockerfile for TensorFlask

• A set of scripts to deploy the resulting Docker image to GCP Kubernetes Engine

Files

• build-cloud-container-builder.sh

  • script to build the Dockerfile via the Google Cloud Container Builder

• cloudbuild.yaml

  • declarative build instructions to build the Dockerfile via the container registry
  • an alternative to build-cloud-container-builder.sh, to be used via a build trigger (see below)

• common.sh

  • shared environment variables

• deploy-on-kubernetes-engine.sh

  • create a TensorFlask deployment and LoadBalancer service on an existing Kubernetes Engine cluster
  • requires (and checks) that the current $IMAGE_NAME is available on gcr.io for the current project

• Dockerfile

  • the Dockerfile for TensorFlask
  • automatically runs the server on container start, unless overriden
  • exposes port 8000

• run-client-locally.sh

  • runs the TensorFlask client locally against the deployed and exposed tensorflask instance
  • requires (and checks) that the LoadBalancer service is up and running

General flow

# Create cluster and get credentials first, see ASCIIcast

# Wait a few minutes between each step. The scripts will check for their prerequsites

# Instead of build-cloud-container-builder.sh, you can also use the gcr.io build triggers
./build-cloud-container-builder.sh

./deploy-on-kubernetes-engine.sh

./run-client-locally.sh

Don't forget to clean up your services, deployments and clusters.

Build image automatically via Cloud Container Builder

• Go to https://console.cloud.google.com/
• -> "Container Registry"
• -> "Build Triggers"
• Add a new build trigger and choose to use cloudbuild.yaml for build configuration

Deployment via cloudbuild.yaml is possible, but not configured right now.

ASCIIcast of some concepts used here