Deploying a custom build of Datalab to Google Cloud platform

Question

For a project we are trying to expand Google Cloud Datalab and deploy the modified version to the Google Cloud platform. As I understand it, the deploying process normally consists of the following steps:

1. Build the Docker image
2. Push it to the Container Registry
3. Use the container parameter with the Google Cloud deployer to specify the correct Docker image, as explained here.

Since the default container registry, i.e. gcr.io/cloud_datalab/datalab:<tag> is off-limits for non-Datalab contributors, we pushed the Docker image to our own container registry, i.e. to gcr.io/<project_id>/datalab:<tag>.

However, the Google Cloud deployer only pulls directly from gcr.io/cloud_datalab/datalab:<tag> (with the tag specified by the containerparameter) and does not seem to allow specification of the source container registry. The deployer does not appear to be open-source, leaving us with no way to deploy our image to Google Cloud.

We have looked into creating a custom deployment similar to the example listed here but this never starts Datalab, so we suspect the start script is more complicated.

Question: How can we deploy a Datalab image from our own container registry to Google Cloud?

Many thanks in advance.

Answer 1

The deployment parameters can be guessed but it is easier to get the Google Cloud Datalab deployment script by sshing to the temporary compute node that is responsible for deployment and browsing the /datalab folder. This contains a runtime configuration file for use with the App Engine Flexible Environment. Using this configuration file, the google preview app deploy command (which accepts an --image parameter for Docker images) will deploy this to the App Engine correctly.