In this short article, I list the steps you need to follow in order to publish a Docker image to the Google Container Registry(GCR) in the Google Cloud Platform.

Pre-requisites

1. Docker installed
2. GCloud SDK
3. User or service account with access required to push to GCR.

Steps to push docker image to GCR

Create Dockerfile

In a previous article on how to install Tensorflow with Docker, I showed you how to create a Dockerfile that extends the default Tensorflow image:

How to setup Tensorflow on Ubuntu Linux with multiple GPUs using Docker

Let’s use that Tensorflow Docker image as the base container, and let’s publish this image to Google’s Container Registry:

FROM tensorflow/tensorflow:latest-gpu-jupyter

RUN pip install jupyterlab

Build and tag Image with docker build

docker build -t codemental/tensorflow .

Now I am going to assume that you want to publish to GCR directly from your Google account.

Login to GCloud

First, we log in:

gcloud auth login

If you would like to use instead a service account then we should call gcloud auth activate-service-account:

gcloud auth activate-service-account ACCOUNT --key-file=KEY-FILE

Docker Registry Login with Google Cloud

In order to push an image to GCR with Docker we run:

gcloud auth configure-docker

Tagging Docker Image

And then we give an appropriate tag to our image:

docker tag codemental/tensorflown gcr.io/<gcp-project-id>/tensorflown:0.1

Pushing Docker Image to gcr.io

To push to GCR.IO we simply have to do:

docker push gcr.io/<gcp-project-id>/tensorflown:0.1