Jupyter+Spark+Mesos: An “Opinionated” Docker Image

Check "opinionated" Docker-based stacks for Jupyter, including one to combine Jupyter and Spark right out of the gate.

By Steve Moore, Spark Technology Center author; Founder and CEO, Physical Plant Theater

Project Jupyter has been steadily growing in popularity since it spun it off from IPython in 2014. It currently boasts over 200 committers and over 9000 commits to its notebook repository on github — not to mention 597 Jupyter-tagged questions on stack overflow. (See the post co-written by Peter Parente and Gino Bustelo for more about what makes Jupyter exciting.)

Despite this popularity and traction, there was a general feeling in the open-source community that Jupyter didn't play well with Apache Spark, in particular users had the impression that you couldn't access Spark through a Jupyter Notebook. That perception struck IBM's Emerging Technology (ET) team as strange. After all, Spark is like any other library that coders would want to use in a notebook.

The ET team reached out to the community to propose and discuss the idea of offering a set of Docker-based stacks for Jupyter, including one to combine Jupyter and Spark right out of the gate. The community took up the challenge and within a week the effort was underway. The different images combine various pieces for convenience. These came to be called "opinionated" Docker images since rather than keeping Jupyter perfectly agnostic, the images bolted together technology that the ET team and the community knew would fit well — and that they hoped would make life easier.


For the Jupyter+Spark "all-spark-notebook", Apache Mesos was added to do cluster management for Spark. Incorporating Mesos meant users could pull the image and reasonably expect to set up an enterprise-level architecture with it. Of course, they don't have to use Mesos if they don't want to; the Docker container still supports connecting to a Spark standalone cluster (as described here), but if users do decide to deploy Spark on Mesos, the container can simply connect to the Mesos cluster without extra configuration.

The Jupyter community knew that the images would be handy — and they hoped the images would be discovered and used by other coders — but they were in for a surprise. When the team checked back last month, they saw that the Jupyter+Spark+Mesos container had been pulled over 150K times. (It's currently up over 168K.) At first, they thought it might be a bot pulling the image again and again, but then noticed that the other images that had been posted weren't getting pulled nearly as much. In fact, the all-spark-notebook container was getting pulled ten times as much as any of the other posted images. A nice feeling. :-)

Now, with the help of the community, the containers continue to get updated and maintained. The Jupyter community invites you to pull the Spark image (or the others) to try them out. Feel free to post any issues:

Repo: https://github.com/jupyter/docker-stacks/tree/master/all-spark-notebook

Issues: https://github.com/jupyter/docker-stacks/issues

The Spark Technology Center makes contributions to Apache Spark and publishes research on Spark from the open source community: www.spark.tc