Install Oracle Instant Client

[calumabarnett]

This PR will install Oracle Instant Client (19.6.0.0.0). This will enable analysts to directly query Oracle databases using the cx_Oracle package. This is in support of an interim measure to provide access to the Delius database on the Analytical Platform while work on the pipeline is ongoing.

[xoen]

As said in Slack, with the limited knowledge of this Oracle thing this seems fine to me.

I'd advice you to try this on your own JupyterLab instance for awhile before mass-releasing to all users as there are few changes in the environment which could cause compatibility problems to the users (e.g. the base image change so probably quite few system libraries and potentially version of Python and conda, etc...so proceed with caution ☺️ )

[calumabarnett]

The image has built OK and I've managed to deploy it to my own JupyterLab instance with a couple of small tweaks to the Helm chart and it looks to be working on the face of it, though I'm probably not a good person to test further.

Once the database connection is created I can deploy it some other users for further testing.

[davidread]

@calumabarnett We talked about making this image with the Oracle client available only to limited users, who are manually given this image - see https://docs.google.com/document/d/1ISoxf47qF3l0VqBo3d-yY4SekCQoBmmGiqlnVowtoaE/edit#
How about we achieve this in a similar way to the all-spark image, by putting the Dockerfile in a separate directory in this repo? It just needs someone to setup Quay to build it and then it's available for users to specify it in their Deployment.

[calumabarnett]

Thanks @davidread – sounds like a sensible approach. I've reverted the datascience-notebook Dockerfile, moved the updated Dockerfile to a new directory called oracle-datascience-notebook and created a new Quay repo of the same name to build from this directory.

[davidread]

👍 Seems pretty straightforward. Nice to have these oracle bits together in the Dockerfile. The rest is a copy of the other Dockerfile, but I think it's worth a bit of DRY to have a separate image built, so that this functionality is not made available to most users.

[davidread]

That build error about missing [jobs] suggests the syntax in the https://github.com/ministryofjustice/analytics-platform-jupyter-notebook/blob/master/circle.yml needs a little update, if you can work that out too before merge?

[xoen]

Actually the comment about the repetition as I wonder whether this Dockerfile could use the standard JupyterLab image as BASE and install this on top of that to make it clear it's the only difference.

@calhutchison sorry I didn't think/suggest about this before 😬

Basically it's similar to what they do in the official R/Jupyter images where they have few common/base images and they have more specialised images which add/inherit on top of these images to add functionalities.

It makes a lot of sense as it clear what's special about this docker image as its Dockerfile would contain only the difference.

Just a thought.

[calumabarnett]

That build error about missing [jobs] suggests the syntax in the https://github.com/ministryofjustice/analytics-platform-jupyter-notebook/blob/master/circle.yml needs a little update, if you can work that out too before merge?

@davidread It's a very old v1 Circle config. I'm not sure it's used/needed anymore anyway as all PRs trigger a build on Quay directly. Perhaps what's needed is just to remove it?

[davidread]

@calumabarnett Makes sense. I've removed the Circle config in this PR: #31

[davidread]

This is good to merge now I think

[calumabarnett]

@xoen that does make sense – I've quickly tried to build from the base image but have run into some errors with apt. I'm sure they could be fixed but given we have something working and I don't think the benefits are huge, I'm inclined to stick with things as they are.