Information for system integrators
These notes are a work-in-progress. This page provides an overview as guidance only, and not a series of definite instructions. We plan to further expand on this information.
This page is aimed at people who want to run their own installation of OpenSAFELY as a proof-of-concept trial prior to integrating into the live system.
These people are likely to have one of the following professional roles:
- software developers
- clinical data providers
- system integrators
It is taken that you have some familiarity with OpenSAFELY.
If not, you should first refer to:
The OpenSAFELY technical architecture diagram shows all of the platform software components and how they interact.
The specific steps required to create a minimal setup are:
Deploy a job runner within your secure environment. A minimal configuration simply runs Docker containers and stores any resulting container output on a local disk.
Deploy a job server that the job runner polls for jobs that end users request to be run.
It is possible to use our existing instance of this server at our jobs site; contact us if you would like us to configure this for you.
Create a secure network with our GitHub proxy. This provides access to repositories with research study code to run and Docker images used to run the code.
To provide access to your database from within your setup, integrate into our cohort-extractor ETL tool:
- via an implementation of a backend interface; this is an example for TPP
- and, if you are using an as-yet unsupported database, a database connector
A simpler replacement for cohort-extractor, DataBuilder, is in development.
You may prefer to wait for a stable version of DataBuilder before attempting to integrate.
Releasing job outputs requires:
- the release-hatch tool for reviewing outputs
- the output-publisher tool for publishing outputs
We use Ubuntu in our deployments. Our deployments use deploy scripts that you can refer to.
See our support page for details of how to get more assistance on integration.