In this series on DevOps for Data Science, I’ve explained the concept of a DevOps “Maturity Model” – a list of things you can do, in order, that will set you on the path for implementing DevOps in Data Science. The first thing you can do in your projects is to implement Infrastructure as Code (IaC), and the second thing to focus on is Continuous Integration (CI). However, in order to set up CI, you need to have as much automated testing as you can – and in the case of Data Science programs, that’s difficult to do. From there, the next step in the DevOps Maturity Model is Continuous Delivery (CD). Once you have that maturity level down, you can focus on Release Management. And now we’re off to the next Maturity Level: Application Performance Monitoring (APM).
Now that the creation, testing and delivery of your organization’s solution is automated (along with the Data Science part of that application) you need a method to make it faster, and more efficient. Once again, starting at the very beginning of the process, the Data Science team should think about building performance monitoring and reporting directly into the code. This actually isn’t as hard as you might think – simply take a methodical approach to performance just as you would the predictive or classification algorithm itself.
Start with the basics – instrumentation. Your code should have at least three levels of monitoring down to a central log file or metrics collection system:
- Minimal – Log the time the prediction/classification started, and when it ended
- Standard – Log the start and stop times of the event, the name of the event, and the calling function(s)
- Debug – Log the start and stop times, the name of the event, the calling function, as much of the call-stack as you can securely record, and any information you can securely and legally gather on the user and user environment
Next, devise a method to view the collected metrics. Finally, do a little Data Science over that data – standard reporting, fivenum summaries, and even predictive and classification work to optimize the system.
And that brings us to the next point – if you think about performance monitoring and management at the beginning of the process, and then build it in as you go, you can use those metrics in analyzing your automated tests, including stress-testing the system. It’s a virtuous cycle, and the very point of DevOps.
To be sure, there are more formal methods and concepts you should study to fully implement APM – there’s a good reference on that here: https://dzone.com/articles/what-is-application-performance-monitoring-apm-app . But if you simply metric and monitor the application, you’ll have a great start to this process. Want a complete series on APM? Here’s one that introduces Application Insights, which simplifies this process: https://channel9.msdn.com/Series/DevOps-Fundamentals/Application-Performance-Monitoring-and-Availability-Monitoring
See you in the next installment on the DevOps for Data Science series, where I’ll cover the next level in your DevOps Maturity Model for Data Science teams.
For Data Science, I find this progression works best – taking these one step at a time, and building on the previous step – the entire series is here:
- Infrastructure as Code (IaC)
- Continuous Integration (CI) and Automated Testing
- Continuous Delivery (CD)
- Release Management (RM)
- Application Performance Monitoring
- Load Testing and Auto-Scale
In the articles in this series that follows, I’ll help you implement each of these in turn.
(If you’d like to implement DevOps, Microsoft has a site to assist. You can even get a free offering for Open-Source and other projects: https://azure.microsoft.com/en-us/pricing/details/devops/azure-devops-services/)
- Need a quick introduction to DevOps? Check out this series: https://channel9.msdn.com/Series/DevOps-Fundamentals
- Here’s a complete, full course on DevOps on the Microsoft Virtual Academy – https://mva.microsoft.com/en-us/training-courses/devops-with-visual-studio-team-services-and-team-foundation-server-16779#!