Unblocking the database bottleneck in enterprise DevOps deployments
In this guest post, DevOps consultant and researcher Nicole Forsgren, PhD, shares her advice on what enterprises can do to overcome the database problem when scaling up their DevOps endeavours.
High performing DevOps teams that have transformed their software delivery processes can deploy code faster, more reliably, and with higher quality than their low-performing peers.
They do this by tackling technical challenges with effective and efficient automation, adopting processes drawn from the lean and agile cannon, and fostering a culture that prioritises empathy and information flow.
A key to these successful transformations is fast feedback loops, and integrating key stakeholders early in the development and delivery process.
In the earliest iterations of DevOps, the first key stakeholder was IT operations: taking feedback about maintainability and scalability of your code. You could frame it this way: learn why IT operations so often put up barriers to code deploys, address that feedback sooner in the pipeline, and continue to work together.
Once dev and ops are humming along smoothly, many teams find they still hit bottlenecks in their deployments, particularly as they scale. And the trend I’m hearing more and more often is that this bottleneck is happening at the database.
Unfortunately this is not a challenge you can turn a blind-eye to. It’s not going away anytime soon and the tooling you are using to automate the build and deployment of your apps will not solve this problem.
No matter how fast you can get your application releases going, it’s more than likely they’ll be waiting for changes to the database before it can get shipped. There’s no time like the present to bring data into DevOps as the next move toward shifting left.
The devil’s in the database
Teams and organisations are humming along and delivering code. Things start to get more exciting as applications start to scale. Under the covers, however, the database is a shared resource across dev, test, and production.
In addition, the database release process is usually manual, making it a carefully orchestrated process that is slow, risky, and prone to errors. The database administrators (DBAs) guard this process carefully, and with good reason.
Early on, performance is acceptable, but as your application continues to grow and scale, this database release process gets more difficult. Or perhaps your application is already at scale, and you are shipping code faster, when suddenly you find the application code outpacing your database schema changes. “Brute force” only works so long before your DBAs are burning out and just can’t keep up.
When faced with more requests for changes to the database than are feasible to do or safe to do, the DBAs by protecting their resource and saying “no.” Suddenly, database changes can take weeks, when the competing software releases are using continuous delivery practices and pushing to production daily or even hourly.
Addressing the issue
Solving the database constraint in DevOps takes a few forms, and includes culture and tools.
Let’s start with culture. You’ll want to start shifting your database work upstream into the development phase and find problems before they get into production. This is similar to the “shift left” emphasis we’ve seen in other critical areas that are often left to the end of the delivery pipeline, like security.
To truly shift data left allow your engineers to follow the same process they use today for the app. Check your change scripts into source control, run automated builds, get notified when database changes ‘break the build’ and provide your engineers with the tooling they need to find and fix database problems long before they become an issue in production.
As a database professional, give your engineering teams the tools they need to treat database code just like app code. This will equip them to know when the database is the problem, and teach them where and how to look for it. This work is important and can be shared upstream. These practices elevate DBAs to valuable consultants, not downstream constraints.
This may feel foreign at first, as the role of the DBA expands from building tools to collaborating and consulting across whole teams of people.
Where they were once protective and working on technology alone or with only other DBAs, they are now open and consultative with others. Where their work with the database was once the last stop in the chain of application release automation, this work is (hopefully) triggered much earlier in the pipeline.
Done right, this approach frees the DBA team from reviewing every change in every change script, so they can spend their valuable time on more important tasks such as; patching and upgrading, performance tuning, data security, and capacity planning.
Expanding the reach of a DBA
Engineering teams usually start with just a few DBAs, so you have to scale the person — and this is done with technology and automation.
This helps teams and organisations work faster, protect their data, and increase productivity. Just as with any other part of the DevOps process, automation speeds up our work: As the application and the databases start to scale, DBAs will find themselves needing to scale their services.
Deploying 100 database servers is much faster (and more effective) with scripts compared to doing it manually.
Rolling database schema changes is a high-risk move; using tooling and technology helps mitigate this risk in two ways: First, by introducing and testing these changes faster in the application development pipeline, you discover errors sooner, allowing teams to find and fix changes before they hit production.
Second, using automation provides traceable steps and verification if any steps need to be repeated or reversed. Third, automation increases productivity by creating repeatable processes that allow you to manage production and database schema migration.
When you’re developing and delivering code with speed and stability, but you start to hit a database constraint, don’t panic. Just apply the same DevOps principles you used for software: focus on bridging culture with your database team, improving and streamlining process with your database team, and leveraging a smart investment in tooling and automation.
Many thanks to Silvia Botros and Darren Pedroza for sharing their thoughts and experiences with databases in DevOps transformations. I would also like to thank Camille Fournier for reading an early draft of this post.
Also, for more detail, I suggest you check out Database Reliability Engineering, by Laine Campbell and Charity Majors. It is essential reading for those digging into databases and database administration in DevOps today.