We Solved DevOps. What’s Next?
A presentation at DevOps Vision 2024 in in Clearwater, FL, USA by Baruch Sadogursky
We solved DevOps. What’s next?
Baruch Sadogursky - @jbaru ch × Developer Productivity Advocate × DevRel Advisor for Gradle Inc × Development -> DevOps -> #DPE
shownotes × × × × speaking.jbaru.ch Slides Video All the links!
A decade of DORA x Deployment frequency x Change lead time x Change fail rate x MTTR
Puppet dora report 2015
Accelerate Dora report 2018
Google cloud DORA report 20 24
Cluster changes over time
We figured the “devops” shit out. Done.
Lead time over time
Why does lead time decline?!
Dev Ops DevOps QA
Process People DevOps Tools
Productive Process Productive People DevOps Productive Tools
Empowered Teams ?!?! DevOps Kubernetes
Production system of your production systems
The pain is real
Developer Productivity Engineering DevOps Kubernetes Empowered Teams
Developer productivity Enginee ring!
Developer Productivity Enginee Foster Faster Feedback Collaborate through Effective Tooling Eliminate Toil for Developers Dedicated Organizational Mindset ring Embrace Rigorous Observability for Proactive Improvement Prioritize Automation and Eliminate Bottlenecks Outcomes Over Output
Talk is cheap, show me the goods! 24
Small DPE improvements make a × × × × × huge difference Generate code faster: Better IDE Test better: Testcontainers Enforce better code: Sonar Test more reliably: Flaky test detection Foster Faster Feedback:
feedback efficiency × × × × IDE: Sub-seconds (I type, it marks it red) Build: Seconds CI: Minutes Production: Hours/Days
Reverse dependency on distance from developers IDE Build CI Production Feedback Time Faster Slower Distance from Developers Expected Real
It is slow!
It is slow and the developers h ave no idea why!
What is build? × × × × × Project setup Downloading the Internet Artifact generation: Compilation, packaging, etc Tests Artifact deployment
What can go wrong? × × × × × Project setup Downloading the Internet Artifact generation: Compilation, packaging, etc Tests Artifact deployment
When can it go wrong? At any time.
The Build frustrates the developers
Let’s ask Chatgpt
What the actual f*ck?! × Skipping tests defeats the purpose of the build! × How about skipping compilation? × We want faster feedback, not less feedback
What feedback do we want?
Ci/cd pipeline quality gates It compiles Basic Integration tests Basic Unit tests Basic Quality Linting Quality Code coverage Dependency scanning SAST/DAST Quality Sec Static code analysis Sec Sec Secrets scanning Resource Utilization Nonfunc Nonfunc Load Testing Nonfunc Compliance
Two types of feedback x e.g., CI/CD x we never wait for it x results are distracting x e.g., build x we’ll wait for it in the flow x we’ll be pissed off when it’s slow
Reverse dependency on distance from developers Faster IDE Build CI Feedback Time Commit time synchronous asynchronous Slower Distance from Developers Production
Ideal build time feedback It compiles Basic Integration tests Basic Unit tests Basic Quality Linting Quality Code coverage Dependency scanning SAST/DAST Quality Sec Static code analysis Sec Sec Secrets scanning Resource Utilization Nonfunc Nonfunc Load Testing Nonfunc Compliance
Delightful build (pick two): ☑ provides max feedback ☑ fast
Skip what can be skipped (but no more!)
Avoidance: Incremental build × Don’t build what didn’t changed × Don’t build what isn’t affected
Avoidance: Incremental build s hortcomings × Relies on produced artifacts × Relies on architectural decisions
Avoidance: Caching × × × × Makes the build faster Makes the build faster for everybody Makes the build faster always Makes all parts of the build faster
Avoidance: Predictive test selec tion × Learns code changes effects de-facto × Skips tests with high degree of confidence
How test prediction works × Code changes and test results are thrown into learning model × After a while, the model predicts which changes fail which tests
Test prediction What changed Where it changed Correlate with observed test failures Predictions which changes will fail which tests
Black magic in action × The more tests a project has, the less they break × Refactorings in Java break tests less than in JavaScript
Speed up what can’t be skipped
Test parallelization × Use max power of local machine × (Yes, your boss should buy you the bleeding edge)
Test distribution × × × × CI uses fan-out to speed-up tests Shouldn’t you enjoy it for local tests? Use the cloud to distribute test load RUN ALL THE TESTS!
Why not just using ci fan-out? × Relying on shared CI infrastructure × CI infrastructure is not optimized for real-time feedback! × Are the agents as fast as they can be?
Don’t let it slide
Observe and improve × Measure local build times across time and environments × Detect downfacing trends × Find root causes and improve
The gains are real!
Developer productivity is the n ext frontier × We figured out (most of) DevOps × If you want to excel in your production environment, you know how × But what about your path to production? × There is work to be done there. × DPE is the way to go!
Learn more and try it today! × × × × Take the Gradle/Maven Speed Challenge! Be DPE Agent of Change! Read the DPE Handbook! Watch the DPE Summit keynotes! x speaking.jbaru.ch
THANKS! Q&A and Twitter X/Bsky/Mastodon/LinkedIn ads: x @jbaruch x #DevOpsVision x speaking.jbaru.ch
Hear me out: We solved DevOps. I am not saying that everyone is doing it right (or even trying), but we have the solution if you want to do it right. Here is the theory, the technology, the communities, the docs, and the books—go knock yourself out. Do DevOps: empower your teams, make deployments a breeze, build invincible CI/CD pipelines, and create production environments that are secure, stable, and observable. Done.
Now, in perfect alignment with DevOps and the Theory of Constraints, it’s time to turn our gaze to the next bottleneck: the production environments of our production environments. They’re a mess. We accept soul-sucking toil, slow and unstable tests and builds, and constant distractions as the norm, just as we accepted the insufferable, manual “throw garbage between silos” all-night deployments of the pre-DevOps era. But, as with DevOps, we shouldn’t take these for granted. It’s time to rise again and resist our development environments’ “Pre-DevOps” state. Let’s engineer better developer productivity for all!
Video
Resources
The following resources were mentioned during the presentation or are useful additional information.
-
DevOps Vision 2025 - Register Today And Save $500 of Early Bird Pricing
Use promo code dv5barw
-
Baruch’s Book “The Liquid Software”
-
State Of DevOps Report 2015
-
Baruch’s Book “DevOps Tools for Java Developers”
Get it signed by Baruch at DevOps Vision 2025!
-
Baruch’s Book “The Liquid Software”
Get it signed by Baruch at DevOps Vision 2025!
-
State Of DevOps Report 2018
-
State Of DevOps Report 2021
-
State Of DevOps Report 2024
-
The Problem with Gradle (any build tool actually) article by Bruce Eckel
-
Gradle Build Cache
-
Gradle Remote Cache Manual
-
Meta’s Predictive Test Selection Research
-
Develocity Predictive Test Selection
-
Spring Project PTS Simulations
-
Gradle Parallel Execution
-
Develocity Test Distribution
-
Develocity Reporting and Visualisation
-
Developer Productivity Engineering Handbook
-
Get Started with DPE University
-
DPE Summit 2024: Watch the Keynotes!
