Things we’ve learned about better software delivery principles through a pandemic
A presentation at SPOTCon 2022 by Jeremy Meiss
Learnings on Better Software Delivery Principles Through a Panini
Pixabay via athree23 2
3
“
performance described vs performance derived
Jeremy Meiss Director, DevRel & Community
Dataset 257 mil+ 44,000+ 290,000+ 1,000x workflows orgs projects Larger than surveys 8
Four classic metrics Deployment frequency Lead time to change Change failure rate Recovery from failure time
CI/CD Benchmarks for high performance teams Suggested Benchmarks Throughput The average number of workflow runs per day Duration The average length of time for a workflow to run Mean time to recovery The average time between failures & their next success Success rate The number of successful runs / the total number of runs over a period of time Merge on any pull request 10 minutes Under 1 hour 90% or better on default branch
11
The Data
Photo by: Matthew Henry 13
Throughput the average number of workflow runs per day 14
Throughput
Throughput ~ Mar/Apr 2020 16
Throughput 17
Most teams are not deploying dozens of times per day
Goal: Continuous validation of your codebase via your pipeline
Image by Pawan Kolhe from Pixabay
Duration Image by Pawan Kolhe from Pixabay the length of time it takes for a workflow to run 21
Duration Image by Pawan Kolhe from Pixabay 22
Duration Image by Pawan Kolhe from Pixabay ~ Mar/Apr 2020 23
Duration Image by Pawan Kolhe from Pixabay 95th Median ~ Mar/Apr 2020 24
Photo by Brett Sayles from Pexels
Mean time to recovery average time between a pipeline’s failure and its next success
Mean time to recovery shortest MTTR ∝ Duration
“…the most robust — and certainly the fastest — solution to a broken build is to simply revert the offending commit, allowing troubleshooting to happen in a way that doesn’t interfere with the rest of the team. You can’t know whether a new build works or not unless you’re starting from a known good position, which means you should never allow a new build to start on a red build unless it’s explicitly designed to fix it, and it’s hard to imagine a commit more likely to fix a broken build than simply reverting the one that broke it to begin with.” - Brandon Byers, Head of Technology, NA @ Thoughtworks Photo by Brett Sayles from Pexels 28
Recovery Time ~ Mar/Apr 2020
Recovery Time ~ Mar/Apr 2020
Recovery Time ~ Mar/Apr 2020
Recovery Time ~ Mar/Apr 2020
Photo by Lukas from Pexels
Success rate The number of passing runs ÷ total number of runs over a period of time 34
Success rate ~ Mar/Apr 2020 35
Success rate ~ Mar/Apr 2020 36
Success rate ~ Mar/Apr 2020 37
Duration The average length of time for a workflow to run TTR The average time between failures & their next success 2019 (median) 2020 (median) This Year (median) Benchmark 3.38 min 3.96 min 3.7 min 5-10 minutes 52.5 55.11 73.6 min < 60 minutes 77% Average should be +90% on default branch 1.43/day As often as your business requires not a function of your tooling Success rate The number of successful runs / the total number of runs over a period of time 60% 61% Throughput The average number of workflow runs per day 0.80/day 0.70/day 38
Extra Insights
202x has been a year.
“Don’t deploy on Friday” is not a thing.
“Don’t Deploy on Friday” is not a thing ○ 70% less Throughput on weekends ○ 11% less Throughput on Friday (UTC) ○ 9% less Throughput on Monday (UTC)
Language shifts over the last few years 43
44
45
46
47
Vertical splits 48
Elite Performer validation 50th percentile on CircleCI fit into the “Elite performer” category on the 2021 State of DevOps report
2020 Report Full 2022 Report https://circle.ci/ssd2020 https://circle.ci/ssd2022 50
Timeline.jerdog.me Thank you. For feedback and swag: circle.ci/jeremy IAmJerdog jerdog /in/jeremymeiss
Since early 2020, software delivery has undergone quite a shift in process and practices and how we understand “high-performing engineering teams.” With anonymous data collected from over 1M+ devs, 45K+ orgs, and 250M+ workflows, we’ve found some interesting insights into better DevOps practices. In this session, we will seek to uncover what this unique dataset can tell us about high-performing teams, and steps to get there as a business.
Buzz and feedback
Here’s what was said about this presentation on social media.
