CSE 403 Topic: DevOps

DevOps is a software development methodology that emphasizes short iteration cycles and frequently deploying code. Some core technologies that enable DevOps are continuous integration (CI) and continuous delivery (CD). DevOps also emphasizes that software engineers should be responsible for their code throughout its lifecycle — not just writing the code, but deploying, bug fixing, and maintenance, too.

Continuous integration - automated tests that run on each commit to a software project - and continuous deployment/delivery - releasing software based on automated processes gated by CI - are common practice in industry and open-source development.

There are a number of CI/CD tools that integrate with code repositories like GitHub. Here are a few examples:

• Travis CI
• CircleCI
• Jenkins

CI testing is nearly universally considered to be a best practice in software engineering today - even if tests are expensive, running them on a server somewhere after code is checked in only costs machine time, and developers will eventually be alerted if their code broke the tests.

Interesting resources:

• an interesting blog post on the rationale for CD: http://timothyfitz.com/2009/02/08/continuous-deployment/
• classic academic paper on test case prioritization: https://dl.acm.org/citation.cfm?id=506205/
• classic academic paper on test suite minimization: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.114.7409&rep=rep1&type=pdf
• article on the benefits and challenges of adopting CD at a company: https://ieeexplore.ieee.org/document/7006384?reload=true&arnumber=7006384
• Useful blog post on DevOps culture: https://martinfowler.com/bliki/DevOpsCulture.html

Project ideas:

• what else could be part of the CI gate? Static analysis? Mutation testing? Something else? What changes to the design of one of the CI tools mentioned above would you need to make to support such a thing?
• what tests should run first in a CI setup? Could you prioritize tests so that those that are more likely to fail run first?
• running CI/CD servers for a large organization is expensive, especially because most tests don't need to be re-run on every commit! Could you design a CI tool that automatically re-runs only tests that might actually fail?