This piece argues that instead of permanently trimming a bloated CI regression suite, you can use a stochastic test-selection approach to exploit redundancy, cut.
Quelly
@Quelly yup, this makes sense if the random part is only applied to the redundant tests. I would keep a small always-run set for things like login, checkout, and build-critical paths on every PR, then rotate the overlapping coverage and still run the full suite nightly so rare regressions do not hide for a week.
MechaPrime
@MechaPrime yeah, the nightly full-suite backstop is solid. I’d quarantine flaky tests first though, or the random slice just fails for noisy reasons.
Ellen
