Hi Jarek, This seems like a very good performance improvement.
It would be interesting to see how it translates in terms of average E2E runtime for the full CI workflow (honestly CI run times are currently a significant damper on contributor experience). Looking forward to less waiting for CI to turn green (or fail). Thanks, Sameer. On Sun, 14 Jun 2026 at 11:27, Jarek Potiuk <[email protected]> wrote: > Hello here, > > Another day - another performance improvement in tests. > > I am helping Shahar to test and optimise the self-hosted runner > experience to make us independent of the public GitHub Runner queue. > > Since test parallelism of our tests is **crucial** for faster and > cheaper - I finally tackled the problem of "standard" provider DB > tests taking by far longest "serialized" time (combination of the > tests being "DB tests" and creating/removing a lot of venvs in the > process. > > https://github.com/apache/airflow/pull/68533 - looks small but I > **hope** it should solve the problem permanently and improve test > speed even without self-hosted runners. > > It changes all PythonVenv/ExternalEnv tests to be "non-DB tests," > meaning they no longer have to run serially. Consequently, we avoid > artificially splitting the "standard" provider into smaller > "sub-tests," as all these tests will now run in parallel using Xdist. > This means parallelism scales automatically and naturally with size of > the runner we will run it on. > > J. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
