OK. I got the tests green and I decided to proceed in the very minimally impacting way - just to see how it works and so that we can learn from it.
For now I disabled the "pull request" trigger - so PRs should not be affected. The tests will run on: * canary runs on schedule (shifted by 3 hours from the regular AMD schedules) - including failure notifications on slack * when we merge to any of the v*test branches * when we trigger it manually with workflow-dispatch This should not have an impact on regular PRs and we will have time to observe stability and fix issues in canary if we find them.. We can continue discussion here if anyone has other comments, but this will allow us to observe and see how it goes. If we see things are stable, we will turn off the experimental status of the ARM image finally. J. On Mon, May 5, 2025 at 2:32 PM Jarek Potiuk <ja...@potiuk.com> wrote: > Go back - just a few tests are failing on MySQL ... we have ARM containers > there :) > > On Mon, May 5, 2025 at 2:30 PM Jarek Potiuk <ja...@potiuk.com> wrote: > >> Ah.. And MySQL tests will have to be skipped - until we switch to ARM >> containers it seems. >> >> On Mon, May 5, 2025 at 2:26 PM Jarek Potiuk <ja...@potiuk.com> wrote: >> >>> Hello here, >>> >>> Following the recent change in ARM image building, I attempted to >>> convert our CI to allow both AMD and ARM tests. >>> >>> This works rather easily (and our CI framework got a bit more generic >>> this way) - https://github.com/apache/airflow/pull/50198 . >>> >>> There are a few tests that we will have to skip on ARM because some >>> libraries are missing for some providers, but there are just a few of those. >>> >>> Running those tests on ARM for me was the condition to remove >>> "experimental" status of our ARM image and I think that would be a good >>> thing. >>> >>> I do not think we should run it in all PRs and likely we do not even >>> have to do it, but I think it would make sense to run it as part of our >>> "canary" builds. We already get notified when such "canary" tests are >>> failing in main (we run them every 6 hours now). >>> >>> It will increase cost for the ASF (but not 2x more like 1.2x) and >>> decrease stability a bit I guess but I think it might be worth it. >>> >>> WDYT? >>> >>> J. >>> >>>
