On Fri, Apr 3, 2026 at 12:30 PM Jakub Kicinski <[email protected]> wrote: > > The test seems flaky, we hit this twice in the last 24h:
Hmm it waits for the active port to change before checking the packet counts, but that appears insufficient. I will make it wait for the newly inactive link to stop receiving traffic, using a generous polling timeout, which should fix this race condition. > What's even worse the retry of the test then fails like this: When starting up teamd, it seems to check for a pid file, which persists after using sigkill on the previous test run. So if we don't kill teamd gracefully, then this file must be deleted manually. It's strange that it didn't cause any problems on my end, but I can fix it in the next iteration.
