Azure is stuck with "This agent request is not running because you have reached the maximum number of requests that can run for parallelism type", restarting it did not help.

Azure is stuck with "This agent request is not running because you have reached the maximum number of requests that can run for parallelism type", restarting it did not help.

There's a massive backlog, but I don't see how to cancel all running jobs at once.

Going to merge this, before it creates some problems. That hanging test is weird, but not really worrisome. Thanks again!

The azure pipeline is also trying to read master. I edited the pipeline yml, lets see if that helps. If not, the easiest fix may be to delete the old pipeline and make a new one.

Wheel builds will need fixing, they currently keep a submodule with the numpy master branch.

We should probably rename to main in numpy-wheels as well. It seems much cleaner if the branch it builds has the identical name (I do not have access to rename) – otherwise have to carefully distinguish whether the occurance of "master" refers to the wheel repo or the numpy one, making the change more than a trivial replace.

@seberg I'm going to fix the wheel build in two steps

• Fix the yml files
• Do the rename

The first step is running now, we will see how it goes. The NumPy submodule should be OK, it points to a hash. I checked that I could also checkout main from numpy after doing git submodule sync --recursive. I'm going to think a bit as to how to do the branch rename in order to avoid the current Numpy problems.

Turns out that the job limit applies to all the pipelines, so the azure wheels build is also stuck waiting.

Microsoft isn't letting me log into azure easily (claiming my private email is a job email). It seems it restarted the jobs for all PRs on azure, not sure if we can clear the whole pipeline somehow? I guess on wheel repo that shouldn't trigger hundrets of jobs at least...

Some jobs are running, so maybe we just need to wait.

The US has had several days of degraded azure performance: https://status.dev.azure.com/_history. Seems to be a azure problem.

Thanks for helping with the loose threads @seberg @charris @rossbar

I'm going to suggest that if this is done for SciPy that it might be a good idea to disable running tests on PRs before making the switch, otherwise azure will probably spend a couple of days rerunning tests on all the outstanding PRs.

Or maybe set up a second pipeline and point it at "main". I'm not quite sure what we did to get all the retests.