When attempting to run the test harness, I receive the following:

Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "C:\Public\Devel\cpython\main\Lib\test\__main__.py", line 2, in <module>
    main()
    ^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\main.py", line 736, in main
    Regrtest().main(tests=tests, **kwargs)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\contextlib.py", line 155, in __exit__
    self.gen.throw(typ, value, traceback)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\contextlib.py", line 155, in __exit__
    self.gen.throw(typ, value, traceback)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\support\os_helper.py", line 396, in temp_dir
    yield path
    ^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\contextlib.py", line 155, in __exit__
    self.gen.throw(typ, value, traceback)
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\support\os_helper.py", line 427, in change_cwd
    yield os.getcwd()
    ^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\support\os_helper.py", line 449, in temp_cwd
    yield cwd_dir
    ^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\main.py", line 658, in main
    self._main(tests, kwargs)
    ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\main.py", line 704, in _main
    self.win_load_tracker = WindowsLoadTracker()
                            ^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\win_utils.py", line 41, in __init__
    self.start()
    ^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\win_utils.py", line 70, in start
    counter_name = self._get_counter_name()
                   ^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Public\Devel\cpython\main\Lib\test\libregrtest\win_utils.py", line 90, in _get_counter_name
    system = counters_dict['2']
             ~~~~~~~~~~~~~^^^^^
KeyError: '2'

This is due to my machine missing the localized names for the performance counters.  Other performance monitoring tools operate just fine.

While I have been working around this issue for some time, it has become difficult to seperate the workarounds from actually changes in the test harness.

The PR (https://github.com/python/cpython/pull/26578) from https://bugs.python.org/issue44336 also solves this issue by accessing the counters directly instead of relying on their localized names.

OK, I know it has been a busy month for Python, but this issue is really hampering my bug fixing efforts.  It makes the complete regrtest useless for me.  I am required to run each affected test directly so it is possible to miss side-effects of changes made.

The original PR is now 9 months old.

You're closer to this stuff than I am, so if you say it works for your scenarios and it doesn't break the existing ones, I'll merge it.

Do we need to backport to 3.8 at this point?

With 3.8 so close to security only, I would doubt it is worth it anymore.  I've just run the PR against HEAD and it still works as is, so should be good to go.

"So close" being some months ago ;)

I need to retrigger address sanitiser so that it passes before merging, but after that, I'll merge (and hopefully the backport is clean).

Copying Eryk Sun's comment here for posterity:

This would be less fragile and more useful in general if support were added to _winapi for PdhOpenQueryW(), PdhAddEnglishCounterW(), PdhCollectQueryData(), PdhCollectQueryDataEx(), PdhGetFormattedCounterValue(), and PdhCloseQuery(). The English counter name is "\System\Processor Queue Length" -- no need to worry about localized names, hard-coded values, or hard-coded struct sizes and offsets.

I was just wondering whether it's worth implementing it using the API. To be clear, I wasn't implying to hold off on applying Jeremy's PR. The existing code is causing him problems, and he has a working solution.

Jeremy - backports need to be done manually. Can I leave that with you and just ping me on here when you're ready for a merge?

Backports state that they are ready...  I'm just a little uneasy as I've never used cherry_picker before.  3.10 went smooth, but 3.9 required manual merging.

Backports have been merged