Drop usage of WhoHas & WhatHas from Client

[jakirkham]

PR ( #4853 ) ran into some issues, which was largely reverted by PR ( #4860 ). However there are some lingering issues, this tries to revert the remaining pieces.

cc @dantegd @quasiben @jrbourbeau

• Closes #xxxx
• Tests added / passed
• Passes black distributed / flake8 distributed / isort distributed

[jrbourbeau]

I think we need to comment out / revert these checks too

distributed/distributed/tests/test_client.py

Lines 3633 to 3634 in 93e2869

	assert type(who_has) is WhoHas
	assert type(has_what) is HasWhat

[jakirkham]

Thanks marked that xfail. Guessing we will want to come back and straighten this out after the release. Though if we want to change more, happy to do so.

FWIW here's a CI log where we were seeing this issue, which traces back to here in cuML:

>       return WhoHas(result)
07:29:48 E       TypeError: 'coroutine' object is not iterable

Dante and I have been trying to reproduce with variations of this:

from time import sleep
from distributed import Client

def inc(i):
    sleep(60)
    return i + 1

async with Client(asynchronous=True, n_workers=1) as c:
    f = c.submit(inc, 0)
    d = await c.who_has(f.key)
    print(d)

However we haven't quite narrowed down a reproducer. Guessing something like that would trigger the error. Though there is likely something simple we are overlooking.

Sorry this isn't more helpful. Hopefully that at least gives some context though 🙂

[jrbourbeau]

Hmm that's interesting, it looks like self.sync(self.scheduler.who_has, keys=keys, **kwargs) is returning a coroutine even though self.asynchronous is False-y. I wonder if using inspect.isawaitable instead of self.asynchronous would be a more robust check

[jakirkham]

Yeah it's odd. Also confused as to why we haven't been able to reproduce it. That makes sense. I think for now we are ok just reverting, but happy to try that after the release

[jrbourbeau]

The changes here look good for getting the release out today. @jakirkham could you remove the draft status of this PR when you're ready for it to be merged?

[jakirkham]

Should be good to go. Sorry we were just testing on our end to make sure things were cleared up

[jakirkham]

FWIW this CI log shows the issue is fixed 🎉