[cookie-store] Improve cleanup logic

[jugglinmike]

While researching for a WPT infrastructure project, I found these tests to be behaving erratically. In a handful of cases, the cleanup invocation was not properly awaited. Since we've been meaning to take advantage of the new "async cleanup" functionality, I took the opportunity to use that API. I'm submitting this in two commits to make the bug fix easier to identify.

To validate this change set, I ran the tests in Chromium (70.0.3538.77) on master (3960fff) and on this branch via the command

 $ xvfb-run --auto-servernum ./wpt run --no-pause --log-mach - --log-mach-level DEBUG --channel dev --no-manifest chrome cookie-store/

The results were identical:

On master:

web-platform-test
~~~~~~~~~~~~~~~~~
Ran 338 checks (35 tests, 303 subtests)
Expected results: 328
Unexpected results: 10
  subtest: 10 (10 fail)

With patch applied:

web-platform-test
~~~~~~~~~~~~~~~~~
Ran 338 checks (35 tests, 303 subtests)
Expected results: 328
Unexpected results: 10
  subtest: 10 (10 fail)

[jugglinmike]

It looks like these tests are still unstable even with my patch applied. I tried to reproduce the instability locally using the following command:

./wpt run --no-pause --log-mach - --channel dev --verify chrome cookie-store

...but that didn't catch anything on my local system (Chromium 70.0.3538.77) or in the Docker image used by Taskcluster (Google Chrome 72.0.3622.0 dev). Even the exact command reported in the Taskcluster logs (too lengthy to post here) didn't demonstrate the instability. Those CRASHes are suspicious, though. I don't think it's a memory limitation because the utilization is documented in the logs:

2018-12-05 01:37:06,210 - tc-run - INFO - Identifying tests affected in range 'HEAD^'...
mem avail: 29462 MiB (96 %), swap free: 0 MiB (0 %)
2018-12-05 01:37:34,339 - tc-run - INFO - Identified 17 affected tests

[...]

PID 1140 | Only local connections are allowed.
mem avail: 27550 MiB (89 %), swap free: 0 MiB (0 %)
PID 1452 | Starting ChromeDriver 2.44.609551 (5d576e9a44fe4c5b6a07e568f1ebc753f1214634) on port 4444

[...]

PID 10044 | Only local connections are allowed.
mem avail: 28787 MiB (93 %), swap free: 0 MiB (0 %)
PID 10320 | Starting ChromeDriver 2.44.609551 (5d576e9a44fe4c5b6a07e568f1ebc753f1214634) on port 4444

It's strange that the container has no swap memory, but I'm not sure that's a problem. It doesn't seem as though we'd need to do any paging with 28 gigabytes of free memory. Unfortunately, I can't replicate that condition via Docker running on my system due to kernel limitations.

@Hexcles do you have thoughts on any of this?

[Hexcles]

Right, our Docker containers don't have swap, which shouldn't have any negative affects. We should never use anything even close to the total available memory (guaranteed to be at least 8GB).

[pwnall]

LGTM.

Thank you very much for the cleanup! I don't know what's up with the bots, but this looks like a step forward for the tests.

[jugglinmike]

This was addressed in Chromium directly and merged to WPT via gh-17949. I've verified that patch addressed all of the instances of asynchronous cleanup that we identified here. It's good that the change eventually made it through stability checks!