Handle blocked I/O of exec'd processes, and remove flaky TestExecWindowsOpenHandles

[crosbymichael]

fixes #39326

This is the second part to
containerd/containerd#3361 and will help process
delete not block forever when the process exists but the I/O was
inherited by a subprocess that lives on.

Signed-off-by: Michael Crosby crosbymichael@gmail.com

[thaJeztah]

addresses containerd/containerd#3286 docker exec hang if earlier docker exec left a zombie process

@crosbymichael does this also relate to;

• docker cannot terminate a container when processes linger after TERM #38978 Zombie reaping not occurring as expected
• docker cannot terminate a container when processes linger after TERM #38978 docker cannot terminate a container when processes linger after TERM

Should we have a test-case for this? (not sure if it's easy to test)

[thaJeztah]

Haven't seen this one before; https://jenkins.dockerproject.org/job/Docker-PRs-WoW-RS1/25656/console

21:06:29 FAIL: docker_cli_exec_test.go:538: DockerSuite.TestExecWindowsOpenHandles
21:06:29
21:06:29 assertion failed: 2 (int) != 1 (int)

I'll kick CI for RS1 to see if it was just flaky

[crosbymichael]

I read through that test and its depending on the "bad" functionality, forking off the sleeps into the background and the actually cmd.exe is exiting. I think it's safe for this test to be removed with this change as well as the test is horribly racy and depending on how many seconds the sleeps wait until they exit, its bad!

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@52c1667). Click here to learn what that means.
The diff coverage is 0%.

@@            Coverage Diff            @@
##             master   #39383   +/-   ##
=========================================
  Coverage          ?   37.34%           
=========================================
  Files             ?      609           
  Lines             ?    45269           
  Branches          ?        0           
=========================================
  Hits              ?    16905           
  Misses            ?    26074           
  Partials          ?     2290

[thaJeztah]

SGTM

@kolyshkin @dmcgowan PTAL

[cpuguy83]

LGTM

[cpuguy83]

I/O waiters seems like a good candidate for metrics as well.