runtests: drop logic calling the handle tool (Windows)

[vszakats]

In the cases observed throughout the last year, handle64 run once per
test run, but with no action (match or task kill). It did not help with
flakiness and seems redundant.

runtests launched it (if present) in Cygwin/MSYS jobs too, where it
probably shouldn't have, because we have seen no flakiness there. In CI
the tool was present and launched in MSYS2 jobs, but not in Cygwin.

After this patch the "clearlocks" warning remain in the log. They are
consistently appearing once in every MSVC CI log, early in the tests:

  test 3207 SKIPPED: curl lacks OpenSSL support
[...START-OF-TESTS...]
  test 0003...[HTTP POST with auth and contents but with content-length set to 0]
  --pd---e--- OK (3   out of 1596, remaining: 17:50, took 1.423s, duration: 00:02)
  test 0007...[HTTP with cookie parser and header recording]
  --pd--oe--- OK (7   out of 1596, remaining: 07:51, took 1.485s, duration: 00:02)
  test 0006...[HTTP with simple cookie send]
  --pd---e--- OK (6   out of 1596, remaining: 09:11, took 1.488s, duration: 00:02)
  test 0005...[HTTP over proxy]
  --pd---e--- OK (5   out of 1596, remaining: 11:03, took 1.491s, duration: 00:02)
CUSTOMBUILD : error : 169: cleardir(log/8/lock) failed [D:\a\curl\curl\bld\tests\test-ci.vcxproj]
  test 0001...[HTTP GET]
  --pd---e--- OK (1   out of 1596, remaining: 55:34, took 1.466s, duration: 00:02)
  test 0004...[Replaced internal and added custom HTTP headers]

Ref: https://github.com/curl/curl/actions/runs/13546192228/job/37858323380?pr=16484#step:14:167

Ref: e53523f #14859
Ref: 311c31e #6179
Follow-up to 3a8920e #16600

[vszakats]

@mback2k, @jay: Your thoughts on this? I'd be interested in ways to trigger handle64 finding an actual kill. It's also an option to keep the logic in runtests.pl and just drop handle64 installs from CI, if it does help outside CI.

[jay]

I'm going to defer to Marc on this one. My general thinking is it should not happen however weird things in CI happen a lot. I really could go either way. #6058 for background. I'd leave this open a bit to give him some time to respond.

[mback2k]

@mback2k, @jay: Your thoughts on this? I'd be interested in ways to trigger handle64 finding an actual kill.

I think this happened with stuck tests/testservers in the past, especially with those related to timeouts.

It's also an option to keep the logic in runtests.pl and just drop handle64 installs from CI, if it does help outside CI.

Fine with me. You could also drop the '-rm' flag from being active on CI first.

[vszakats]

@mback2k, @jay: Your thoughts on this? I'd be interested in ways to trigger handle64 finding an actual kill.

I think this happened with stuck tests/testservers in the past, especially with those related to timeouts.

It's also an option to keep the logic in runtests.pl and just drop handle64 installs from CI, if it does help outside CI.

Fine with me. You could also drop the '-rm' flag from being active on CI first.

@mback2k Just to clarify, are you also okay to drop the clearlocks() function
from servers.pm (making the call to handle*.exe), or is it still needed
in some envs?

Meanwhile tested without -rm in CI, and it builds OK.

[mback2k]

@mback2k Just to clarify, are you also okay to drop the clearlocks() function

from servers.pm (making the call to handle*.exe), or is it still needed

in some envs?

Meanwhile tested without -rm in CI, and it builds OK.

Maybe keep the code around for a little bit longer until the CI runs have proven to be stable without it for some weeks?

[vszakats]

@mback2k Just to clarify, are you also okay to drop the clearlocks() function
from servers.pm (making the call to handle*.exe), or is it still needed
in some envs?
Meanwhile tested without -rm in CI, and it builds OK.

Maybe keep the code around for a little bit longer until the CI runs have proven to be stable without it for some weeks?

It works for me, I'll merge the CI patch separately now. And keep this
one around for a while.

I think the main issue is that even though handle runs exactly once for
every MSYS2 and MSVC CI job, it never finds anything to kill. I've added
verbose logs to follow this, and tested it extensively a while ago. Couldn't
figure out why that happens. The code, filters all seem correct.

[vszakats]

I could observe no effect of disabling handle almost 2 weeks ago.

Windows CI looks just like before, with mostly:

1. greyed out test run step
2. 2304 errors
3. pipe errors
4. the occasional flaky test, mostly just a single test

1., 2., 3. are sometimes combined and/or the result of hangs.

This is about the same as we observed always. handle doesn't seem to play a role, and when run, it was never seen to kill a job. The condition keeps triggering once for every test run, at the beginning, sometimes right in the beginning (maybe always, hard to say from the logs).

Planning to merge this today.