[PATCH] runtests: check ports on IPv6 address

Martin von Zweigbergk martinvonz at google.com
Wed Feb 15 23:24:58 EST 2017


On Wed, Feb 15, 2017 at 8:23 PM, Martin von Zweigbergk
<martinvonz at google.com> wrote:
> On Wed, Feb 15, 2017 at 8:12 PM, Augie Fackler <raf at durin42.com> wrote:
>>
>>> On Feb 9, 2017, at 8:59 AM, Jun Wu <quark at fb.com> wrote:
>>>
>>> # HG changeset patch
>>> # User Jun Wu <quark at fb.com>
>>> # Date 1486648674 28800
>>> #      Thu Feb 09 05:57:54 2017 -0800
>>> # Node ID 93e23f7b87a4ab456053b6ba573615be16c6c4b0
>>> # Parent  a68510b69f413545722c086eaeb840dd5e8305b4
>>> # Available At https://bitbucket.org/quark-zju/hg-draft
>>> #              hg pull https://bitbucket.org/quark-zju/hg-draft -r 93e23f7b87a4
>>> runtests: check ports on IPv6 address
>>
>> This has made test-bundle2-remote-changegroup.t flaky when run in parallel with other tests. I was able to bisect with this:
>>
>> cd tests && python run-tests.py -j 100 --runs-per-test 400 test-bundle2-remote-changegroup.t
>>
>> fails in about 70 out of the 400 runs. I also see it periodically when running the tests with -j120 on gcc112 from the gcc compile farm, which is a POWER8 machine with 160 hardware threads. I suspect you can reproduce it with a lower -j if you run the test enough times, but I’m up later than I should be already, so I didn’t bother logging in to a smaller linux machine to confirm.
>>
>> Can you take a look?
>
> Heh, I reported the flakiness on #mercurial earlier today. That's why
> Jun sent a series of patches maybe 2 hours ago that you then queued.
> Have you tested with those patches applied? I have not had time to
> look at his patches, but maybe he should have explained their
> motivation better if this is indeed the same problem that I reported.

Or maybe I just incorrectly assumed that the problems I reported with
ports not being available had the same root cause as the flaky
test-bundle2-remote-changegroup.t?

>
>>
>> Thanks!
>> Augie
>>
>>> Previously, checkportisavailable only checks ports on the IPv4 address. This
>>> patch makes it check IPv6 as well. It'll be useful if "localhost" does not
>>> have an IPv4 address, or its IPv4 address does not exist somehow.
>>>
>>> diff --git a/tests/run-tests.py b/tests/run-tests.py
>>> --- a/tests/run-tests.py
>>> +++ b/tests/run-tests.py
>>> @@ -115,13 +115,17 @@ wifexited = getattr(os, "WIFEXITED", lam
>>> def checkportisavailable(port):
>>>     """return true if a port seems free to bind on localhost"""
>>> -    try:
>>> -        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
>>> -        s.bind(('localhost', port))
>>> -        s.close()
>>> -        return True
>>> -    except socket.error as exc:
>>> -        if not exc.errno == errno.EADDRINUSE:
>>> -            raise
>>> -        return False
>>> +    families = [getattr(socket, i, None)
>>> +                for i in ('AF_INET', 'AF_INET6')
>>> +                if getattr(socket, i, None) is not None]
>>> +    for family in families:
>>> +        try:
>>> +            s = socket.socket(family, socket.SOCK_STREAM)
>>> +            s.bind(('localhost', port))
>>> +            s.close()
>>> +            return True
>>> +        except socket.error as exc:
>>> +            if exc.errno not in (errno.EADDRINUSE, errno.EADDRNOTAVAIL):
>>> +                raise
>>> +    return False
>>>
>>> closefds = os.name == 'posix'
>>> _______________________________________________
>>> Mercurial-devel mailing list
>>> Mercurial-devel at mercurial-scm.org
>>> https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel
>>
>> _______________________________________________
>> Mercurial-devel mailing list
>> Mercurial-devel at mercurial-scm.org
>> https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel


More information about the Mercurial-devel mailing list