You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Johan Corveleyn <jc...@gmail.com> on 2018/10/02 00:38:07 UTC
conflicts-test very slow over svnserve (Windows)
I haven't looked deeply into it yet, but it seems that the
conflicts-tests (the C tests) are very slow when testing over
svnserve. I noticed this while testing 1.11.0-rc2, but I see the same
with trunk and with 1.10.x.
Running the entire conflicts-test.exe (on a ramdisk):
- ra_local: 23s
- ra_serf: 70s
- ra_svn: 982s (16m22s)
The difference between ra_local and ra_serf strikes me as normal
(after all, the conflict resolver contacts the repository a lot to
fetch information). The time of ra_svn not really :-).
The slowdown seems to be spread over all the tests of conflicts-test
(the dots of the test progress come evenly spread).
Is anyone else seeing this? Only on Windows, or on *nix too?
A performance problem with tree conflict resolution specific to svnserve?
--
Johan
Re: conflicts-test very slow over svnserve (Windows)
Posted by Johan Corveleyn <jc...@gmail.com>.
On Tue, Oct 2, 2018 at 5:37 PM Bert Huijben <be...@qqmail.nl> wrote:
> On Tue, Oct 2, 2018 at 4:30 PM Branko Čibej <br...@apache.org> wrote:
>>
>> On 02.10.2018 02:38, Johan Corveleyn wrote:
>> > I haven't looked deeply into it yet, but it seems that the
>> > conflicts-tests (the C tests) are very slow when testing over
>> > svnserve. I noticed this while testing 1.11.0-rc2, but I see the same
>> > with trunk and with 1.10.x.
>> >
>> > Running the entire conflicts-test.exe (on a ramdisk):
>> > - ra_local: 23s
>> > - ra_serf: 70s
>> > - ra_svn: 982s (16m22s)
>> >
>> > The difference between ra_local and ra_serf strikes me as normal
>> > (after all, the conflict resolver contacts the repository a lot to
>> > fetch information). The time of ra_svn not really :-).
>> >
>> > The slowdown seems to be spread over all the tests of conflicts-test
>> > (the dots of the test progress come evenly spread).
>> >
>> > Is anyone else seeing this? Only on Windows, or on *nix too?
>> > A performance problem with tree conflict resolution specific to svnserve?
>>
>> How do you set the URL for win-tests.py? I've noticed that using
>> svn://localhost is orders of magnitude slower than svn://127.0.0.1,
>> apparently because the Windows resolver attempts a DNS search for
>> 'localhost' first (and failing) before resolving it locally.
>>
>> -- Brane
>>
> It resolves to both ::1 and 127.0.0.1 on Windows, and as (per RFC) ipv6 is tried before ipv4... things timeout before trying the address used by default (hardcoded) in svnserve.
> (We used to default to ipv6 for some time... But we decided to explicitly default to ipv4 when not passing the later added -6)
>
> More recent Windows versions have a workaround/speedup for this problem. They will swap the DNS results after a few failed attempts.
>
> Bert
Yep, that's it. Thanks all.
When using svn://127.0.0.1 conflicts-test.exe takes around 28s for
1.11.0-rc2 on my machine. Much better :-).
--
Johan
Re: conflicts-test very slow over svnserve (Windows)
Posted by Bert Huijben <be...@qqmail.nl>.
It resolves to both ::1 and 127.0.0.1 on Windows, and as (per RFC) ipv6 is
tried before ipv4... things timeout before trying the address used by
default (hardcoded) in svnserve.
(We used to default to ipv6 for some time... But we decided to explicitly
default to ipv4 when not passing the later added -6)
More recent Windows versions have a workaround/speedup for this problem.
They will swap the DNS results after a few failed attempts.
Bert
On Tue, Oct 2, 2018 at 4:30 PM Branko Čibej <br...@apache.org> wrote:
> On 02.10.2018 02:38, Johan Corveleyn wrote:
> > I haven't looked deeply into it yet, but it seems that the
> > conflicts-tests (the C tests) are very slow when testing over
> > svnserve. I noticed this while testing 1.11.0-rc2, but I see the same
> > with trunk and with 1.10.x.
> >
> > Running the entire conflicts-test.exe (on a ramdisk):
> > - ra_local: 23s
> > - ra_serf: 70s
> > - ra_svn: 982s (16m22s)
> >
> > The difference between ra_local and ra_serf strikes me as normal
> > (after all, the conflict resolver contacts the repository a lot to
> > fetch information). The time of ra_svn not really :-).
> >
> > The slowdown seems to be spread over all the tests of conflicts-test
> > (the dots of the test progress come evenly spread).
> >
> > Is anyone else seeing this? Only on Windows, or on *nix too?
> > A performance problem with tree conflict resolution specific to svnserve?
>
> How do you set the URL for win-tests.py? I've noticed that using
> svn://localhost is orders of magnitude slower than svn://127.0.0.1,
> apparently because the Windows resolver attempts a DNS search for
> 'localhost' first (and failing) before resolving it locally.
>
> -- Brane
>
>
Re: conflicts-test very slow over svnserve (Windows)
Posted by Branko Čibej <br...@apache.org>.
On 02.10.2018 02:38, Johan Corveleyn wrote:
> I haven't looked deeply into it yet, but it seems that the
> conflicts-tests (the C tests) are very slow when testing over
> svnserve. I noticed this while testing 1.11.0-rc2, but I see the same
> with trunk and with 1.10.x.
>
> Running the entire conflicts-test.exe (on a ramdisk):
> - ra_local: 23s
> - ra_serf: 70s
> - ra_svn: 982s (16m22s)
>
> The difference between ra_local and ra_serf strikes me as normal
> (after all, the conflict resolver contacts the repository a lot to
> fetch information). The time of ra_svn not really :-).
>
> The slowdown seems to be spread over all the tests of conflicts-test
> (the dots of the test progress come evenly spread).
>
> Is anyone else seeing this? Only on Windows, or on *nix too?
> A performance problem with tree conflict resolution specific to svnserve?
How do you set the URL for win-tests.py? I've noticed that using
svn://localhost is orders of magnitude slower than svn://127.0.0.1,
apparently because the Windows resolver attempts a DNS search for
'localhost' first (and failing) before resolving it locally.
-- Brane
Re: conflicts-test very slow over svnserve (Windows)
Posted by Bert Huijben <be...@qqmail.nl>.
This could be related to log requests that are not somehow limited (by last
rev and/or number of results). ra_serf and ar_local allow cancelling
halfway through, while ra_svn doesn't.
(No time to investigate this myself righ now)
Bert
On Tue, Oct 2, 2018 at 2:38 AM Johan Corveleyn <jc...@gmail.com> wrote:
> I haven't looked deeply into it yet, but it seems that the
> conflicts-tests (the C tests) are very slow when testing over
> svnserve. I noticed this while testing 1.11.0-rc2, but I see the same
> with trunk and with 1.10.x.
>
> Running the entire conflicts-test.exe (on a ramdisk):
> - ra_local: 23s
> - ra_serf: 70s
> - ra_svn: 982s (16m22s)
>
> The difference between ra_local and ra_serf strikes me as normal
> (after all, the conflict resolver contacts the repository a lot to
> fetch information). The time of ra_svn not really :-).
>
> The slowdown seems to be spread over all the tests of conflicts-test
> (the dots of the test progress come evenly spread).
>
> Is anyone else seeing this? Only on Windows, or on *nix too?
> A performance problem with tree conflict resolution specific to svnserve?
>
> --
> Johan
>
Re: conflicts-test very slow over svnserve (Windows)
Posted by Daniel Shahaf <d....@daniel.shahaf.name>.
Johan Corveleyn wrote on Tue, 02 Oct 2018 02:38 +0200:
> Running the entire conflicts-test.exe (on a ramdisk):
Running «make {check,davautocheck,svnserveautocheck}
TESTS=subversion/tests/…/conflicts-test APACHE_MPM=event» with the build tree
in a ramdisk, I get (average of three runs in each case):
> - ra_local: 23s
~2.4sec
> - ra_serf: 70s
12sec (+ 3sec for davautocheck.sh overhead)
> - ra_svn: 982s (16m22s)
4sec to 5sec
>
Cheers,
Daniel