You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Tony Butt <tj...@cea.com.au> on 2004/10/12 01:38:07 UTC
Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52 on Windows 2000
We have been running the subversion 1.0.x series with Apache 2.0.48, then Apache 2.0.50 on W2000 SP3 for some months, and have recently encountered a major problem.
I recently updated our subversion server to 1.1.0 from 1.1.0 rc3
At the same time, I updated my client to TortoiseSVN 1.1.0, from the prior RC.
My first sympton was that when using the repo-browser with Tortoise, the performance was poor, and sometimes just failed to return. At the same time, the apache process on the W2K repository server would spin, growing in size to 24M or so, and consuming 100% CPU for an extended period of time.
I first suspected Tortoise, and reverted to a prior version, giving the same problem.
I tried a vendor merge of a source tree from the command line, as I have done several times in the past. It ran very slowly, exhibiting the same apache server problem, and at some point failed(!) I have totally reinstalled apache, updating to 2.0.52 from 2.0.50, reinstalled subversion on the server, and this problem persists. I checked the repository with svnadmin, and everything is Ok.
Access to the repository works fine via file:/// or svn:/// protocols. Strangely enough, web browser access seems to work OK
I have even swapped in a backup of the repository, and still have the same problem.
For the time being, we will limp along with svn: protocol access, but we would much prefer to use the apache server.
Is there some apache configuration I should look at, or is this a real bug with the subversion apache modules.
Tony Butt
CEA Technologies
Canberra
Australia
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52 on Windows 2000
Posted by Simon Large <sl...@blazepoint.co.uk>.
<kf...@collab.net> writes:
> > If anyone is willing to pick this up it would be greatly
appreciated.
>
> #2011 is marked as "fixed", and the last note says it has been
> nominated for 1.1.1.
>
> So... ?? :-)
So ... whaddaya mean by fixing things when I'm not looking ;-)
> Re: 1.1.1 scheduling (19 Oct 2003 19:41)
> =====Not Fixed=====
>
> * UNC paths broken (Issue #2011)
That was swift work, Ben :-)
Thanks,
Simon
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52 on Windows 2000
Posted by kf...@collab.net.
"Simon Large" <sl...@blazepoint.co.uk> writes:
> Does that mean a fix for issue 2011 is being punted? We have had several
> TSVN users complaining that their setup no longer works with 1.1.0. I
> know there is a drive mapping workaround, but it is not always a good
> solution (one user needs more drive mappings than there are drive
> letters available).
>
> If anyone is willing to pick this up it would be greatly appreciated.
#2011 is marked as "fixed", and the last note says it has been
nominated for 1.1.1.
So... ?? :-)
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Erich Oliphant <er...@vantixweb.com>.
Solved my problem. I upgraded to 2.0.52 to no avail, then started
disabling services. I am also running Resin 3.0 and it's Apache plugin,
looks like if the resin server is not up the plugin slows down apache
processing somehow. Sorry for the false alarm :)
Simon Large wrote:
>>>Looking forward to 1.1.1 as I am demoing a new software dev.
>>>
>>>
>environment
>
>
>>>centered around SVN soon and this does not look good.
>>>
>>>
>>It should be out in a few days.
>>
>>
>
>Does that mean a fix for issue 2011 is being punted? We have had several
>TSVN users complaining that their setup no longer works with 1.1.0. I
>know there is a drive mapping workaround, but it is not always a good
>solution (one user needs more drive mappings than there are drive
>letters available).
>
>If anyone is willing to pick this up it would be greatly appreciated.
>
>Simon
>
>
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
>For additional commands, e-mail: users-help@subversion.tigris.org
>
>
>
>
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52 on Windows 2000
Posted by Simon Large <sl...@blazepoint.co.uk>.
> > Looking forward to 1.1.1 as I am demoing a new software dev.
environment
> > centered around SVN soon and this does not look good.
>
> It should be out in a few days.
Does that mean a fix for issue 2011 is being punted? We have had several
TSVN users complaining that their setup no longer works with 1.1.0. I
know there is a drive mapping workaround, but it is not always a good
solution (one user needs more drive mappings than there are drive
letters available).
If anyone is willing to pick this up it would be greatly appreciated.
Simon
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Erich Oliphant <er...@vantixweb.com>.
No prob, I am running via http (apache 2.0.46)
OS: Red Hat 9 (kernel 2.4.18-14)
HW: PIII 900 768Mb RAM
Revisions: ~220
The 1.1.0 repository was created by loading a dump of the .36
repository. When I first encounterd the problem, I ran 'svnadmin
recover' (after shutting down httpd :)), but it reported no probs and
returned quickly (less than 1 min). Initially my windows client was
outdated (.36), upgraded to 1.1.0 still saw the same response times.
1.0 client on OSX shows similar behavior so it appears to be it's a
server side prob.
As far as actual performance, here's a commit I just did:
---
[eoliphan-receiptimaging]:date ; svn commit -m "Added context param to
set expect timeout prior to mget"
src/com/trw/dts/receiptimaging/GetFaxImage.java ; date
Wed Oct 20 11:37:27 EDT 2004
Sending src/com/trw/dts/receiptimaging/GetFaxImage.java
Transmitting file data .
Committed revision 221.
Wed Oct 20 11:49:25 EDT 2004
[eoliphan-receiptimaging]:
---
So around 11 mins.
I just hit some other URLs and it appears that apache in general is
slow. However, the SVN upgrade is the only change i've made. I am
going to upgrade apache to the latest rev and see what that does.
Tobias Ringström wrote:
> Erich Oliphant wrote:
>
>> Don't know if this is related, but I just upgraded from .36 to 1.1.0 on
>> RH Linux (server) and Windows (client). Everything seems to complete
>> but is agonizingly slow (status, ls, checkout, update, etc). Repos is
>> only ~ 250 Mb and was blazingly fast before the upgrade.
>
>
> Assuming that you're using http[s]://, I'm not surprised that ls is
> hideously slow, and that log and blame are slower than before, but I'm
> very surprised that the other operations are "agonizingly slow". Can
> you give us some numbers and a description of your setup? An
> reproduction recipe would be great.
>
>> Looking forward to 1.1.1 as I am demoing a new software dev. environment
>> centered around SVN soon and this does not look good.
>
>
> It should be out in a few days.
>
> /Tobias
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Erich Oliphant wrote:
> Don't know if this is related, but I just upgraded from .36 to 1.1.0 on
> RH Linux (server) and Windows (client). Everything seems to complete
> but is agonizingly slow (status, ls, checkout, update, etc). Repos is
> only ~ 250 Mb and was blazingly fast before the upgrade.
Assuming that you're using http[s]://, I'm not surprised that ls is
hideously slow, and that log and blame are slower than before, but I'm
very surprised that the other operations are "agonizingly slow". Can you
give us some numbers and a description of your setup? An reproduction
recipe would be great.
> Looking forward to 1.1.1 as I am demoing a new software dev. environment
> centered around SVN soon and this does not look good.
It should be out in a few days.
/Tobias
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Erich Oliphant <er...@vantixweb.com>.
Don't know if this is related, but I just upgraded from .36 to 1.1.0 on
RH Linux (server) and Windows (client). Everything seems to complete
but is agonizingly slow (status, ls, checkout, update, etc). Repos is
only ~ 250 Mb and was blazingly fast before the upgrade.
Looking forward to 1.1.1 as I am demoing a new software dev. environment
centered around SVN soon and this does not look good.
Tobias Ringström wrote:
> Tony Butt wrote:
>
>> Now, the ECos source tree is ~ 600 MB, and the merge process used to
>> take 5-10 minutes, using apache/dav_svn on a Windows 200 Server (PIII
>> 800MHz)
>> Currently, the merge process takes over 30 minutes, and fails to
>> complete, returning this error message:
>> svn: REPORT request failed on '/repos/cea/!svn/vcc/default'
>> svn: REPORT of '/repos/cea/!svn/vcc/default': Could not read
>> response body: An existing connection was forcibly closed by the
>> remote host. (http://svnserver)
>>
>> We have switched to the svn: protocol as a workaround, and the
>> problem has gone away.
>> Additionally, it runs MUCH faster than the http: protocol ever did -
>> as does browsing, (and probably checkouts, etc)
>> Is this a common experience? We would have expected better
>> performance from the apache based repository server.
>
>
> Yes, svn:// is faster than http:// in general. How much faster is the
> merge in your case?
>
> It would also be very interesting to hear how http:// performs with
> "SVNPathAuthz Off", but since that directive is broken in 1.1.0, we
> have to wait until Subversion 1.1.1 is released (which will probably
> happen in a week or so).
>
> If you're using the default BDB filesystem right now, it would
> probably make things faster if you switch to FSFS ("svnadmin create
> --fs-type fsfs repos"), especially since you're on Windows.
>
> /Tobias
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Tony Butt wrote:
> Now, the ECos source tree is ~ 600 MB, and the merge process used to
> take 5-10 minutes, using apache/dav_svn on a Windows 200 Server (PIII
> 800MHz)
> Currently, the merge process takes over 30 minutes, and fails to
> complete, returning this error message:
> svn: REPORT request failed on '/repos/cea/!svn/vcc/default'
> svn: REPORT of '/repos/cea/!svn/vcc/default': Could not read
> response body: An existing connection was forcibly closed by the
> remote host. (http://svnserver)
>
> We have switched to the svn: protocol as a workaround, and the problem
> has gone away.
> Additionally, it runs MUCH faster than the http: protocol ever did -
> as does browsing, (and probably checkouts, etc)
> Is this a common experience? We would have expected better performance
> from the apache based repository server.
Yes, svn:// is faster than http:// in general. How much faster is the
merge in your case?
It would also be very interesting to hear how http:// performs with
"SVNPathAuthz Off", but since that directive is broken in 1.1.0, we have
to wait until Subversion 1.1.1 is released (which will probably happen
in a week or so).
If you're using the default BDB filesystem right now, it would probably
make things faster if you switch to FSFS ("svnadmin create --fs-type
fsfs repos"), especially since you're on Windows.
/Tobias
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Tony Butt <tj...@cea.com.au>.
Tobias Ringström wrote:
> Tony Butt wrote:
>
>> We have been running the subversion 1.0.x series with Apache 2.0.48,
>> then Apache 2.0.50 on W2000 SP3 for some months, and have recently
>> encountered a major problem.
>>
>> I recently updated our subversion server to 1.1.0 from 1.1.0 rc3
>> At the same time, I updated my client to TortoiseSVN 1.1.0, from the
>> prior RC.
>>
>> My first sympton was that when using the repo-browser with Tortoise,
>> the performance was poor, and sometimes just failed to return. At the
>> same time, the apache process on the W2K repository server would
>> spin, growing in size to 24M or so, and consuming 100% CPU for an
>> extended period of time.
>
>
> Thanks for the report. This is a know problem that was introduced in
> 1.0.8 and 1.1.0-rc4. There will be a 1.0.9 and a 1.1.1 release this
> week to fix that problem.
>
>> I tried a vendor merge of a source tree from the command line, as I
>> have done several times in the past. It ran very slowly, exhibiting
>> the same apache server problem, and at some point failed(!) I have
>> totally reinstalled apache, updating to 2.0.52 from 2.0.50,
>> reinstalled subversion on the server, and this problem persists. I
>> checked the repository with svnadmin, and everything is Ok.
>
>
> I wonder if this is the same problem. Can you show us exactly what you
> did?
>
> /Tobias
>
Thanks for the quick reply.
The short response is that I followed the Vendor branch merge
instructions in the 'Version Control with Subversion' manual, chapter 7
(Advanced Topics)
More generally, we use an RTOS called ECos, which provides updates via
CVS, so:
* Check out latest ecos code from the ecos CVS server to our
internet gateway.
* Check in the ecos tree to the subversion repository, ecos project
(http://svnserver/repos/CEA/ecos/ecos-2.0/vendor/ecos)
* Copy the current ecos vendor tree to a tag
(http://svnserver/repos/CEA/ecos/ecos-2.0/vendor/ecos-yymmdd)
* Merge in the changes to my working directory:
o svn merge
http://svnserver/repos/CEA/ecos/ecos-2.0/vendor/ecos-040817
http://svnserver/repos/CEA/ecos/ecos-2.0/vendor/ecos ecos-2.0
* Resolve any merge conflicts
* Rebuild ECos, test for a week, or so, then check in to the trunk
branch of the ecos project.
Now, the ECos source tree is ~ 600 MB, and the merge process used to
take 5-10 minutes, using apache/dav_svn on a Windows 200 Server (PIII
800MHz)
Currently, the merge process takes over 30 minutes, and fails to
complete, returning this error message:
svn: REPORT request failed on '/repos/cea/!svn/vcc/default'
svn: REPORT of '/repos/cea/!svn/vcc/default': Could not read
response body: An existing connection was forcibly closed by the remote
host. (http://svnserver)
We have switched to the svn: protocol as a workaround, and the problem
has gone away.
Additionally, it runs MUCH faster than the http: protocol ever did - as
does browsing, (and probably checkouts, etc)
Is this a common experience? We would have expected better performance
from the apache based repository server.
Tony Butt
Re: Poor Performance - Subversion 1.1.0 and Apache 2.0.50/2.0.52
on Windows 2000
Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Tony Butt wrote:
> We have been running the subversion 1.0.x series with Apache 2.0.48,
> then Apache 2.0.50 on W2000 SP3 for some months, and have recently
> encountered a major problem.
>
> I recently updated our subversion server to 1.1.0 from 1.1.0 rc3
> At the same time, I updated my client to TortoiseSVN 1.1.0, from the
> prior RC.
>
> My first sympton was that when using the repo-browser with Tortoise,
> the performance was poor, and sometimes just failed to return. At the
> same time, the apache process on the W2K repository server would spin,
> growing in size to 24M or so, and consuming 100% CPU for an extended
> period of time.
Thanks for the report. This is a know problem that was introduced in
1.0.8 and 1.1.0-rc4. There will be a 1.0.9 and a 1.1.1 release this week
to fix that problem.
> I tried a vendor merge of a source tree from the command line, as I
> have done several times in the past. It ran very slowly, exhibiting
> the same apache server problem, and at some point failed(!) I have
> totally reinstalled apache, updating to 2.0.52 from 2.0.50,
> reinstalled subversion on the server, and this problem persists. I
> checked the repository with svnadmin, and everything is Ok.
I wonder if this is the same problem. Can you show us exactly what you did?
/Tobias
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org