You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by Bryan Dyck <bd...@mac.com> on 2005/06/22 19:00:50 UTC
Checkout over HTTP fails with seemingly random errors
After poking somewhat methodically at this, I'm not getting anywhere,
so it's time to ask the experts. :)
First, a short leadup to the problem: I recently upgraded my
Subversion box to OS X 10.4, which of course entails rebuilding the
necessary components to get Subversion running again. I was
previously using the fink packages, but decided to be less lazy this
time around and build everything myself. My current setup includes
the following:
APR/APR-Util 0.9.6 (I built this separately to make double-certain
that Apache and Subversion use the same APR)
Neon 0.24.7 (also built separately from Subversion)
Apache 2.0.54
Subversion 1.2.0
SWIG 1.3.21 (not really relevant to this, but it's there)
I have not installed BDB, as I've had no problems with using an FSFS
backend and didn't feel there were any advantages (for me) to have
the option of a BDB backend.
After building everything and setting up Apache (with the Subversion
modules built as DSOs), I proceeded to recreate the repository and
loaded my dumpfile back in. The problems began when I did a cursory
checkout to see if everything was working properly - it wasn't. The
checkout failed, and subsequent attempts also failed with a seemingly
random assortment of errors:
svn: Invalid diff stream: [tgt] insn 1 starts beyond the target view
position
svn: Unexpected end of svndiff input
svn: REPORT request failed on '/svn/prima-survey/!svn/vcc/default'
svn: REPORT of '/svn/prima-survey/!svn/vcc/default': Chunk delimiter
was invalid
svn: Checksum mismatch for 'prima-survey/lib/log4j-1.2.9.jar';
expected: 'c2ced5f8505fe9d1cae685201e9cba07', actual:
'be72ae3be573c017a8906647c8b13dd4'
These occurred in no particular order, and sometimes at different
points in the checkout. Having not once had even a hint of a problem
with the actual operation of Subversion, this was rather disturbing
to me. The checksum failure occurred only a couple times, but that
and the REPORT failure led me to initially wonder if the repository
was corrupted, so I ran 'svnadmin verify' on it, which came back with
no errors.
Googling for the various error messages didn't really produce
anything terribly interesting, other than that other people have
apparently suffered similar problems, some of which seemed to be
traced back to a repository that had become corrupt at some point.
I'm fairly certain that my repository isn't corrupt, partly because
I've never had a problem committing to it (it's fairly young; only 31
revisions), and partly because I can run a checkout just fine if I
use a file:// URL.
Thus far, I have disabled HTTP compression in the client-side SVN
options (I built Apache with mod_deflate), and I have changed
Apache's LogLevel to debug, which hasn't produced much of note other
than this:
[info] (35)Resource temporarily unavailable: core_output_filter:
writing data to the network
and after a REPORT failure error, this:
[error] [client 192.168.1.1] Provider encountered an error while
streaming a REPORT response. [500, #0]
[error] [client 192.168.1.1] A failure occurred while driving the
update report editor [500, #190004]
One other point that might be germane is that I saved the dumpfile to
an SMB share on a Win2k box before wiping my OS X box and installing
Tiger. I've done this before, and not had a problem with it, but
there you go..
With all that out of the way, does anyone have any bright ideas as to
what I might be doing wrong, or how this can be fixed? Many thanks in
advance, and my apologies for the long-windedness.. :)
Cheers,
Bryan
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Checkout over HTTP fails with seemingly random errors
Posted by kf...@collab.net.
Bryan Dyck <bd...@mac.com> writes:
> > This might be the famous bug in APR that only revealed itself on
> > Tiger recently, something to do with APR using poll instead of
> > select, or vice versa. The symptom was apache essentially unable
> > to stream >64k files to the client.
> >
> > The bug has been fixed in the latest HEAD of apr and apr-util.
> >
> > As an experiment, you might want to checkout the latest apr/apr-
> > util/httpd, and build them all from source and install. Then
> > rebuild Subversion, since apr 1.X is binary-incompatible at runtime
> > with apr-0.X. See if it clears things up.
>
> Hmm... that sounds like an ugly bug. If that's the case, I'm relieved
> to hear it, because I was beginning to wonder if I was going slightly
> crazy.
>
> I'll try checking out the latest apr/apr-util/httpd and rebuilding
> the whole show... Hopefully that clears up the problem. Either way,
> I'll report back with the results.
>
> Thanks for the quick response and keep up the awesome work, Ben (and
> the rest of the SVN team).
And if there's not an entry for this already in
http://subversion.tigris.org/faq.html
...then please consider writing one for us. Searching for "64 and
"tiger" on that page doesn't reveal any, so I suspect we don't have
anything about this. See
http://subversion.tigris.org/mailing-list-guidelines.html#patches
for patch submission guidelines.
Thanks,
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Checkout over HTTP fails with seemingly random errors
Posted by Bryan Dyck <bd...@mac.com>.
On 22-Jun-05, at 3:15 PM, Stephen Davis wrote:
> On Jun 22, 2005, at 2:36 PM, Bryan Dyck wrote:
>
>
>
>> On 22-Jun-05, at 1:30 PM, Ben Collins-Sussman wrote:
>>
>>
>>
>>> On Jun 22, 2005, at 2:00 PM, Bryan Dyck wrote:
>>>
>>>
>>>
>>>>
>>>>
>>>> svn: Invalid diff stream: [tgt] insn 1 starts beyond the target
>>>> view position
>>>>
>>>> svn: Unexpected end of svndiff input
>>>>
>>>> svn: REPORT request failed on '/svn/prima-survey/!svn/vcc/default'
>>>> svn: REPORT of '/svn/prima-survey/!svn/vcc/default': Chunk
>>>> delimiter was invalid
>>>>
>>>> svn: Checksum mismatch for 'prima-survey/lib/log4j-1.2.9.jar';
>>>> expected: 'c2ced5f8505fe9d1cae685201e9cba07', actual:
>>>> 'be72ae3be573c017a8906647c8b13dd4'
>>>>
>>>> [error] [client 192.168.1.1] Provider encountered an error while
>>>> streaming a REPORT response. [500, #0]
>>>> [error] [client 192.168.1.1] A failure occurred while driving
>>>> the update report editor [500, #190004]
>>>>
>>>>
>>>
>>> This might be the famous bug in APR that only revealed itself on
>>> Tiger recently, something to do with APR using poll instead of
>>> select, or vice versa. The symptom was apache essentially unable
>>> to stream >64k files to the client.
>>>
>>> The bug has been fixed in the latest HEAD of apr and apr-util.
>>>
>>> As an experiment, you might want to checkout the latest apr/apr-
>>> util/httpd, and build them all from source and install. Then
>>> rebuild Subversion, since apr 1.X is binary-incompatible at
>>> runtime with apr-0.X. See if it clears things up.
>>>
>>>
>>
>> Hmm... that sounds like an ugly bug. If that's the case, I'm
>> relieved to hear it, because I was beginning to wonder if I was
>> going slightly crazy.
>>
>> I'll try checking out the latest apr/apr-util/httpd and rebuilding
>> the whole show... Hopefully that clears up the problem. Either
>> way, I'll report back with the results.
>>
>> Thanks for the quick response and keep up the awesome work, Ben
>> (and the rest of the SVN team).
>>
>> Cheers,
>> b
>>
>>
>
> I believe you can also just set the following in your environment
> before configuring apache2. This prevents apache2 from using poll.
>
> export ac_cv_func_poll=no
Apologies for the late response; I had to cut my workday short and
didn't get to finish dealing with this.
Thanks for the tip - I got part way into doing a bleeding-edge build
of everything but since I still have the old version of apr/apr-util
installed, I'll try this first and see if that solves the issue. If
not, I'll grab a recent Apache version (I'm a bit leery of just
checking out HEAD on the 2.1.x tree and building that, so I thought
I'd take the most recent tagged version, since APR 1.x doesn't work
with Apache 2.0.x) and go that route.
Cheers,
b
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Checkout over HTTP fails with seemingly random errors
Posted by Bryan Dyck <bd...@mac.com>.
On 22-Jun-05, at 3:15 PM, Stephen Davis wrote:
> I believe you can also just set the following in your environment
> before configuring apache2. This prevents apache2 from using poll.
>
> export ac_cv_func_poll=no
Well, the early report from a rebuild of apr/apr-util-0.9.6,
httpd-2.0.54 and subversion-1.2.0 is that that little flag fixed the
problem - the first checkout from the repository completed without
error.
I'm going to do some more testing just to make sure that the issue is
resolved, as I haven't loaded all my data back into Subversion and
I'll probably create a test repos with some > 64K files to see how it
behaves - the only files in my Java projects that are larger than 64k
are .jar files. :)
Thanks again for the help.
Cheers,
b
Re: Checkout over HTTP fails with seemingly random errors
Posted by Stephen Davis <su...@soundgeek.org>.
On Jun 22, 2005, at 2:36 PM, Bryan Dyck wrote:
> On 22-Jun-05, at 1:30 PM, Ben Collins-Sussman wrote:
>
>
>> On Jun 22, 2005, at 2:00 PM, Bryan Dyck wrote:
>>
>>
>>>
>>>
>>> svn: Invalid diff stream: [tgt] insn 1 starts beyond the target
>>> view position
>>>
>>> svn: Unexpected end of svndiff input
>>>
>>> svn: REPORT request failed on '/svn/prima-survey/!svn/vcc/default'
>>> svn: REPORT of '/svn/prima-survey/!svn/vcc/default': Chunk
>>> delimiter was invalid
>>>
>>> svn: Checksum mismatch for 'prima-survey/lib/log4j-1.2.9.jar';
>>> expected: 'c2ced5f8505fe9d1cae685201e9cba07', actual:
>>> 'be72ae3be573c017a8906647c8b13dd4'
>>>
>>> [error] [client 192.168.1.1] Provider encountered an error while
>>> streaming a REPORT response. [500, #0]
>>> [error] [client 192.168.1.1] A failure occurred while driving the
>>> update report editor [500, #190004]
>>>
>>
>> This might be the famous bug in APR that only revealed itself on
>> Tiger recently, something to do with APR using poll instead of
>> select, or vice versa. The symptom was apache essentially unable
>> to stream >64k files to the client.
>>
>> The bug has been fixed in the latest HEAD of apr and apr-util.
>>
>> As an experiment, you might want to checkout the latest apr/apr-
>> util/httpd, and build them all from source and install. Then
>> rebuild Subversion, since apr 1.X is binary-incompatible at
>> runtime with apr-0.X. See if it clears things up.
>>
>
> Hmm... that sounds like an ugly bug. If that's the case, I'm
> relieved to hear it, because I was beginning to wonder if I was
> going slightly crazy.
>
> I'll try checking out the latest apr/apr-util/httpd and rebuilding
> the whole show... Hopefully that clears up the problem. Either way,
> I'll report back with the results.
>
> Thanks for the quick response and keep up the awesome work, Ben
> (and the rest of the SVN team).
>
> Cheers,
> b
>
I believe you can also just set the following in your environment
before configuring apache2. This prevents apache2 from using poll.
export ac_cv_func_poll=no
hth,
stephen
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Checkout over HTTP fails with seemingly random errors
Posted by Bryan Dyck <bd...@mac.com>.
On 22-Jun-05, at 1:30 PM, Ben Collins-Sussman wrote:
>
> On Jun 22, 2005, at 2:00 PM, Bryan Dyck wrote:
>
>>
>>
>> svn: Invalid diff stream: [tgt] insn 1 starts beyond the target
>> view position
>>
>> svn: Unexpected end of svndiff input
>>
>> svn: REPORT request failed on '/svn/prima-survey/!svn/vcc/default'
>> svn: REPORT of '/svn/prima-survey/!svn/vcc/default': Chunk
>> delimiter was invalid
>>
>> svn: Checksum mismatch for 'prima-survey/lib/log4j-1.2.9.jar';
>> expected: 'c2ced5f8505fe9d1cae685201e9cba07', actual:
>> 'be72ae3be573c017a8906647c8b13dd4'
>>
>> [error] [client 192.168.1.1] Provider encountered an error while
>> streaming a REPORT response. [500, #0]
>> [error] [client 192.168.1.1] A failure occurred while driving the
>> update report editor [500, #190004]
>>
>>
>
> This might be the famous bug in APR that only revealed itself on
> Tiger recently, something to do with APR using poll instead of
> select, or vice versa. The symptom was apache essentially unable
> to stream >64k files to the client.
>
> The bug has been fixed in the latest HEAD of apr and apr-util.
>
> As an experiment, you might want to checkout the latest apr/apr-
> util/httpd, and build them all from source and install. Then
> rebuild Subversion, since apr 1.X is binary-incompatible at runtime
> with apr-0.X. See if it clears things up.
Hmm... that sounds like an ugly bug. If that's the case, I'm relieved
to hear it, because I was beginning to wonder if I was going slightly
crazy.
I'll try checking out the latest apr/apr-util/httpd and rebuilding
the whole show... Hopefully that clears up the problem. Either way,
I'll report back with the results.
Thanks for the quick response and keep up the awesome work, Ben (and
the rest of the SVN team).
Cheers,
b
Re: Checkout over HTTP fails with seemingly random errors
Posted by Ben Collins-Sussman <su...@collab.net>.
On Jun 22, 2005, at 2:00 PM, Bryan Dyck wrote:
>
>
> svn: Invalid diff stream: [tgt] insn 1 starts beyond the target
> view position
>
> svn: Unexpected end of svndiff input
>
> svn: REPORT request failed on '/svn/prima-survey/!svn/vcc/default'
> svn: REPORT of '/svn/prima-survey/!svn/vcc/default': Chunk
> delimiter was invalid
>
> svn: Checksum mismatch for 'prima-survey/lib/log4j-1.2.9.jar';
> expected: 'c2ced5f8505fe9d1cae685201e9cba07', actual:
> 'be72ae3be573c017a8906647c8b13dd4'
>
> [error] [client 192.168.1.1] Provider encountered an error while
> streaming a REPORT response. [500, #0]
> [error] [client 192.168.1.1] A failure occurred while driving the
> update report editor [500, #190004]
>
This might be the famous bug in APR that only revealed itself on
Tiger recently, something to do with APR using poll instead of
select, or vice versa. The symptom was apache essentially unable to
stream >64k files to the client.
The bug has been fixed in the latest HEAD of apr and apr-util.
As an experiment, you might want to checkout the latest apr/apr-util/
httpd, and build them all from source and install. Then rebuild
Subversion, since apr 1.X is binary-incompatible at runtime with
apr-0.X. See if it clears things up.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org