subversion-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Steffen Moser <li...@steffen-moser.de>
Subject Re: Apache 2.4, Worker MPM, SVN 1.7.20, MOD_DAV_SVN and Post-Commit Hooks
Date Fri, 12 Feb 2016 16:24:16 GMT
Hi Philip,

On 2/12/16 2:19 AM, Philip Martin wrote:
> Steffen Moser <lists@steffen-moser.de> writes:
> 
>>  | Running tests in commit_tests.py [43/84]......................success
> 
> One of the tests in commit_tests.py is 'hook_test' which runs start-,
> pre- and post- hooks.  So it appears that your build can run hooks.

Thank you very much - this was a very important piece of information
which I was not aware of.

> Not clear why you get a hang in your normal configuration.  Perhaps one
> of the other Apache modules? 

It seems that this was the solution! After knowing that the test script
also checks the hooks, I was indeed sure that there must have been
something different between my build system and my standard configuration.

I bisected the list of modules which are loaded in my standard
configuration. Finally, it turned out that "mod_ssl" was the cause of
the problem. Actually, the problem was not caused by loading the Apache
module. It was rather triggered by the fact that in the test-clone of my
working environment where I encountered it, I accessed all of my SVN
repositories via HTTPS (of course, because our users login to access the
SVN repositories).

So I saw that I could avoid the problem by accessing the SVN repository
via HTTP instead of HTTPS.

After learning this, I had a look at the HTTPS part of my site-specific
Apache configuration (/etc/apache2/2.4/sites.d/...). There I found a
strange configuration line:

  SSLCryptoDevice pkcs11

from which I think that is was supposedly a relict of an UltraSPARC T2
configuration. The newer CPUs of the UltraSPARC platform come with a
hardware device to support hardware-based cryptography and Oracle seems
to have an individual set of patches for SSL to enable the access to
this device. Most probably (I am not sure yet), this line is just
useless on an x86_64 architecture which we use.

At least, the code introduced or activated with this line doesn't seem
to be thread-safe. So disabling this option solved the problem for me.

So far I can say that SVN now seems to work via HTTPS on our system
(Solaris 11.3, x86_64, Apache 2.4.12, SVN 1.7.20).

Thanks again for your help. Your hint about the fact that the test
script includes hook tests was very important and helpful.

Kind regards,
Steffen


-- 
------------------------------------------------------------------------
Dipl.-Inf. Steffen Moser
School of Advanced Professional Studies            Room: O27/317
Ulm University                                     Tel: +49.731.50-24179
Albert-Einstein-Allee 11                           Fax: +49.731.50-24182
89081 Ulm, Germany                               http://saps.uni-ulm.de/

Mime
View raw message