[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Bug: apache/mod_svn intermittently creates transaction dirs with incorrect permissions or ownership such that the transaction breaks.

From: John Duprey <john.duprey_at_gmail.com>
Date: 2005-09-16 19:48:04 CEST

I'm about to file a subversion bug in relation to apache/mod_svn
intermittently creating transaction dirs with incorrect permissions or
ownership such that the transaction breaks.

Many times during the day commits are failing with an error message like the
following:

*[Wed Aug 31 12:55:55 2005] [error] [client
192.189.224.121<http://192.189.224.121>
<http://192.189.224.121>] *
*Could not create activity *
*/svn/ResultsPlus/!svn/act/a24ba835-0155-2042-a734-13818c448007. [500, #0] *
*[Wed Aug 31 12:55:55 2005] [error] [client
192.189.224.121<http://192.189.224.121>
<http://192.189.224.121>] *
*could not begin a transaction [500, #13] *
*[Wed Aug 31 12:55:55 2005] [error] [client
192.189.224.121<http://192.189.224.121>
<http://192.189.224.121>] *
*Can't open file
'/svnroot/repos/ResultsPlus/db/transactions/90-1.txn/node.0.0':
*
*Permission denied [500, #13]
*
Inspection on the server side reveals one of 2 scenarios:

   1. The transaction directory (e.g.
   /svnroot/repos/reposname/db/transactions/90-1.txn) has the wrong
   permissions and cannot be populated by the apache user (apache) - i.e.
   drw-rwSrw-
   2. The props file in the transaction directory (e.g.
   /svnroot/repos/reposname/db/transactions/90-1.txn/props) is owned by
   the root user and ONLY user has write permissions

If the commit is retried, it may or may not succeed. To temporarily fix the
problem, I must restart apache, and delete the broken transaction
directories.

I am now running the latest stable apache(httpd-2.0.54) and subversion(
subversion-1.2.3) built from scratch. (This problem was first observed with
prebuilt RPMS.) I am running Intel RedHat Enterprise Server 3 (rhel-3). I
have audited the system extensively in an effort to identify any external
processes that could be causing this problem and found now.

I have a little bash script that will repeatedly modify, commit, and sleep 1
second that tests subversion. I can reproduce this error within 10 - 30
commits. I posted this problem before (
http://svn.haxx.se/users/archive-2005-09/0228.shtml). More details can be
found in that post. This problem was reported by another user as well in
June (http://svn.haxx.se/users/archive-2005-06/1629.shtml) -- on Solaris. He
was able to work around the problem by using setfacl however, rhel-3 (kernel
2.4) doesn't appear to support setfacl completely.

If anyone can help me debug this problem further I'd appreciated it. I'm
willing to try almost anything - a debug version of subversion, apache etc.
If you'd like more information, please ask me. Otherwise, I'd welcome any
tips for successfully logging a bug on
subversion.trigris.org<http://subversion.trigris.org>.

I feel I've exhausted all configuration possibilities between
apache-subversion. I've tried creating a test repository from scratch -
ensuring all commands out-of-apache-web-server were done as apache (sudo -u
apache svnadmin ...). I've tried different file system locations. I've
removed all but svn functionality from apache. I've used the simplest svn
configuration possible..

Lost,
-John Duprey
Received on Fri Sep 16 19:51:03 2005

This is an archived mail posted to the Subversion Dev mailing list.