As of today, when attempting a commit I get the following: Sending trunk/pym/emerge/__init__.py Transmitting file data .svn: Commit failed (details follow): svn: 'pre-commit' hook failed with error output: (there is no further error output after that, just a blank line) That's with both the portage and the gentoolkit repo. $ svn --version svn, version 1.4.2 (r22196) compiled Nov 21 2006, 02:38:40
Could it be related with neysx@basil ~/gentoo.org/gentoo/xml/htdocs/doc/en $ cvs commit -m'foo' test.xml XML-Checking test.xml ... [ done ] /var/cvsroot/gentoo/xml/htdocs/doc/en/test.xml,v <-- test.xml new revision: 1.74; previous revision: 1.73 cvs [commit aborted]: received broken pipe signal cvs [commit aborted]: received broken pipe signal I guess both mails to cia and gento-doc-cvs ML trigger an error. Those commits do complete, though.
According from what I understood, solar tried to track down the issue with rejecting mails from stork->woodpecker with changing sysctl-variables. Already reverted for the time (see core-messages).
It's happening again.
Hmm. I checked the logs on stork, but can't find any problems there. genone, is the error the same like before? robbat2, can you see something strange?
(In reply to comment #4) > genone, is the error the same like before? Yeah, the symptom I'm seeing with the portage repo is exactly as described in comment #0. Like before, there's just a blank line where the error output is supposed to be.
Yeah, same error (= no error message) as before, and the reports in -dev say that it's also the same error for cvs. If this is related to CIA (what most people seem to think) and there is no known solution short term could we disable the CIA script temporary so we can commit things instead of letting them pile up?
Yes, it could be that script. exim's last log entry was at 2007-05-17 12:41:25 1HofIK-00020T-98 Completed I just restarted exim to be sure, that this is not the troublemaker. Disabling ciabot does not make any difference. Commit-diffs to the docs which should be mailed to the doc-cvs-mailing-list are not generated. So the error must be somewhere earlier. svn's pre-commit hook does not call ciabot… Also, there were no updates on the machine lately. And no changes at files in /etc. And, cvs commits go through, but the commit-break-message is shown. svn commits break before and are not accepted. I'm currently out of ideas. I would just reboot the machine and see if it still happens ;-) But I guess, we should track down the problem. I'm not an expert with sysctl and ulimit, so a helping hand is needed here.
Some update on this, working on CVS first. I did an strace on a commit on stork directly, and found that ciabot.pl is never actually launched by CVS.
Ok, I traced it all. There were actually two breakages, and the first one made the second one hard to find. 1. Somebody broke the cvslogdate commitinfo task. The magic directory of /var/log/cvs-access/ was completely gone. I recreated it mode 1777 as it should be (I added devrel here, because they were the ones that used that task). 2. nscd had failed, and was causing a lot of other failures when the system failed to look up the uid from LDAP in a suitable time. I restarted it for now. We should get nagios to check for nscd, and auto-restart it as needed.