| blkperl | slangasek: I'm mad about this | 21:46 |
|---|---|---|
| blkperl | https://lists.ubuntu.com/archives/ubuntu-devel/2010-September/031493.html | 21:46 |
| blkperl | grumpy... | 21:46 |
| blkperl | so I have to modify the init script for slapd | 21:48 |
| blkperl | ? | 21:48 |
| slangasek | blkperl: so, subsequent to that mail, the compromise we have in place is that the default *soft* limit for open files is 1024, but the hard limit is higher, at 4096 | 22:06 |
| slangasek | blkperl: which means that any process which needs more files can call setrlimit() to raise its own limit, privileged or not | 22:06 |
| blkperl | slangasek: well slapd is either not doing that or needs more than 4096 | 22:08 |
| slangasek | wasn't it y'all's slapd server that was leaking fds once before? | 22:09 |
| blkperl | yeah we fixed that :P | 22:09 |
| slangasek | and now it's running out of fds for legitimate reasons? :) | 22:09 |
| blkperl | last time is was memory | 22:09 |
| blkperl | not fds | 22:10 |
| slangasek | ok | 22:11 |
| slangasek | do you have more than 1000 simultaneous clients? | 22:11 |
| blkperl | nope | 22:11 |
| blkperl | Oct 10 11:58:03 arya slapd[929]: warning: cannot open /etc/hosts.deny: Too many open files | 22:12 |
| blkperl | getting those in the logs | 22:12 |
| slangasek | how many simultaneous clients *do* you have? | 22:13 |
| slangasek | raising the fds is doable, but if the real problem is an fd leak... | 22:13 |
| slangasek | (lsof -p $(pidof slapd) -n?) | 22:13 |
| blkperl | 827 at the moment | 22:15 |
| blkperl | now 858 | 22:15 |
| blkperl | hmm maybe your right and its leaking | 22:16 |
| blkperl | well it matches with number of connections though | 22:19 |
| slangasek | how are you measuring the number of connections? *should* there be that many simultaneous clients? | 22:24 |
| slangasek | if you're just saying slapd agrees with itself about how many connections there are"... | 22:25 |
| blkperl | slangasek: idk, nfs/autofs/mail all use it I can't tell if they should be there | 22:27 |
| slangasek | ok | 22:28 |
| slangasek | is that on the right order of magnitude for the number of client systems you have? | 22:29 |
| slangasek | blkperl: I guess I would say, cowboy a change to the slapd init script to add 'ulimit -n 4096' and see if the fd usage levels off | 22:30 |
| slangasek | blkperl: btw, if you feel strongly about not editing the init script, you could drop it in /etc/default/slapd instead ... and it would work ;) | 22:31 |
| blkperl | slangasek: how can you put it in /etc/default/slapd its a config file? | 22:33 |
| slangasek | blkperl: <cough> it's a shell fragment that's sourced by the init script | 22:35 |
| blkperl | oh.. is that because ulimit is a shell builtin | 22:53 |
| blkperl | so source just works | 22:53 |
| blkperl | hmm its back down to 419 | 23:09 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!