On Wed, 25 Aug 2004, Kurt Mosiejczuk wrote: >Periodically, my users stop being able to access email via imap. >Mozilla Thunderbird will just start spinning the hourglass forever. On >the server, the load average starts climbing. What is happening is that >one bincimapd process (usually being one of my users) is going berserk >and monopolizing NFS traffic. My NetApp ends up being pegged at
Whoaa, nasty. :-( >2600-2800 operations/second until I either reboot the mail server or >kill the process. Killing the process I've only done once so far when I >smartened up and added 'intr' to the mount options for the nfs filesystems. >Until I was able to kill the one bincimapd process and have the machine >return to normal, I was leaning towards it being something in OpenBSD's >NFS implementation that binc might happen to tickle. And I suppose that >hasn't been ruled out, but with killing the rogue process returning >things to normal, I'm not suspecting that it may be something binc is doing. Binc doesn't care about NFS, it should work perfectly fine. >Fortunately, because of other bizarreness, I had put recordio in the >pipe and have a protocol trace of the rogue bincimapd from last night. >I haven't yet isolated the trace for the one that happened an hour ago. >If there's any other detail I've missed or might be able to gather for >next time, let me know and I'll get it. For now, I've attached the >protocol trace to this message. Thanks! Also, please attach to the haywire bincimapd process with gdb if you can, and do a backtrace on it. This will show exactly where Binc is spinning. I'll have to look at it this weekend, but it'll speed things up if you can provide a backtrace. (Binc needs to be compiled with debug symbols). Thanks for the report! I'm sure we can fix this. Andy :-) -- Andreas Aardal Hanssen | http://www.andreas.hanssen.name/gpg Author of Binc IMAP | "It is better not to do something http://www.bincimap.org/ | than to do it poorly."
