geom_raid5 livelock?

[CyberLeo Kitsana]

R. B. Riddick wrote:
> --- CyberLeo Kitsana <cyberleo at cyberleo.net> wrote:
>> http://home.cyberleo.net/cyberleo/workspace/Zip/graid5-testing.png
>>
> Aha...
> Looks like gmirror is quite busy, too, while idle at the same time...
> This might mean, that we have a new GEOM problem...
> 
>> As shown in top, one rsync and one find process are stuck in biord 
>> waiting for data from the filesystem, while the CPU is otherwise mostly 
>> quiescent. I can't find evidence anywhere of what could be using the 
>> raid5 and mirror providers so much, though.
>>
> Me, too...
> 
>> The frozen processes are one locate database update (find) and one rsync 
>> process copying data from an old backup machine to the new one. The 
>> frozen processes do not respond to attempts to kill them until the 
>> filesystem begins responding.
>>
> Hmm... Can u try without gmirror? :-)
> I mean: Just if it is not too complicated for u (I have no hard evidence, that
> gmirror is the cause of something)...
> 
>> Also, tuning the sysctl had no effect prior to, nor during, the 
>> livelock. Nothing new appeared in /var/log/messages while set to 1 or 2.
>>
> Hmm...
> That might show, that ur /etc/syslog.conf does not log every message... :-) Can
> u try the line "*.* /var/log/all" to /etc/syslog.conf (after editing that file,
> touch the log file ("touch /var/log/all") and then reload the syslogd conf
> ("/etc/rc.d/syslogd reload" or so))?

http://home.cyberleo.net/cyberleo/workspace/Zip/graid5-testing2.png

Disabled gmirror and booted from a single member of the array. The 
lower-right xterm is tailing all.log, enabled as you suggested in 
syslog, with kern.geom.raid5.debug = 2.

It was logging to /var/log/messages, four lines every 10 seconds or so. 
I was just impatient. ^.^

--
Fuzzy love,
-CyberLeo
Technical Administrator
CyberLeo.Net Webhosting
http://www.CyberLeo.Net
<CyberLeo at CyberLeo.Net>

Furry Peace! - http://www.fur.com/peace/