geom_raid5 livelock?

[R. B. Riddick]

--- CyberLeo Kitsana <cyberleo at cyberleo.net> wrote:
> R. B. Riddick wrote:
> http://home.cyberleo.net/cyberleo/workspace/Zip/graid5-all.log
> 
> That log contains two full seconds of requests, under the same 
> circumstances as before (rsync and find processes stuck in wait, along 
> with an 'ls' or two). If you need more, I can produce it.
> 
> It appears to be completing requests, but always the same requests over 
> and over. Watching the log for about a minute doesn't show any other 
> requests during this time.
> 
> [start] raid5/raid5[READ(offset=779825545216, length=16384)]
> [start] raid5/raid5[READ(offset=764410200064, length=16384)]
> 
Hmm...
Thats strange...
Possibly there happens an ENOMEM error, which would explain the repititions...

A further debug line (it is again at debug level 2) before the current line
1165 (inside the inbed==children IF but in the end of it):
G_RAID5_LOGREQ(bp, "[ready err%d cmp%jd]", obp->bio_error, obp->bio_completed);
Or turn on "bootverbose"...

For better data safety (e. g. in case of a power loss), I would recommend to
reduce kern.geom.raid5.wdt to 0 or 1 (the lower the safer).

For less memory consumtion I would use lower values for .maxmem and for
.maxwql...

-Arne


 
____________________________________________________________________________________
Now that's room service!  Choose from over 150,000 hotels
in 45,000 destinations on Yahoo! Travel to find your fit.
http://farechase.yahoo.com/promo-generic-14795097