[sheepdog-users] sheepdog 0.8 cache issues

Gerald Richter - ECOS richter at ecos.de
Fri Feb 21 14:14:36 CET 2014


I finally upgraded to sheepdog 0.8 and qemu 1.7, which should have a better cache behavior.

Basically I am doing the same as with my previous setup. I import 5 images at the same time into sheepdog. I have make sure that there is enough space on disk, so that all 5 images would fit into the cache without running out of diskspace.

I am running a command like “qemu-img convert -p -t writeback vm.vhd sheepdog:vm-61036-disk-1”, doing this with 5 different images in parallel (using & on the shell). Everything gets imported, but most of the data remains in the cache and will not be distributed to the other nodes:

#dog vdi cache info

Name   Tag         Total      Dirty      Clean

vm-61025-disk-1                             15 GB    15 GB    36 MB

vm-61026-disk-1                             20 GB    20 GB    40 MB

vm-61037-disk-1                             20 GB    0.0 MB  20 GB

vm-61035-disk-1                             18 GB    18 GB    36 MB

vm-61036-disk-1                             20 GB    20 GB    40 MB

Cache size 100 GB, used 93 GB, non-directio

Regardless how long I wait, that cache never gets flushed.

Sheep is started with:

/usr/sbin/sheep --pidfile /var/run/sheep.pid --nosync /var/lib/sheepdog/ /var/lib/sheepdog//disc1/data,/var/lib/sheepdog//disc2/data -w dir=/var/lib/sheepdog//cache size=100G

Regards Gerald

P.S. By the way it would be nice to make the Name column wider, so there is enough space to display names that not only consist of a few characters  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.wpkg.org/pipermail/sheepdog-users/attachments/20140221/461bc6e7/attachment-0004.html>

More information about the sheepdog-users mailing list