[sheepdog-users] object cache questions

[Maxim Terleckiy]

02.02.2014 13:53, Valerio Pachera ?????:
>
> 2014-02-01 Maxim Terletskiy <terletskiy at emu.ru 
> <mailto:terletskiy at emu.ru>>:
>
>     the reason before found that "nodiratime" is root of the evil. 
>
>     Hope this information would be usefull for somebody.
>
>
> Sure it  is.
> Has any body else been able to reproduce this problem?
> What version of sheep are you running?
>
For me "nodiratime" problem arises with QEMU 1.6.1, 1.6.2, 1.7.0. Using 
CentOS 6.5 x86_64, sheep version 0.7.6 with zookeeper, object cache on 
ext4 partition, dual-nic with bonded ethernet for cluster communication 
and IPoIB for IO. Sheep options I'm using on VM nodes:
sheep -g -c 
zookeeper:172.16.0.15:2181,172.31.200.249:2181,172.31.200.250:2181,timeout=30 
-y 172.16.0.10 -i host=192.168.10.10,port=7001 -w 
dir=/mnt/sheep_objcache,size=213000

>     Now I'm curious about object cache failover. What happen if volume
>     with cache will fail? Will sheep and VMs live or will they die?
>
>
> I'm also curious.
> I think it's not handled as of now.
> It would be good that, in case of I/O error from the cache device, 
> sheep disables cache (like a node md unplug).
Morita Kazukata writes that VMs must survive. I will try to do some tests.
>
>     (I found that with restart of sheep process VMs connected to it dying)
>
>
> They shouldn't, but this depends on your qemu version.
> Qemu supports auto-reconnect. As soon as sheep daemon is available 
> again, the vm will issue their I/O requests.
Tried twice with QEMU 1.6.1. Both times result was negative.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.wpkg.org/pipermail/sheepdog-users/attachments/20140203/26f4c1de/attachment-0005.html>