[sheepdog-users] About vdi check

Valerio Pachera sirio81 at gmail.com
Wed May 22 18:49:31 CEST 2013


2013/5/22 Liu Yuan <namei.unix at gmail.com>:
> notice any situation that 'vdi check' saturate the network?

Not exactly, but I saw eth0 running at 430M/s!
I'm not using my test cluster now, I'm on the production cluster that
is still running 0.5.5_335_g25a93bf.

The situation is that:
I killed a node (using md) and Insert it back (as you might read in a
previous mail).
Recovery started and finished (two guests were running).
Later I noticed my guests were not working (qemu proccess using
100%cpu, not even vnc was working).
I kill -9 them (I forgot to flush the cache first ... damn!)
I tried to start them back, but they stop right after, complaining
about dry I/O error.

I run vdi check to "see" if cluster was intact.
Right now I'm running vdi check on big disks (100G and 1T).
Eth0 runs even over 400M/s

sheepdog004
NET | eth0     37%  |              |  pcki  312609 | pcko   51712  |
           |  si  372 Mbps | so   36 Mbps
sheepdog002
NET | eth0     22%  |              |  pcki  194262 | pcko  189016  |
           |  si  222 Mbps | so  185 Mbps

Till now it didn't print any missing chunk.
On sheepdog004 I'm checking 2 vdi's.
On sheepdog002 I'm checking 1 vdi's.


More information about the sheepdog-users mailing list