[sheepdog-users] Write till the end of the cluster

Valerio Pachera sirio81 at gmail.com
Thu Dec 13 12:38:53 CET 2012

Hi, I'm testing what happens if write data to a guest till I consume
all the phisical space of the cluster.

3 nodes; 1 device of 1G for each node; 1 vdi disk of 10G virtual space.

I run the guest by a live cd (systemrescue)
  collie vdi create test 10G
  kvm -drive file=sheepdog:test -cdrom
/tmp/systemrescuecd-x86-3.1.1.iso -boot order=d -k it -vnc :1 -m 512

Inside the guest I create a single xfs partition and mount it.
Then I run
  dd if=/dev/zero of=/mnt/sda1/bigfile bs=1M

It fast reaches the end of the physical space
  collie node info
  Id      Size    Used    Use%
   0      982 MB  980 MB   99%
   1      982 MB  980 MB   99%
   2      982 MB  980 MB   99%
  Total   2.9 GB  2.9 GB   99%

I stop dd by ctrl+c.
If run dmesg on the guest I see lot's of I/O errors.

* dd never hangs up: it never gives 'no space left on the device',
probably because the virtual size i bigger.
* I tried to run dd several times using different output files
(bigfile1, bigfile2...). Their size is:
    bigfile 1,5G; bigfile1 0G; bigfile2 0G;
  What is strange is that 'df -h' shows 3.2G!
  And if I continue running dd, it continues to grow up.
* I'm able to unmount and mount the device (sda1), so it seem the
filesystem doesn't get corrupt.

I'm soon going the mess up the things even more :-)
Do you have any observation to share about this test?

More information about the sheepdog-users mailing list