Open Stack

On 08/03/2016 09:13 AM, Tuan Luong wrote:
> Hi Jay, Thank you for the reply. Actually, I did ask this question to make
> sure that everything is still going well with disk usage report of Nova. We
> had the problem related to wrong report (IMHO it is wrong). The situation is
> below:
>
> - We launch an instance with 600Gb of ephemeral disk of flavor without
> specifying extra ephemeral disk. The total free_disk before launching is
> 1025Gb, the total disk_available_least is 1016Gb. - After successful
> launching, what we have is the value of disk_available_disk = -184Gb
> therefore we could not launch the next instance.
>
> It seems like the value of disk_available_disk decreases 2*600Gb when I think
> it should be 600Gb. The value of free_disk after launching is calculated
> well. I would not believe that the disk_over_commit in this case is 600Gb.
> Otherwise, we also check the instance that indeed it has only 600Gb mounted
> as vdb.

Assuming you're using qcow, the maximum disk space consumed by a qcow disk is 
"size of backing file" + "size of differences from backing file".  Thus, if you 
have a single instance using a given backing file, the worst-case consumption is 
actually twice the size of your disk.

I think the backing file for an ephemeral disk is actually a sparse file, but on 
a cursory examination it will appear to be consuming 600GB.

Chris