[openstack-dev] [oslo] rpc concurrency control rfc

Duncan Thomas duncan.thomas at gmail.com
Tue Mar 4 16:15:03 UTC 2014


On 28 November 2013 10:14, Daniel P. Berrange <berrange at redhat.com> wrote:

> For this specific block zero'ing case it occurred to me that it might
> be sufficient to just invoke 'ionice dd' instead of 'dd' and give it
> a lower I/O priority class than normal.

Excuse the thread necromancy, I've just been searching for thoughts
about this very issue. I've merged a patch that does I/O nice, and it
helps, but it is easy to DoS a volume server by creating and deleting
volumes fast while maintaining a high i/o load... the zeroing never
runs and so you run out of allocatable space.

I'll take a look at writing something with more controls than dd for
doing the zeroing...



More information about the OpenStack-dev mailing list