Open Stack

On 17 April 2014 17:37, Deepak Shetty <dpkshetty at gmail.com> wrote:

> forgot to add, that for the above to work consistently, we definitely need
> to have a framework / mechanism in cinder where the driver is provided a
> function/callback to gracefully cleanup its mounts (or any other thing)
> during service goign down. Today we don't have such a thing and hence
> drivers don't cleanup their mounts and hence when service starts up, it
> ensure_shares_mounted sees the mount being present and does nothing. this
> would work nice if drivers were given ability to clean up the mounts as part
> of service going down.

While adding a graceful shutdown hook is not necessarily a bad idea,
the fact that drivers don't check all of the details of the mounts and
update as necessary is a bug and should be filed & addressed. Config
changes should not be missed just because there was an unclean
shutdown, and all services and drivers should correctly handle
recovery from an unclean shutdown.


-- 
Duncan Thomas