Open Stack

On Tue, Jul 21, 2015 at 02:24:14PM +0200, Dmitry Tantsur wrote:
> Hi folks!
> 
> If you're not aware already, I'm working on solving "node is locked"
> problems breaking users (and tracking it at
> https://etherpad.openstack.org/p/ironic-locking-reform). We have retries in
> place in client, but we all agree that it's not the eventual solution.
> 
> One of the things we've figured out is that we actually have server-side
> retries - in task_manager.acquire. They're nice and configurable. Alas, we
> have one place that checks reservations without task_manager: https://github.com/openstack/ironic/blob/master/ironic/api/controllers/v1/node.py#L401-L403
> (note that this check is actually racy)
> 
> I'd like to ask your opinions on how to solve it? I have 3 ideas:
> 1. Just implement retries on API level (possibly split away a common
> function from task_manager).
> 2. Move update to conductor instead of doing it directly in API.
> 3. Don't check reservation when updating node. At all.
> 
> Ideas?

So, it looks like the only reason we check the reservation field here is
because we want to return a 409 for "node is locked" rather than a 400,
right? do_node_deploy and such will raise a NodeLocked, which should do
the same as this check. It's unclear to me why we can't just remove this
check and let the conductor deal with it.

// jim

> 
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev