[openstack-dev] [Quantum] RPC support for OVS

Isaku Yamahata yamahata at valinux.co.jp
Tue Aug 7 06:56:38 UTC 2012


On Mon, Aug 06, 2012 at 02:56:16PM +0300, Gary Kotton wrote:
> Hi,
> Please take a look at https://review.openstack.org/#/c/10879/ for the RPC
> support for the OVS. I would appreciate it if the guys familiar with the OVS
> and RYU agents would take a look and let me know if they have any comments.

Regarding to Ryu plugin, the plain is to make agent simpler.
Ideally, agent would do only local initializations, and polling would be
totally eliminated. Instead, Ryu server will communicate with each ovsdbs
remotely.
So the effort to convert Ryu agent for RPC notification would be wasted.

Using RPC seems very good for that direction. Probably Ryu server would
take advantage of RPC. Not agent.


> At the moment I have labeled it as WIP due to the fact that there are a number
> of things that still need to be done:
> 1. I need to treat the tunneling, that is, in the current implementation the
> local IP is added to the database and then the agent polls the database to read
> if there are additional tunnels (please let me know if I have misunderstood
> something here). This needs to be supported by the RPC mechanism (I am
> currently coding)
> 2. The code depends the RPC support for the Linux bridge https://
> review.openstack.org/#/c/9591/ (this has some utilities that are used by this
> code).
> 3. There is a lot of "common" code between the Linux bridge and ovs plugins for
> the RPC support. Once the above patch is approved I'll make the code common so
> that it can be used by other agents (ryu, dhcp agent etc.)
> 4. I need to add unit tests. The initial testing has been done manually.
> 
> In addition to this I have stumbled upon a problem with the ovs agent. This
> happens when the agent starts. There are two traces below. The first trace is
> where everything works OK and the second trace is where there is an exception.
> The same ovs-vsctl command is used. There are just different return codes.

It seems that self.patch_int_ofport is initialized to -1.
I'm not sure why it happens, though.
-- 
yamahata



More information about the OpenStack-dev mailing list