[ovs-dev] [PATCH v4 3/3] netdev-dpdk: vHost client mode and reconnect

Daniele Di Proietto diproiettod at ovn.org
Mon Aug 15 17:09:36 UTC 2016


Hi Sean,

2016-08-15 9:04 GMT-07:00 Mooney, Sean K <sean.k.mooney at intel.com>:

> Hi Daniele
> Sorry to top post but I have just read back over the last
> couple of revisions of this patch.
>

No problem, thanks for stepping in.  As I said many times during the review
of this series I'm not sure what the best interface would be, and I really
appreciate any feedback on this.


> I noticed that you requested that the vhost-driver-mode flag be removed
> From the Open_vSwitch table. The vhost-driver-mode flag was included in
> The original patch for two reasons 1 to configure the global driver mode
>

In my opinion Open vSwitch is not the right place to store this global
configuration parameter.  I think we should put it as high in the stack as
possible.  Isn't there another place to store it in OpenStack?


> And 2 to provide a way to detect if reconnect/qemu server mode was
> available.
>

I see your point, we need feature detection for this.

Perhaps we can use another type for client ports, like
"dpdkvhostuserclient".  I think it make senses to have a separate class,
since the interface is different anyway.  It would be easy then to detect
the feature based on the available iface_types.

What do you think?

Thanks,

Daniele


>
> Without the global flag or a similar mechanism to expose the capability via
> The ovsdb I cannot complete the OpenStack integrations.
> https://review.openstack.org/#/c/344997/
>
> I have one proposal which is to store the feature list currently in the faq
> https://github.com/openvswitch/ovs/blob/master/FAQ.md#q-are-all-features-
> available-with-all-datapaths
> in the ovsdb. This can be retrieve remotely via the ovs db by openstack or
> any other orchestrator to
> make dession based on the feature detected.
>
> If you have another suggestion I would be glad to adapt my OpenStack change
> To use another mechanism to detect the support for reconnect but without
> the
> vhost-driver-mode flag I am currently blocked.
>
> I will make as a separate thread not to discuss feature discovery in
> general
> Not to  distract from this review
> Regards
> Sean.
>



More information about the dev mailing list