[openstack-dev] [neutron][nova][stable][sr-iov] Status of physical_device_mappings

Jay Pipes jaypipes at gmail.com
Wed Mar 23 20:01:04 UTC 2016

+tags for stable and nova

Hi Vladimir, comments inline. :)

On 03/21/2016 05:16 AM, Vladimir Eremin wrote:
> Hey OpenStackers,
> I’ve recently found out, that changing of use neutron sriov-agent in Mitaka from optional to required[1] makes a kind of regression.

While I understand that it is important for you to be able to associate 
more than one NIC to a physical network, I see no evidence that there 
was a *regression* in Mitaka. I don't see any ability to specify more 
than one NIC for a physical network in the Liberty Neutron SR-IOV ML2 agent:


> Before Mitaka, there was possible to use any number of NICs with one Neutron physnet just by specifying pci_passthrough_whitelist in nova:
>      [default]
>      pci_passthrough_whitelist = { "devname": "eth3", "physical_network": "physnet2”},{ "devname": "eth4", "physical_network": "physnet2”},
> which means, that eth3 and eth4 will be used for physnet2 in some manner.

Yes, *in Nova*, however from what I can tell, this functionality never 
existed in the parse_mappings() function in neutron.common.utils module.

> In Mitaka, there also required to setup neutron sriov-agent as well:
>      [sriov_nic]
>      physical_device_mappings = physnet2:eth3
> The problem actually is to unable to specify this mapping as "physnet2:eth3,physnet2:eth4” due to implementation details, so it is clearly a regression.

A regression means that a change broke some previously-working 
functionality. This is not a regression, since there apparently was 
never such functionality in Neutron.

> I’ve filed bug[2] for it and proposed a patch[3]. Originally physical_device_mappings is converted to dict, where physnet name goes to key, and interface name to value:
>      >>> parse_mappings('physnet2:eth3’)
>      {‘physnet2’: 'eth3’}
>      >>> parse_mappings('physnet2:eth3,physnet2:eth4’)
>      ValueError: Key physnet2 in mapping: 'physnet2:eth4' not unique
> I’ve changed it a bit, so interface name is stored in list, so now this case is working:
>      >>> parse_mappings_multi('physnet2:eth3,physnet2:eth4’)
>      {‘physnet2’: [‘eth3’, 'eth4’]}
> I’d like to see this fix[3] in master and Mitaka branch.

I understand you really want this functionality in Mitaka. And I will 
leave it up to the stable team to determine whether this code should be 
backported to stable/mitaka. However, I will point out that this is a 
new feature, not a bug fix for a regression. There is no regression 
because the ability for Neutron to use more than one NIC with a physnet 
was never supported as far as I can tell.


> Moshe Levi also proposed to refactor this part of code to remove physical_device_mappings and reuse data that nova provides somehow. I’ll file the RFE as soon as I figure out how it should work.
> [1]: http://docs.openstack.org/liberty/networking-guide/adv_config_sriov.html
> [2]: https://bugs.launchpad.net/neutron/+bug/1558626
> [3]: https://review.openstack.org/294188
> --
> With best regards,
> Vladimir Eremin,
> Fuel Deployment Engineer,
> Mirantis, Inc.
> __________________________________________________________________________
> 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

More information about the OpenStack-dev mailing list