[ovs-dev] OVS bridges always report UP + hotplugging support?

Thadeu Lima de Souza Cascardo cascardo at redhat.com
Fri Nov 13 18:35:25 UTC 2015


On Fri, Nov 13, 2015 at 04:30:12PM -0200, Thadeu Lima de Souza Cascardo wrote:
> On Fri, Nov 13, 2015 at 09:56:50AM -0800, Karol Mroz wrote:
> > Hi Thadeu,
> > 
> > On Thu, Nov 5, 2015 at 4:51 AM, Thadeu Lima de Souza Cascardo <
> > cascardo at redhat.com> wrote:
> > 
> > > On Wed, Nov 04, 2015 at 02:06:56PM -0800, Karol Mroz wrote:
> > > > Hello,
> > > >
> > > > Support for configuring OVS bridges has been added to our (SUSE) network
> > > > manager.
> > > > During test (OVS 2.1.2, kernel 3.12.x), a couple things have been
> > > uncovered
> > > > that
> > > > I'd love to get some feedback on.
> > > >
> > > > 1. OVS bridges appear to report up state, regardless if this is true. For
> > > > example,
> > > > up is reported even if the underlying bridge port is down, or internally
> > > > OVS has
> > > > not yet set up the necessary flows to allow for data transmission. Is
> > > this
> > > > something
> > > > that is planned to be addressed? I realize 2.1.2 is an older version, so
> > > > perhaps
> > > > this has already been addressed in later versions of OVS?
> > > >
> > > > 2. OVS does not appear to handle hotplugging. For example, if a port,
> > > which
> > > > is attached
> > > > to the an OVS bridge (thus is correctly reflected in ovsdb) arrives on
> > > the
> > > > system _after_
> > > > the actual `add-port` operation was performed, it will not get enslaved
> > > > under the ovs-system
> > > > interface. In this case, a manual `del-port/add-port` is needed. Is there
> > > > any plan to
> > > > add netlink monitoring of interface states (NEWLINK/DELLINK, etc) +
> > > > enslavement
> > > > when appropriate?
> > > >
> > >
> > > This has been implemented. Check commit
> > > e21c6643a02c6b446d2fbdfde366ea303b4c2730, which is not present in any
> > > release
> > > yet.
> > >
> > 
> > Yes, I see. Thank you. This commit appears to rely on some additional
> > infrastructure being
> > in place (ie. various rtnetlink_*() APIs). Do you have some idea what
> > upcoming release would
> > contain this feature.
> 
> Well, if you are on Linux, there is no problem. In fact, this was even
> implemented and tested on FreeBSD. Unless you have a kernel configured without
> rtnetlink, it will work. And if you don't have rtnetlink, other thinks wouldn't
> work on OVS as well. So, I don't understand what's your concern about it relying
> on that API on OVS.

I missed the fact that you may consider backport the feature. In that case,
using new API may be a problem. But I believe it would be easy to backport it to
the 2.4 series. 2.1, on the other hand, I guess it may be harder.

Cascardo.

> 
> This feature should appear in the next release of OVS, whatever it's numbered,
> 2.5.0, or 3.0.
> 
> > 
> > Further, would this be useful in addressing point #1 above?
> > 
> 
> I don't know too much about point #1, but this feature does not address that,
> sorry.
> 
> Do you mean that OVS reports the UP state to a controller using OpenFlow?
> 
> Cascardo.
> 
> > Thanks,
> > Karol
> _______________________________________________
> dev mailing list
> dev at openvswitch.org
> http://openvswitch.org/mailman/listinfo/dev



More information about the dev mailing list