Our support team suggested out-of-band that you were using OVS 1.0.99.

I set up OVS 1.0.99 (the tip of the "vlan-maint" branch) and
experimented briefly with a setup similar to the one that you
provided.  I did not set up a restrictive flow table or configure a
controller; that is unlikely to be key to the problem.  I did set up
just a single continuous "ping" between two VMs that otherwise had
idle networks, with nothing else going on.

The behavior I see so far is that the MAC learning table is doing what
I expect: the age of each entry never goes beyond 1 second, so the MAC
learning entries should not expire, and I don't see them expiring.  So
I am not yet able to reproduce the problem.

Here is an experiment that you can conduct in your setup: run "watch
ovs-appctl fdb/show <bridge>" in the Dom0 and observe the changes in
the MAC learning table.  As is my setup, you should not see the
entries for the laptop and vm1 age beyond 1 second or so, certainly
not enough to expire.  What do you actually see?

Thanks,

Ben.

On Thu, Dec 22, 2011 at 10:45:37AM -0800, Ben Pfaff wrote:
> OK.  So it seems that MAC learning entries are expiring in cases where
> we expect them to persist.  I can look into that, if you can give me
> some more details; to start, the version of OVS involved.  (I think
> that you might have already given detail to our support team in
> parallel; I'm trying to find out how I get direct access to that
> information.)
> 
> Let me reiterate that the "normal" action isn't an effective way to
> enforce ACLs.  Nevertheless, there appears to be a bug that I should
> investigate here.
> 
> Thanks,
> 
> Ben.
> 
> On Thu, Dec 22, 2011 at 06:35:50PM +0000, Mike Bursell wrote:
> > I believe that there is nothing else going on at all.
> > 
> > The CLI tools were used to construct the rules: no DVSC in play.
> > 
> > -Mike.
> > --
> > Mike Bursell.
> > 
> > 
> > 
> > Ben Pfaff <[email protected]> wrote:
> > 
> > 
> > On Thu, Dec 22, 2011 at 04:35:45PM +0000, Mike Bursell wrote:
> > > We've discovered what we suspect is a bug, and are looking for
> > > thoughts, please!
> > >
> > > Observed behaviour:
> > > - Continuous pings being sent from laptop to vm1
> > > - vm2 is quiescent
> > > - Intermittently, the response to a ping from laptop is seen on vm2
> > 
> > Is anything else going on?  Certain kinds of changes to a bridge
> > (adding and removing ports, etc.) can cause the MAC learning table, or
> > particular entries in it, to be flushed.  If VMs are being brought up
> > or down, VLANs being created or destroyed, etc., one might expect to
> > see a need to re-learn MAC addresses immediately after those events.
> > 
> > I have not carefully looked over your flow table.  Is this flow table
> > constructed by hand, generated by DVS, or generated by some other
> > controller?  I ask because the "normal" action may not be an effective
> > way to enforce ACLs--it is an implementation of a MAC learning switch,
> > which is not itself an effective way to enforce ACLs--so I wonder what
> > assumptions lie behind this flow table construction.
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to