[ovs-dev] Revert "ovs-lib: Try to call exit before killing."

Ben Pfaff blp at ovn.org
Mon Feb 1 18:09:40 UTC 2016


On Mon, Feb 01, 2016 at 08:05:33AM -0800, Guru Shetty wrote:
> On 31 January 2016 at 23:24, Ilya Maximets <i.maximets at samsung.com> wrote:
> 
> > On 29.01.2016 22:02, Guru Shetty wrote:
> > > This reverts commit 9b5422a98f817b9f2a1f8224cab7e1a8d0bbba1f.
> > >
> > > Reason:
> > > Calling 'ovs-appctl exit' on ovs-vswitchd will cause ovs-vswitchd
> > > to destroy all the ports from the datapath. This is an unacceptable
> > > behavior while restarting the daemons.
> >
> > Killing is more unacceptable here, because ovs with vhost-user port will
> > not be able to open this port anymore.
> >
> If 'ovs-appctl exit' solves the above problem, can one could argue that
> killing the daemon via SIGTERM should do the same? i.e you should register
> the cleanup functions that is so important for proper functioning after
> restart (irrespective of how this particular problem is solved).
> 
> 
> >
> > > Couple of reasons that come
> > > to mind are:
> > >
> > > 1. Any configured ip addresses will be lost
> >
> > Why this can't be solved using network-scripts or something else?
> >
> This is very hard since OVS runs on multiple hypervisors each with
> different way of configuring interfaces.
> 
> 
> >
> > > 2. Traffic will stop flowing.
> >
> > I think, this was just a side effect. IMHO, stop of traffic flowing
> > is an expected behaviour while you restarting the switch.
> >
> When OVS daemons are restarted, the flows in the kernel remain making old
> traffic continue to flow. There has been quite a bit of effort done to
> prevent traffic disruption during upgrades so you cannot introduce a
> regression here.
> 
> 
> I think it is probably easier to prevent 'ovs-appctl exit' from deleting
> the bridges in the first place.

I'm going to try for that first.



More information about the dev mailing list