Hi! * Ilya Maximets [Thu Feb 09, 2023 at 09:44:41PM +0100]: > On 1/31/23 16:37, Michael Prokop via discuss wrote:
> > On a Proxmox VE cluster using openvswitch v2.15.0 (as present in > > Debian/bullseye[1]) we ran into a serious problem: after the > > openvswitch-switch/ovs-vswitchd service restart the OVS bridge lost > > its IP address and the interface stayed down. > > > > As it turned out this issue got already fixed by Ilya Maximets with > > https://github.com/openvswitch/ovs/commit/bc0aa785a83c11dab482b3e20736b969174d9f86 > > ("ovsdb-idl: Fix the database update signaling if it has never been > > connected.") > > > > What was special about this issue for us is that we noticed > > the problem *only* on multi-CPU systems (systems with >=2 CPUs). > > Furthermore the issue didn't show up on any of our systems running > > with Intel CPUs, but only on systems running with AMD CPUs. :) > > > > I'm curious, is this related to a concurrency/race condition? > > And I'd like to understand whether this Intel vs. AMD behavior is > > only coincidence related to different kernel/userland/libdpdk/... > > environments, or if that could really show up as vendor specific > > behavior as we saw it? :) > Hi. I don't think it's vendor-specific. There is nothing really > vendor-specific here. There might be a slight race window, IIRC, > i.e. how many main loop iterations ovs-vswitchd does before getting > database updates, so the number of CPU cores (1 vs 2) may play > some role. But even that is unlikely. In general, you should be > able to reproduce the issue on any system. I see, we'll keep an eye on it, thanks for your feedback and estimation, Ilya - much appreciated! regards -mika-
signature.asc
Description: PGP signature
_______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
