On 2017/09/23 09:27, Claudio Jeker wrote: > On Thu, Sep 21, 2017 at 03:45:28PM +0100, Stuart Henderson wrote: > > Two routers on a point-to-point ethernet, the link between them dropped > > out but ports stayed up. The link came back after ~2h or so but ospfd > > doesn't recover - > > > > first router: > > > > 195.95.187.3 1 FULL/OTHER 00:00:03 195.95.187.16 ix1 1d14h45m > > > > Neighbor 195.95.187.3, interface address 195.95.187.16 > > Area 0.0.0.0, interface ix1 > > Neighbor priority is 1, State is FULL, 92 state changes > > DR is 195.95.187.16, BDR is 0.0.0.0 > > Options -|O|-|-|-|-|E|- > > Dead timer due in 00:00:03 > > Uptime 1d14h49m > > Database Summary List 0 > > Link State Request List 0 > > Link State Retransmission List 0 > > > > > > second router: > > > > 195.95.187.1 1 FULL/OTHER 00:00:03 195.95.187.17 vlan730 1d14h45m > > > > Neighbor 195.95.187.1, interface address 195.95.187.17 > > Area 0.0.0.0, interface vlan730 > > Neighbor priority is 1, State is FULL, 12 state changes > > DR is 195.95.187.17, BDR is 0.0.0.0 > > Options -|O|-|-|-|-|E|- > > Dead timer due in 00:00:03 > > Uptime 1d14h46m > > Database Summary List 0 > > Link State Request List 0 > > Link State Retransmission List 1 > > > > > > i.e. they both think that they're DR and that there's no BDR. > > (1d14h ago, yeah I was a bit slow spotting it this time..) > > > > Not the first time I've seen it (on various links over various releases).. > > In this case they're on a /31 though I've seen it on /30's previously. > > > > Are other people seeing this? > > > > Does anyone have an idea where the bug might be, or a better suggestion > > than removing the interfaces from ospfd.conf, reloading, re-adding, > > reloading again? > > > > Do you have tcpdumps of the OSPF traffic by any chance?
I don't, but I'll try to scrape together a test network to reproduce and collect them. > I think there is a case where the hello protocol gets confused and so you > end up with multiple DRs that stick to themself. > You could try to down/up the interface which may be enough to get out of > this state... I have some recollection of hitting other problems when I've done that in the past, remove/reload/add/reload is a bit annoying but seems fairly low impact.
