On 2020-04-13, Claudio Jeker <cje...@diehard.n-r-g.com> wrote: > On Mon, Apr 13, 2020 at 02:08:31PM +0200, Remi Locherer wrote: >> On Mon, Apr 13, 2020 at 12:05:10PM +0100, Richard Chivers wrote: >> > On Mon, 13 Apr 2020, 10:18 Remi Locherer, <remi.loche...@relo.ch> wrote: >> > > >> > > On Mon, Apr 13, 2020 at 08:38:31AM +0100, Richard Chivers wrote: >> > > > We have been having a strange issue, whereby OSPF stops updating >> > > properly. >> > > > >> > > > We can see an entry for an ip route in the database but it is not in >> > > > the >> > > > kernel routing table, and when it is the DR, other routers then do not >> > > have >> > > > the route at all. >> > > > >> > > > We are seeing this across multiple boxes. We have 10+ ospf speakers, >> > > > and >> > > > seem to see the issue at different times. >> > > > >> > > > The problem starts with: >> > > > >> > > > ospfd[6960]: recv_db_description: neighbor ID x.x.x.x: seq num >> > > > mismatch, >> > > > bad flags >> > > >> > > The neighbor sent a db desc with the master flag set differently than >> > > what >> > > this ospfd instance recorded before for that particular neighbor. >> > > >> > > See 2nd last item on page 100 of RFC 2328: >> > > https://tools.ietf.org/html/rfc2328#page-100 >> > >> > >> > Thanks, should the routers just recover then from this scenario even if it >> > was happening due to lost packets, CPU pause etc. >> >> I think so. But it may take quite a while. It might also be an bug in ospfd >> or in another implementation.
On my 6.6/current boxes it seems to recover fairly quickly from this (30 seconds or so). I've definitely seen it take a long time in the past though. > Since this issues happen with 5.8 and 6.4 ospfd I would suggest to update > to at least 6.6 (especially the 5.8). IIRC there was some issue with ospfd > neighbor selection that caused troubles when sessions flapped. This was > fixed some time ago but I doubt 5.8 has that fix in. That one was fixed in 6.3. If you also run bgpd then be aware there are crashes with the version in 6.6 release - fixed in syspatches (and of course in snapshots), but one of the crashes is at startup with some configurations and it's hard to run syspatch if you have no routing ;) so either be ready to cope with that in case you run into it (e.g. pre-download the syspatch directory and make sure you have console access), or consider skipping 6.6 (go straight to a -current snapshot).
