Hi, I have narrowed down the problem so far as to that I know what peer is causing the issue on both my routers. I have all information I have regarding my setup and tests etc shared with Martin Winter and he has created a test bed but in this test bed we have not been able so far to reproduce the issue.
What we see is that the moment the peering session with the partner comes up (EBGP) the router with this session stops sending all updates or part of its updates to it's iBGP peers. This results over time in an incomplete set of routes on the other routers in the iBGP mesh. A last test that I have to do is to see if there is a difference in the peer being in a peergroup or having it's own config outside of the peergroup. I have done this test but part of the logs I created (pcap files) weren't downloaded because I was under the impression that the problem had disappeared. But a few days later I got alerts from my monitoring servers that the number of routes in the iBGP mesh was getting below the alert thresh-hold which proved to me that the problem was actually still there but a little bit less visible. I think Martin has created a complete overview of my configuration and documented most of it in his debugging work set. My production routers are running the latest 0.99 version because I can't upgrade to 1.1 because of a serious OSPF6 bug that completely destroys my internal IPv6 routing. Jan Hugo On 01/23/2017 11:33 PM, Paul Jakma wrote: > On Mon, 23 Jan 2017, Mike Tancsa wrote: > >> There are a few people who have run into the issue. The peer with >> the problem seems to exhibit the behaviour when there are "many" ebgp >> peers. Routes learned from the ebgp peers do not all get propagated >> to ibgp peers. It seems random prefixes do not get sent to the ibgp >> peers. tcpdump confirms they are never sent. Jan Hugo I think has >> narrowed it down much more than I have. > > Ok, any details you have would be great. > > I did BGP stress testing late last summer, of Quagga and BIRD. Those > tests depended on all prefixes getting through to a 'monitor' bgpd to > detect the end of the test run. Though, not complex topologies. > > Need to work to build a comprehensive, free software, routing > test-rig, to catch those issues (and incorporate BIRD, OpenBGPd, etc., > to enable comparison testing and find strength/weaknesses - and > provide them with the same kind of testing as a benefit). > > If there's people interested in that... > > regards, _______________________________________________ Quagga-dev mailing list [email protected] https://lists.quagga.net/mailman/listinfo/quagga-dev
