Replies: 1 comment
-
I have been doing some testing with this and the problem seems to be related to the FRR specific neighbor cache. For my particular test case the following ip6table rules provide a work around.
For the actual FRR problem the output of "show int XXX" has a list of ipv6 neighbors, however this list seems to be overwritten every time FRR receives a router-advertisement on that interface. So with hundreds of ipv6 speakers on an interface, the ipv6 neighbor table is constantly being overwritten.
Notice the number of advertisements received. I suspect this is a bug here: https://github.com/FRRouting/frr/blob/master/zebra/rtadv.c in the function: rtadv_recv_packet but I'm not a C Coder, so the issue may be elsewhere. |
Beta Was this translation helpful? Give feedback.
-
FRR is running on a VM with 400 other cloned VMs and attempting to peer with two identically configured Arista switches. Given enough time, 3-4days, all VMs will eventually peer with both top of rack switches as expected. However any individual VM takes a longtime to establish.
Debugging on a random VM, I see this message in the console every 180secs until eventually the peering is established:
However Dynamic neighbors are configured.
On the Top of Rack Switch
And I only reset one neighbor:
So both neighbors are configured identically.
Once the neighbor is finally established debug see this:
I have attached a packet capture that shows the Arista switch establishing a TCP session and then sending a BGP Open message, but getting an explicit reset from FRR. Until eventually FRR decides to accept and turn up the session at Packet 6742 in the capture (around 300Min after I restarted the session).
bgp_ens224.pcap.zip
Beta Was this translation helpful? Give feedback.
All reactions