Duplicated PEER_INDEX_TABLE in rit mrt file

[Wydrych, Piotr]

Hello Ondrej,

On 14.01.2021, 04:03, "Ondrej Zajicek" <santiago at crfreenet.org> wrote:
> Thanks for the bugreport. For IPv6 tables (due to ugliness of how MP-BGP
> handles bgp_nexthop), we forgot to generate fake MP_REACH_NLRI attribute
> with bgp_nexthop.
> 
> The resulting files worked with bgpdump (which i used for tests), just
> displayed some default bgp_nexthop.
> 
> Fixed here (now it works with bgpdump, mrt2bgpdump and bgpscanner):
>
> https://gitlab.nic.cz/labs/bird/-/commit/d774f6d721b0e52ed800c4b9a3a482c8ce9dd074

I did some checks in my environment and it works only for small dumps. I checked both vanilla 2.0.7 + cherrypicked fix as well as vanilla master. After some* number of entries dumped, it stops encoding MP_REACH_NLRI until end of the dump. Next dump appended to the same file is exactly the same. It seems reproducible for me - for given set of routes, it stops encoding exactly on the same route. I'll send you the dumps off-list to not pollute archives.

I did check with mrtparse/examples/print_all.py for two dumps and it stopped encoding somewhere below sequence number 2^14 (I saw last NLRI @ 0x3FF2 and 0x3FF0), FWIW.

*some, e.g.:
When I have 15687 ipv4 routes + 1465 ipv6 routes, 730 of them miss MP_REACH_NLRI
When I have 15533 ipv4 routes + 1448 ipv6 routes, 557 of them miss MP_REACH_NLRI
When I have 18010 ipv4 routes + 1801 ipv6 routes, all of them miss MP_REACH_NLRI

Piotr