[c-nsp] NX-OS 6.0(1) BGP graceful restart on SUP failover

[Phil Mayers]

Does anyone have this working?

I am testing an N7k with dual SUP, and when I perform a failover, BGP 
graceful restart is not working as expected. I have a TAC case open, but 
was wondering if anyone else had run into it.

The symptoms are:

  1. The initial BGP session has the "graceful restart" capability 
negotiated, and the expected AFI/SAFI marked for GR.

  2. When I perform "system switchover", the box continues to forward 
traffic, and the neighbour continues to send it traffic.

  3. Once the 2nd SUP becomes active, it establishes a new BGP peering. 
In the OPEN message, the "flags" of the GR capability has the R-bit set 
(indicating restart has occurred) but the F-bit for each AFI/SAFI is set 
to 0 - the neighbour thus assumes forwarding was not preserved, and 
discards the stale routes.

So, I have the odd situation that, after a switchover, forwarding works 
for about 20 seconds, then stops as the SUP effectively commits "BGP 
suicide" by telling the peer that forwarding didn't work!

This is on M-series linecards, in an MPLS/L3VPN config if that matters - 
though I am fairly sure it's a problem at the BGP level, since packet 
forwarding did continue, stopping immediately upon the new BGP session 
startup i.e. step #3 above.