[c-nsp] Odd problem with N7K and multicast

[Phil Mayers]

On 31/07/14 21:42, John Neiberger wrote:

> Have any of you seen this before?

I've seen something similar before in an MVPN context; *,g would get 
created but s,g would not.

This was on an earlier 5.2 and TAC were able to reproduce on that same 
version, but not on later 5.2 so we upgraded and it went away.

When I was fiddling with it I found that "restart pim" and/or "restart 
bgp" would sometimes tickle it to work. This made me suspect there was 
some internal datastructure that was getting corrupted, presumably by a 
bug, causing the s,g triggering process to fail. Maybe this is worth trying?

(FWIW "restart" on NX-OS seems to work really well with very minimal 
disruption and fast recovery, for reasonable table sizes)

TBH I've run into a fair few niggles like this, with unicast and 
multicast, on N7k. It's a pretty decent platform but there seem to be 
some RIB/FIB/state corruption issues lurking in bits of the code. I 
wonder if this stems from porting IOS routing protocol code to the very 
different NX-OS, or just relative youth of the platform.

Other options that spring to mind - for s,g to trigger the packets must 
leak to the SUP, so check CoPP and hw ratelimiters are all configured 
correctly and working; an IGMPv3 s,g join obviously sidesteps this.

I would definitely TAC it if you can reproduce, however.