[c-nsp] IOS BGP AS-PATH Reftable

[David Freedman]

Does anybody know how exactly this is kept clean?

I would assume that when a prefix with a ref to the as-path reftable is
no longer in a rib , the refcount of the entry is decremented and at
zero the ref is cleaned up.

I seem to have stale data in mine, this example concerns a prefix leak
from AS30071 (a peer network) which is no longer happening (i.e, I do
not appear to see nor select these paths)

#sh ip bgp paths _30071_7018
Address    Hash Refcount Metric Path
0x52866F14  954        7     40 30071 7018 26415 i
0x556E1A78 1234        4     40 30071 7018 71 i
0x547CED9C 1234       37     40 30071 7018 71 i
0x54F383F8 2708       10     40 30071 7018 i
0x5E20558C 2885        4     40 30071 7018 40912 i

I see this across a GSR peering router running 12.0(SY),
a 7600 PE router running 12.2(33)SRC1 and a 65K PE router
running 12.2(33)SXH2a.

is this intended behaviour (not to decrement refcounts)
or a bug?

and if so, what constraints exist to prevent this reftable consuming
huge amounts of RAM (i.e , if somebody were to spam it with multiple
bogus AS-paths) ?

Thoughts?

Dave.