Hi, following up on this:
On Tue, Sep 02, 2008 at 08:33:08PM +0300, Adrian Minta wrote:
> Gert Doering wrote:
> >On Tue, Sep 02, 2008 at 07:02:08PM +0300, Adrian Minta wrote:
> >>> - the BGP ghost bug is back :-(
I have now managed to open a TAC case on this - in case you want to
open your own case and attach to it, it's "SR 609533003". We have no
BugID yet, TAC is trying to reproduce it.
In our network, reproducing the problem is fairly straightforward (and
I have demonstrated it to the TAC engineer, who then mumbled something
like "I think you have found a bug here" - surprise, surprise :) ):
Host H, AS 65500 ------ Router A, AS 5539 ---- Router B, AS 5539
Host H announces a single prefix via eBGP to Router A.
Router A has a "bog standard" iBGP session to Router B. No(!) filters
of any kind between A and B.
Now "inject instabilities" -> tear down the BGP session H<->A, bring it
up again, wait a few minutes, tear it down again, and so on. After
somewhat between 2 and 10 "session down", the following happens:
- on Router A, the prefix completely drops from the BGP table
Cisco-A#sh ip b 193.31.7.1/32
% Network not in table
- on Router B, the prefix is still visible, via "A":
Cisco-B#sh ip b 193.31.7.1/32
BGP routing table entry for 193.31.7.1/32, version 16726560
Paths: (1 available, best #1, table Default-IP-Routing-Table)
Flag: 0x820
Advertised to update-groups:
4
65500
195.30.H.H (metric 3584) from 193.149.A.A (194.97.A.A)
Origin IGP, metric 0, localpref 100, valid, internal, best
it doesn't matter whether "next-hop-self" is used or whether "B" is
a normal iBGP peer or a route-reflector-client (or whether a mixture of
"normal peers and RR clients" exists). "A" just forgets - occasionally -
to send withdraws if it doesn't have the prefix any longer.
Of course A and B have full BGP tables - and as there is instability and
withdraws out there, we see this happen to about 5-20 prefixes per day.
It might have to do with the amount of "normal" updates going on in
parallel - "if there is lots of updates, then there is a propability
that things will get lost". But maybe not. (I have not yet tested in
a "pure lab" environment, with no other BGP updates between A and B).
gert
--
USENET is *not* the non-clickable part of WWW!
//www.muc.de/~gert/
Gert Doering - Munich, Germany [EMAIL PROTECTED]
fax: +49-89-35655025 [EMAIL PROTECTED]
pgp223xLafzZU.pgp
Description: PGP signature
_______________________________________________ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
