[c-nsp] BGP over Etherchannel Issue

[Julio Arruda]

Oliver Boehmer (oboehmer) wrote:
> C and C Dominte <> wrote on Monday, October 06, 2008 4:22 PM:
> 
>> I am having a rather unusual issue, and I have no idea on
>> how or where to find any info about it.
>>
>> Basically, a bgp session configured over an etherchannel
>> (2x10G ports) goes down, if one cable is removed. This should not
>> happen under normal circumstances, as the physical disconnection,
> should not
>> affect the Etherchannel operation, as long as there is at least one
> connected
>> port in operation. 
>> The IP addresses are configured over a VLAN, which is then allowed on
>> the etherchannel trunk. 
>> When I removed one of the two fibers, the bgp session went down, and
>> in the reset log the reason was peer closed the session, on both
>> sides.  
> 
> Are you running LACP or PaGP across the channel? If you're not, the
> remote end might not have noticed the member leaving the bundle and
> continued transmitting data over it, and if the BGP control packets
> happened to hash to this member link, the session will go down..

Not sure....he is talking fiber links, so, even without any 
'link-detection-tricks', the fiber port would go down if he unplugged it 
fully.
But yes, I agree some LACP/PaGP/UDLD/whatever of sorts is a good idea :-)

PS:
Unless "removed one of the two fibers" means removing 'one strand' of 
the fiber interface, and if you add on top of this no 
RFI/Autonegotation, you could have the port becoming a black-hole from 
the sender-side point of view..