[nsp] Graceful shutdown

[Pete Templin]

> On Tue, 6 Jul 2004, sam_ml at spacething.org wrote:
> 
>>I've been thinking about some expect scripts that would shut a router down
>>gracefully (ie. without packet loss), by giving backup routers time to
>>takeover traffic forwarding before the power cycle.
>>
>>The existing configuraion would be saved, and then an expect script would
>>modify it to effectively make _it_ the backup router, then reboot it some
>>time later - say 120 seconds - without saving the modifications.
>>
>>For HRSP/VRSP you could lower the priority of each of the groups to below
>>that of a pre-empting backup router. For BGP sessions configuring
>>"prefix-list none in"/"prefix list none out"  for every peer should give
>>BGP time to converge before the router reboots.
>>
>>It might also be possible to do this with some Layer2 ethernet devices by
>>raising the spanning-tree cost of every port in an attempt to deflect
>>traffic away from it.
>>
>>Does this seem reasonable? Does something like this already exist, or 
>>should I get out my expect hat?

 From a BGP perspective, I keep a pair of route maps handy that severely 
depreference the peer (and "myself" to that peer).  YMMV, but some 
combination of prepends, communities, local preferences, weights, 
whatever will guide the network around the path/node.

Depending on your implementation, this may be handy after the power 
cycle as well.  In my case, I was able to "cut my teeth" on the 
Riverstone platform (it wasn't my choice...), where multiple route maps 
can be applied in each direction.  My maintenance route map was last on 
the list, so a quick "comment out" of the earlier sequence route maps 
and the path is routed away but still available in case of crisis.

pt