Alex
In the simplest situation only routing has restarted, all physical
connections are up. Routing has to rebuild the world with minimal
disruption to the network. So during this state the physical links still
have LSPs and the links still have metrics and they should stay. The
one thing you don't want to do is start a path selection from this
node but all other options are possible. So I would say just let
the TE database build until it is synchronized and block local path
selections (which is really easy to do).
By the way I think you can still build LSPs through the node and
feedback and crank back can be used to bridge the "race conditions"
that have been opened wider by graceful restart. Eventually the node
is synchronized with the neighbors and LSAs can be generated.
If you have more elaborate failures such as software for link management
at the same time routing is synchronizing then these local links can
have bandwidth and metric changes but this is typical of a link
failure today and is localized only to the links in question.
All of this is driven by causing the least amount of disruption
possible which is the goal of graceful restart.
Don
> -----Original Message-----
> From: Alex Zinin [mailto:zinin@psg.com]
> Sent: Friday, April 26, 2002 4:00 PM
> To: Fedyk, Don [BL60:1A00:EXCH]
> Cc: Yakov Rekhter; ccamp@ops.ietf.org
> Subject: Re: TE metric and graceful restart
>
>
> Don,
>
> Would you care to elaborate more, please?
> Specifically, what do you think is wrong here
> and what would be the right way in your opinion.
>
> Thanks.
>
> --
> Alex
>
> Friday, April 26, 2002, 12:12:09 PM, you wrote:
> > Yakov
>
> > I have a question. Do you mean that a node that is restarting
> > its routing view and learning from it's neighbors should mark
> > the TE metrics to local neighbors to infinity? I think that during
> > any graceful restart the TE metrics will not have a perfect view
> > and local path selections should be blocked but setting the metrics
> > artificially to infinity seems like the wrong way to do this in my
> > opinion.
>
> > Don
>
>