On Wed, Apr 08, 2026 at 08:39:36AM -0400, Jeffrey Haas wrote:
> On 4/8/26 08:32, Job Snijders wrote:
> > On Tue, Apr 07, 2026 at 08:58:14AM +0000, Wolfgang Tremmel wrote:
> > which is current practice for example for our (DE-CIX) route servers.
> > Right, this is _exactly_ NOT a good practise, and its been extensively
> > documented why it isn't a good practise. Based on what you say, I have
> > to conclude that the DE-CIX route servers are configured in a way that
> > is inefficient and a burden for its peers. Why is it set up this way? :-)
> >
> > Would it help your team if the document used stronger words?
>
> Keeping local marking of routes out of the global BGP tables has obvious
> benefit.
>
> Doing so internally is an operational choice. One that, as the sidrops
> document highlights, generates churn. While the sidrops case is specific,
> the problem is rather more general since many operators do all sorts of
> dynamic things that involve internal marking of routes that can result in
> internal churn.
Sure, but I'm not sure the context here is _internal_ marking.
> Objecting to it harder will not make it go away. At best, we can
> provide for mechanisms to make sure it doesn't show up in the global
> tables and provide for practices and mechanisms that reduce the churn
> internally.
Right. But softening language around unhelpful practices is something
that raises eyebrows and must be carefully considered.
<A> doing XYZ has tangible downsides
<B> but I am doing XYZ! change the document!
<A> errr... (in silent: or perhaps change the practice?)
If this conversation helps kickstart a re-evaluation of the
configuration and practises of making certain internal dynamic state
markings externally visible, at a very large route server, that would be
a win.
Kind regards,
Job
_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]
