On Wed, Mar 6, 2019 at 8:32 Smith, Courtney <Courtney_Smith@comcast.com> wrote:

On 3/5/19, 6:04 PM, "NANOG on behalf of Job Snijders" <nanog-bounces+courtney_smith=comcast.com@nanog.org on behalf of job@instituut.net> wrote:

    On Sun, Mar 03, 2019 at 08:42:02PM -0500, Joshua Miller wrote:
    > A while back I read somewhere that transit providers shouldn't delete
    > communities unless the communities have a specific impact to their
    > network, but my google-fu is failing me and I can't find any sources.
    >
    > Is this still the case? Does anyone have a source for the practice of
    > leaving unknown communities alone or deleting them?

    https://tools.ietf.org/html/rfc7454#section-11


Remember policies between two peers may not be same as customer policies.

Example:  Customers_of_transit_X >>> Transit X >>> Peer_A >> Customers_of_Peer_A

Customers_of_Peer_A may use community A:50 to set local pref to 50 in Peer_A network.  But that doesn’t not mean Customers_of_transit_X can send A:50 to set lpref on their routes in Peer_A's network.  Peer_A's policy with Transit X likely does not take action on customer communities since they are 'peers' not customers.  Transit X can send A:50 to Peer_A but nothing would happen.  What's the benefit of Transit X preserving A:50 from its customers if it means nothing in Transit X?

OP didn’t specify what kind of BGP communities they were referring to. In general we can separate communities into two categories: “Informational” and “Action”. You are right that preserving/propagating “action” communities (such as in your example) probably isn’t that interesting. “informational” communities on the other hand can be very valuable. 

See https://tools.ietf.org/html/rfc8195 for more information on how the two types differ.

Kind regards,

Job