search

p4lang / p4-applications

P4 Applications WG repo
107 stars 45 forks source link

INT extensions for domain specific features #54

Closed rsivakolundu closed 4 years ago

jklr commented 4 years ago

@rsivakolundu I wonder if we can prioritize this PR and merge first than others. The change of this PR is applied to every INT mode (old and new) and the examples. If the key section is updated and merged into master, I can update the other examples (under different transports) and add new examples accordingly.

jklr commented 4 years ago

Thanks @rsivakolundu for the fixes. They look good to me. Can you pls comment on @mickeyspiegel 's questions below? I put my answers.

I see two major differences from the IETF's IOAM namespace beyond the additional fields:

  1. When a transit node receives an INT packet with a Domain Specific ID that it does not recognize, is it still allowed to add its metadata to the packet?

[JK] I didn't think through but the answer depends on the Yang-model-based query mechanism. Yes, if the collector can ping each device and get the metadata semantics. No, if not. Agree we better clarify this in the spec.

  1. Does the Domain Specific ID have any bearing on the semantics and/or metadata units? In the IETF this seems to be the main point of namespace. It would be good to hash this out further, either aligning with the IETF or clarifying at least to ourselves if not in the spec, why should this be different?

[JK] associating the domain ID to the metadata semantics seems a good idea to me.

jklr commented 4 years ago

Let's merge it and address the domain ID questions in a separate PR on top of master.