Re: [dnssd] SRP Update - removing individual services (draft-ietf-dnssd-srp-06)

[Ted Lemon <[e-mail]>]

On Dec 19, 2020, at 10:34 AM, Kangping Dong <[e-mail]> wrote:
> 
> Sorry, another question :)
> 
> In section 2.3.1. Validation of Adds <https://www.ietf.org/archive/id/draft-ietf-dnssd-srp-07.html#name-validation-of-adds>, we have:
> 
>  SRP Updates consist of a set of instructions that together add or remove one or more services. 
> Each instruction consists either of a single add, a single delete, or a delete optionally followed by an add. 
> When an instruction contains a delete and an add, the delete MUST precede the add.
> 
> What is the use case of the "single add"? To my understanding,  the "delete + add" pattern is
> for resetting the service to our current SRP update. So the "single add" pattern is for updating
> the existing service, for example, adding one more AAAA RR, right?

The PTR add can just be an add, because there’s a 1:1 match between a service name PTR record and the service instance name that is its target. But this text is indeed confusing, so I’ve updated it:

            Each instruction
	    consists some combination of delete updates and add updates. When an instruction contains a delete and an add, the
	    delete MUST precede the add.

> I think, for most SRP clients, it will be more convenient for them to track all IP Addresses and
> always update with the whole IP addresses list rather than depending on this "update" model
> and register the addresses separately. The server, I think can also be simplified. Thoughts?

Indeed, the single add doesn’t make sense—the client has no way of knowing if there may be stale data present on the name.

> Another issue with this sentence is that, the Service & Host Description Invalidation requires:
> exactly one "Delete all RRsets from a name" RR,
> Seems disallowing the "single add" pattern? 

That’s right. The model of having the client partially update its data wouldn’t be consistent with the fact that SRP updates are essentially blind updates. So yes, the client always has to send all the data for any instruction, and that means we need a delete for every Host instruction, and a delete for every Service Description instruction. This isn’t needed for the Service Discovery instruction because there can only be one PTR record on a particular service type pointing to a particular service instance name—if it’s there, there’s no need to delete it before adding it.