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

[Kangping Dong <[e-mail]>]

Hi Ted,

I'd like to summary my understanding of what current SRP-draft-06 is
doing/allowing:

   1. We can add services separately. like, we can add "service-1" and in
   the first SRP update and "service-2" in the second SRP update.
   2. We can only remove all the services together with its host because
   "exactly one Host Description Instruction" is a MUST for each SRP update.

But the spec seems, is not enforcing this behavior:

> Therefore, SRP servers MAY remove all
> service instances pointing to a host when a host is removed, even if
> the SRP client doesn't remove them explicitly.
>
> If we cannot remove individual service without removing the host, should
the "MAY" be a MUST?
Otherwise, there may be no efficient way to clean up the service until the
lease time expires.

To remove a service registration, the SRP client retransmits its most
> recent update with an Update Lease option that has a LEASE value of
> zero.
>
> Does it remove only the services in my most recent update or all services
I have ever registered? This is what
I was confused about and thought we can remove individual service. As we
cannot remove individual service,
I think we don't need to retransmit the most recent update: It may be more
efficient to always transmit a
SRP update with only Host Description Instruction and zero lease time.
Thoughts?

For the current use model, I agree it will work for most cases and removing
individual service may be rare
(even there are, it can be done with a "removing all" and an add). So I
don't have a strong opinion to support it.

BRs,
Kangping

On Fri, Dec 18, 2020 at 10:07 AM Ted Lemon <[e-mail]> wrote:

> On Dec 17, 2020, at 7:40 PM, Abtin Keshavarzian <[e-mail]> wrote:
>
> - In "Service Description Instruction", we allow a "Delete all RRSet from
> a name", with possibly no follow-up adds.
>
>
> Yes.
>
> - When a service is being deleted, we can consider allowing KEY RR "Add to
> an RRSet" for instance name to be optionally included, to have the server
> keep the instance name reserved (if desired).  Does it make sense?
>
>
> We could allow this, but KEY RRs are large, so we might prefer not to.
>
> - In "Service Discovery Instruction" I think we need to allow delete as
> well (I guess "Delete an RR from an RR set" for PTR RR)?
>
>
> Yes, and the RR has to be pointing to a Service Description Instruction
> which is also being deleted.
>
> - We continue to require SRP Update msg to always include exactly one
> "Host Description".
>
>
> Yes.
>
> Does this all sound good/reasonable? Anything I may have missed?
>
>
> It sounds fine. I’m curious what others think about this.
>
>