Re: [v6ops] An Update to Happy Eyeballs

[神明達哉 <[e-mail]>]

At Thu, 16 Mar 2017 13:09:57 -0700,
Stuart Cheshire <[e-mail]> wrote:

> >   If, during the course of connection establishment, the DNS answers
> >   change either by adding resolved addresses, or removing previously
> >   resolved addresses (for example, due to expiry of the TTL on that DNS
> >   record), the client should react based on its current progress.
> >
> > Specifically what kind of scenario does this intend to imply?  For
> > example, consider the following scenario:
> >
> > 1. start a AAAA query for a destination name
> > 2. start an A query for the same name almost at the same time
> > 3. get a response to the AAAA query (set of IPv6 addresses)
> > 4. start establishing a connection for one of the IPv6 addresses
> > 5. get a response to the A query (set of IPv4 addresses)
> >
> > At step 5, DNS resolution is fully completed.
>
> No, DNS resolution is not fully completed at step 5. In this
> asynchronous model (as used in Apple products for many years) DNS
> resolution is an ongoing process, which continues until the client
> has successfully made a connection, and only stops when the client
> tells the DNS engine that it can stop.

I guess there's something (undocumented yet and) very specific to the
Apple implementation details of the "asynchronous model", then.  In
the above example steps, we've received an answer to both AAAA and A
queries at step 5.  Whether the resolution is asynchronous or
synchronous, why can't we "tell the DNS engine to stop"?  Does the
"DNS engine" still expect more responses even after getting both AAAA
and A responses?  Perhaps it keeps sending the same queries to see
whether there's a change?  At least it's not clear to me from Section
3 of the draft.

> Many web sites use DNS records with fairly short TTLs, which allows
> them to update answers fairly rapidly.

I know all this.  But I don't think it relevant to the point of my
question.

--
JINMEI, Tatuya