[DNSOP] Re: Comments on "Service Binding Mapping for Service Levels"

[Erik Nygren <[e-mail]>]

As another top-level comment on this draft:

For at least some operators, deploying safely will also require having a
way for clients to signal which SVCB record was used.  For example, load
reporting, debugging/diagnostics, and other operational use-cases would
need to know whether the background or interactive SVCB endpoint was used.

It would make sense to do this more generally rather than as part of this
specific draft.
Here was the backlog ticket from when SVCB was being authored where we
captured future-draft notes:

    https://github.com/MikeBishop/dns-alt-svc/issues/107
The privacy considerations were one of the biggest challenges here.  It's
also unclear which WG such a draft would go into?  (eg, http or maybe even
a future incarnation of happy.)  The top ideas had either been:  1) a
SVCB-Used request header that includes the used SVCB RR (the best solution
if we can be ok with the privacy properties); 2) introducing an identifier
tag SvcParam of at most N bits which then gets sent in a SVCB-Used header
(which would be no worse than what can be trivially done with IPv6
addresses as long as N<=64).

As to this draft, I wonder if having a bitmask of attributes (eg, with one
for "background") or list of tokens could be another approach we could take
that could make this more generally useful for cases other than just
background.

        Erik