BIND 10 #2003: support DDNS/TCP response

[BIND 10 Development]

#2003: support DDNS/TCP response
-------------------------------------+-------------------------------------
                   Reporter:         |                 Owner:  jinmei
  jinmei                             |                Status:  reviewing
                       Type:  task   |             Milestone:
                   Priority:  low    |  Sprint-20120612
                  Component:  DDNS   |            Resolution:
                   Keywords:         |             Sensitive:  0
            Defect Severity:  N/A    |           Sub-Project:  DNS
Feature Depending on Ticket:  DDNS   |  Estimated Difficulty:  5
        Add Hours to Ticket:  0      |           Total Hours:  2.12
                  Internal?:  0      |
-------------------------------------+-------------------------------------

Comment (by jinmei):

 Replying to [comment:10 vorner]:

 > > I agree that's debatable.  Which options other than this are you
 > > thinking about?
 > >
 > > - returning SERVFAIL or REFUSED?  but it requires to keep the
 > >   connection at least until the transmission of the response is
 > >   completed.
 >
 > It would be possible to at least try sending REFUSED in a non-blocking
 way and drop the connection if it doesn't work. I don't think it would
 happen often that the send would block.

 Yes, that could be an option, and I agree this send normally wouldn't
 block as DDNS responses should be pretty small in general.

 > > - killing oldest client to allow a new client?  Maybe, but maybe not;
 > >   if the server is simply busy, it could be possible that the server
 > >   is just killing any TCP client before completing any of the
 > >   requests.
 >
 > We may want to do something like that (or timeout the connections)
 anyway. Otherwise, someone might create few connections and keep them open
 for ever, blocking the server.

 This ("otherwise") is correct, but that's what other implementations
 do.  I'm not necessarily saying that it means that's the only feasible
 behavior, but the fact that multiple different implementations behave
 that way probably indicates it's not a big deal in practice anyway.

 > > > And, generally speaking, isn't 10 as the quota really low?
 > >
 > > This is also debatable, and I'm not sure.  BIND 9's default of
 > > tcp-clients (not specific to ddns) is 100, so maybe we should use the
 > > same value.  In any case, as long as we use select, we cannot increase
 > > it too much due to its scalability.
 >
 > Ah, right, select :-|. Well, hopefully, we won't ever reach the limit.
 >
 > After looking at the test, I think it can be merged.

 Okay, thanks (and thanks for fixing the typo), merge done.  I'm
 closing this ticket.

-- 
Ticket URL: <http://bind10.isc.org/ticket/2003#comment:12>
BIND 10 Development <http://bind10.isc.org>
BIND 10 Development