BIND 10 #2028: socket session forward test fails on solaris11

[BIND 10 Development]

#2028: socket session forward test fails on solaris11
-------------------------------------+-------------------------------------
                   Reporter:         |                 Owner:  jinmei
  jinmei                             |                Status:  accepted
                       Type:         |             Milestone:
  defect                             |  Sprint-20120619
                   Priority:  high   |            Resolution:
                  Component:         |             Sensitive:  0
  Inter-module communication         |           Sub-Project:  Core
                   Keywords:         |  Estimated Difficulty:  7
            Defect Severity:  N/A    |           Total Hours:  0
Feature Depending on Ticket:         |
        Add Hours to Ticket:  0      |
                  Internal?:  0      |
-------------------------------------+-------------------------------------

Comment (by jinmei):

 trac2028 is ready for review.

 The attached file (additions to socketsession_unittest.cc) highlights
 the issue more clearly.  To see the effect you'd also need to disable
 SO_REUSEADDR:
 {{{#!cpp
 #if 0
         const int on = 1;
         if (setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)) ==
 -1) {
             isc_throw(isc::Unexpected, "setsockopt(SO_REUSEADDR) failed: "
 <<
                       strerror(errno));
         }
 #endif
 }}}

 I've not figured out how exactly that happened, but what happened was
 that when we create a socket and forward the socket FD via a UNIX
 domain socket (and then close both the original and forwarded FDs),
 this deviant system somehow prevents the reuse of the port for some
 short period (some shorter than 1sec).

 So, my proposed fix is to just use different ports for the offending
 tests, which is what this branch does.

 Note that TCP tests aren't affected, probably because it forwards a
 TCP socket on accept() whose remote address is always different.

 Note also that this behavior of Solaris won't affect more practical
 cases of BIND 10, because the forwarder side of app should keep the
 UDP socket open, in which case this type of problem doesn't happen (I
 confirmed that).  So this is basically only for the specific set of
 unit tests, and any workaround that avoids the trouble should be okay.

 Since it's only about unit tests and for a newest buildbot
 environment, I don't think we need a change log for it.

-- 
Ticket URL: <http://bind10.isc.org/ticket/2028#comment:4>
BIND 10 Development <http://bind10.isc.org>
BIND 10 Development