Review Request: Send NoMasterDetectedMessage to non-contending detectors. Added a disconnected slave map to the master to track disconnected slaves, in order to disallow slave re-registration after a network partition.

[Ben Mahler]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/10172/
-----------------------------------------------------------

Review request for mesos, Benjamin Hindman and Vinod Kone.


Description
-------

See above. This is a fix of MESOS-305.

This also fixes MESOS-362.


This addresses bugs MESOS-305 and MESOS-362.
    https://issues.apache.org/jira/browse/MESOS-305
    https://issues.apache.org/jira/browse/MESOS-362


Diffs
-----

  src/detector/detector.cpp 7a8355162d543e017505dd58efd2d7bf96f99623 
  src/master/http.cpp 71b04f01f45ee73d9c246f469e1368223903abed 
  src/master/master.hpp 9776a7cb8448e41e5d52288e3c637737cee15a08 
  src/master/master.cpp 5b0e8c03c516f9fc8bb729c21e876bdde89baf9c 
  src/tests/fault_tolerance_tests.cpp 9d3f8b1bfb58d459b1719d2ba1dbb2e93858fc92 
  src/tests/master_detector_tests.cpp fe3b91fb375e0b09f8f2de3e69e736cd5f5b94ba 

Diff: https://reviews.apache.org/r/10172/diff/


Testing
-------

make check

Added tests for the partitioned slave re-registration.
./bin/mesos-tests.sh 
--gtest_filter="FaultToleranceTest.PartitionedSlaveReregistration" --verbose 
--gtest_break_on_failure --gtest_repeat=3000

Ran into MESOS-406, but otherwise no issues.

Would love to add zookeeper master detector tests to ensure these new 
NoMasterDetected messages make it through, but there seems to be no easy way to 
induce the timedout() call at the moment. Suggestions?


Thanks,

Ben Mahler