Race condition when fail-over chlid cluster

[rdemongeot]

Hello,

As discussed per Skype, we have found a possible race condition when fail-over a child cluster.

Parent-stream1 --------+
                                   +-->  child-s1 (old primary) --------> child-s2 (new primary)
Parent-stream2 --------+

to migrate to

Parent-stream1 --------+
                                   +-->  child-s2 (new primary) --------> child-s1 (old primary)
Parent-stream2 --------+

During the operation, the child S1 was reconfigured with "MASTER_USE_GTID=CURRENT_POS" (expected)
but REJOIN Module reconfigure it as "MASTER_USE_GTID=SLAVE_POS" making search for a very old GTID not found into the new primary binary log.

[svaroqui]

Please reopen if not fixed