The whole balancer plans might be aborted if there are more than one plans to move a same region

In SimpleLoadBalancer, plans are generated firstly by average number regions per server for a table. Each server will be randomly assigned either floor(average) or ceiling(average) regions (if the average is not an integer number). But afterwards, the balanceOverall method might generate new plans of some regions of the table to balance server loads in whole cluster scope. As a result, there are plans to move a same region in one call of balance.

Currently, branch-2 is using async procedures to implement balancer plans. But the concurrency of moving the same regions will cause the balance method failed. And all the afterwards plans will not be implement when one plan encounters exception.
We have encountered this problem in our practices, the logs are as follows,

2018-09-26,12:12:38,224 INFO [master/c4-hadoop-tst-ct15:52900.Chore.1] org.apache.hadoop.hbase.master.HMaster: Balancer plans size is 3757, the balance interval is 79 ms, and the max number regions in transition is 25
2018-09-26,12:12:38,224 INFO [master/c4-hadoop-tst-ct15:52900.Chore.1] org.apache.hadoop.hbase.master.HMaster: balance hri=1588230740, source=c4-hadoop-tst-st99.bj,52900,1537522783781, destination=c4-hadoop-tst-st28.bj,52900,1537520009497
2018-09-26,12:12:38,325 INFO [master/c4-hadoop-tst-ct15:52900.Chore.1] org.apache.hadoop.hbase.master.HMaster: balance hri=1588230740, source=c4-hadoop-tst-st99.bj,52900,1537522783781, destination=c4-hadoop-tst-st29.bj,52900,1537522784188
2018-09-26,12:12:38,325 INFO [PEWorker-16] org.apache.hadoop.hbase.master.procedure.MasterProcedureScheduler: pid=119197, state=RUNNABLE:REGION_STATE_TRANSITION_CLOSE; TransitRegionStateProcedure table=hbase:meta, region=1588230740, REOPEN/MOVE checking lock on 1588230740
2018-09-26,12:12:38,325 ERROR [master/c4-hadoop-tst-ct15:52900.Chore.1] org.apache.hadoop.hbase.master.balancer.BalancerChore: Failed to balance.
org.apache.hadoop.hbase.HBaseIOException: rit=OPEN, location=c4-hadoop-tst-st99.bj,52900,1537522783781, table=hbase:meta, region=1588230740 is currently in transition
at org.apache.hadoop.hbase.master.assignment.AssignmentManager.preTransitCheck(AssignmentManager.java:536)
at org.apache.hadoop.hbase.master.assignment.AssignmentManager.createMoveRegionProcedure(AssignmentManager.java:592)
at org.apache.hadoop.hbase.master.assignment.AssignmentManager.moveAsync(AssignmentManager.java:609)
at org.apache.hadoop.hbase.master.HMaster.balance(HMaster.java:1707)
at org.apache.hadoop.hbase.master.HMaster.balance(HMaster.java:1622)
at org.apache.hadoop.hbase.master.balancer.BalancerChore.chore(BalancerChore.java:49)
at org.apache.hadoop.hbase.ScheduledChore.run(ScheduledChore.java:186)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at org.apache.hadoop.hbase.JitterScheduledThreadPoolExecutorImpl$JitteredRunnableScheduledFuture.run(JitterScheduledThreadPoolExecutorImpl.java:111)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)

This is a serious problem because it often occurs when new RSs started or old RSs failover. And what's more, no effective methods can be used to make the balance of the cluster back to normal.

But the solution of this problem may be simple. We can cache Exceptions when implementing a plan, and then just skip it, avoiding failed plans effect later plans in the whole plans list. New calls of balance can fetch up the failed and skipped plans.