Transition zoostore deferral tracking to use System.nanoTime

[keith-turner]

Describe the bug

When a FATE operation is unreserved with a deferral time, the code uses System.currentTimeMillis() to determine when the operation should run again. In the case where wall time changes on the machine this could cause this deferral tracking to be incorrect, in the worse case causing something to not run for long periods of time.

Expected behavior

If would be best if the deferral time tracking in ZooStore used System.nanoTime instead of System.currentTimeMillis as this avoids problems with wall time changing. We could also modify the method signature of

public void unreserve(long tid, long deferTime)

to

public void unreserve(long tid, long deferTime, TimeUnit deferTimeUnuit)

All the code that calls this method would still pass milliseconds, so making the time unit explicit would make the code less confusing if the ZooStore impl uses nano seconds and its routinely passed milliseconds.

[keith-turner]

Not sure of the versions where this should be fixed. It could be fixed in 2.1 and would be a nice fix to have there as long as it does not introduce new bugs.

[keith-turner]

This is one place where the deferral happens in zoostore

accumulo/core/src/main/java/org/apache/accumulo/core/fate/ZooStore.java

Line 289 in 4d38127

defered.put(tid, System.currentTimeMillis() + deferTime);

[cshannon]

The AccumuloStore will also need to be updated that was added in #4049

[kevinrr888]

I would like to work on this

[keith-turner]

Done in #4126