reading processor dirs can hang
As noted by @Chiara in EP1528, processes can hang on startup.
Running with timeStampMaster with non-distributed case directories, but with case directories that contain different portions of the processorNN structure. Eg,
node0:
/scratch/job10/constant, /scratch/job10/system, /scratch/processor{0,1,2,3}
node1:
/scratch/job10/constant, /scratch/job10/system, /scratch/processor{4,5,6,7}
...
The startup optimization logic added in #1946 (closed) results in a false listing since the directory information is scattered from node0, which only has processors (0-3). Proposed fix is to place the optimization onto a switch, and restore the previous behaviour of polling on all nodes by default.