bug in SGEPlugin usage within cluster.py - makes master_is_exec_host always False

[jtriley]

From http://mailman.mit.edu/pipermail/starcluster/2012-March/001109.html

Setup a cluster today (0.93.2) and suddenly noticed that the 'master' node was
not being reported in a "qstat -f" command and was not accepting run jobs from
the queue . . . i.e., with 12 nodes x 8 cpus each (96), when 96 jobs are
submitted, only 88 run (nodes 1-11) while 8 remain in the queue waiting.

I tried restarting the cluster using the 'sge' plugin to manually ensure that
master_is_exec_host was set to 'True'. But the result was the same: 88 running

• 8 waiting.