I'm setting up a cluster running Hadoop 0.19.1. Everything I've read
says that, by default, Hadoop uses a FIFO queue that will run one map-
reduce job to completion before starting the next job. I have a small
4-node cluster, so this is exactly the behavior I want.
However, if I start two instances of the Hadoop examples that came in
the 0.19.1 tarball, the second job starts immediately, even while the
first is still in the map phase. Shouldn't the second job get queued
behind the first? If not, how can I configure the cluster for simple
FIFO scheduling?
I can get the capacity scheduler or the fair scheduler to work
properly, but neither is apparently able to give each job exclusive
use of the cluster.
-Logan
