We use a large Impala/Kudu cluster for our analytic reporting. HDFS is not
on the critical path for this, and after experimenting with a cluster
configuration without it, we simply added a co-located HDFS cluster. It
turns out we use HDFS for our dimension staging, and I'm guessing Impala
uses it
Hi Manuel,
You definitely need a Hive metastore to store the table metadata. You
don't need HDFS specifically - you can run with various other default
filesystems (including the local filesystem) and we test in various of
those configurations. We don't test without any filesystem - it may work