-
your cluster is too big and too small
-
re-size mid job — why not?
-
tuning cluster to the job is way easier than v/v
-
your data isn’t that big (yet)
-
with small teams, worry about downtime not utilization
-
encourage profligacy — humans are important, robots are cheap. "Don’t want data scientist watching a cluster whose hourly rate is smaller than hers."