Meanwhile, my biggest issue with elasticsearch is that for any kind of system where you have indices of different sizes (eg you have several data models being indexed), your cluster will make no attempt at trying to distribute smartly and instead will stop at "same number of indices per node". Which means you very often end up with a very unbalanced cluster with regards to disk usage. For earlier versions there was a plugin that had a smarter allocator, which would look at disk usage when assigning nodes, but for any recent ES cluster (>2.0) you're left with a shrug.