FYI: From my experience (on older ES versions) when you hit problems with it, they will be pretty bad, catastrophic problems. We run dedicated master nodes, but have the master servers running multiple ES instances, so they can serve as masters for multiple clusters.