The default metadata store can cause all kinds of interesting problems for concurrent users ruing jobs against the server; it's an interesting choice for amazon to use it at all (especially since they use a local mysql as the metadata store for hive on other deployments)
So I'm wondering: Is it feasible (and recommended?) to get a 5.0 emr deployment of spark up to par (that is, install mysql locally on the master node and use it as the metadata store), without killing the cluster in the process? (This is a relevant question, since it appears that many services and libraries are installed at very specific versions to result in a stable cluster)
If it is feasible/recommended, has any one publicly posted a script or tested instructions that will achieve this?