We solves complexity with simplicity.
Our Team is working on Elasticsearch since Elasticsearch 0.17.3 version (Mid of 2011). And we are handling elasticsearch having more than 30M documents. where size of our database is 20GB. To maintain such a big database, it really need to go through so many performance tuning iteration at OS level, at elasticsearch level and at data level. Our experience say, it is so important to understand what should be indexed, how mapping of an attribute should be defined. Because elasticsearch does not allow to change mapping of an attribute unless you reindex the whole index again. For large databases, we can not reindex again and again. So better to take expert help instead of hustling in wrong direction.
We have hand on experience to administrate entire cluster of ElasticSearch. We can help to
Note: You should never drop any node while resharding of ElasticSearch cluster regardless of number of replica you may have.