[moe]

generally the merge layer kills you because you must query all nodes to support global indexes. that is not to say that its impossible.

It's not that bad actually. PK lookups and range queries are rather easy because mongos (the sharding daemon) knows the shard layout and only queries nodes that hold data relevant to the query.

Other query types, especially joins, are much harder. But frankly, PK lookups and range queries go a long way for most applications.

The real effort is in making all this, especially the automatic shard rebalancing, not only reliable but absolutely bulletproof. That seems to be on mongo's roadmap and if they get there that would be a major selling point.