Currently we use an in-process indexer over leveldb to index txs by tag (ie. so we can get tx history for a given account). It's design plan is here: #548

Instead, we should be streaming txs to an external process using postgresql to index everything.

There's some related discussion here: #674
and an early PR here: #548

I'm closing those in favor of a more focused issue and a future PR.

Note we should continue using leveldb as is for the blockchain and the state - we're just moving the extra tx indexing (so we can support lots of light clients from a single node and do whatever queries on txs).