- Returns an ack after persisting a document to both master and replica node.
- Resident should be configurable such as FIFO.
- Keeps indexes on node where corresponding to the document is stored in.
- Persists on the disk and then asynchronously load into the RAM
- All documents should be replicated at least on one another node (or exactly one another node???)
- Sharding algorithm is another question. But it will be handled on the server side. Not client side.
- It sacrifices availability during network partition. So it's CP.
- ACID?
- Leader election or leaderless model?