Concepts

The mental model, smallest to largest:

• Documents — JSON values, stored as BSON, addressed by a stable id.
• Collections — named groups of documents within a database.
• Indexes — persistent B-trees over document paths for fast lookups.
• Transactions — atomic multi-document writes, durable through a write-ahead log.
• Replication — followers stream a leader’s changes and can take over.
• Sharding — a stateless router spreads collections across rings.

A single process can host many databases; one .dfdb file is one database with its own page cache, indexes, and write-ahead log.