From query to core.
A document database with a MongoDB-compatible front door and an LSM storage foundation.
Two entry points. One foundation.
The standalone server accepts MongoDB wire-protocol connections. The embedded Go package bypasses the network and uses the engine in process. These are two integration styles, with different operational surfaces.
The standalone path
| Layer | Responsibility |
|---|---|
| Wire handler | Connections, BSON messages, command dispatch, authentication integration. |
| Query processing | Parsing, planning, execution and aggregation operations. |
| Storage engine | Keys and values, memory tables, write-ahead logging and sorted tables. |
The storage model
The LSM design batches writes in memory and maintains a write-ahead log for recovery. Sorted string tables store flushed data on disk; compaction reorganizes tables over time. Bloom filters and caching help avoid unnecessary reads.
Durability depends on the selected WAL policy and the behavior of the filesystem. Measure your actual workload and test crash recovery with your chosen settings.
Beyond a single node
The repository includes Raft replication and sharding packages. A package’s presence does not prove that every distributed topology is ready for a particular deployment. Review the integration paths and exercise failover, resharding and recovery before depending on them.
Read the implementation
The server command is the clearest starting point for understanding which configuration and instrumentation are wired into the standalone runtime. The embedded package exposes a smaller direct API.
Grounded in the upstream source · snapshot 6b186803 · reviewed 19 October 2026.
Architecture guide Storage engine Server wiring