SQL and planner
Parsing, expressions, statistics, and physical operator selection.
RadixDB architecture
Predictable data handling from write to scan
The engine separates mutable state from cold storage while retaining one SQL interface, transactional visibility, and bounded resource use.
The data path
Transaction layer
MVCC, READ COMMITTED and SNAPSHOT, savepoints, and statement atomicity.
Hot row versions
Current changes remain available without rewriting cold blocks.
Columnar blocks
Immutable compressed segments for sequential and selective reads.
WAL and recovery
Write-ahead logging, checkpoints, and verified post-crash startup.
SQL
Familiar SQL. Explicit boundaries.
RadixDB supports DDL, DML, JOINs, aggregates, subqueries, CTEs, window functions, RETURNING, and ON CONFLICT. Where compatibility is limited, the manual says so directly.
BEGIN ISOLATION LEVEL SNAPSHOT;
UPDATE stock_balances
SET quantity = quantity - 4
WHERE product_id = :product
AND revision = :expected
RETURNING quantity, revision;
COMMIT;Index access
Indexes shaped for the query
- BTREE for ranges and ordering
- HASH for exact lookups
- BITMAP for selective attributes
- Composite and partial indexes
Application interfaces
From SQL to applications
Embedded Rust
Open databases and run queries directly in-process.
TCP client
Connect over the network to radixdb-server.
ORM facade
A typed application interface over queries and records.
Server programming
Routines, triggers, and jobs evolve as a separate layer.