原文
An experimental append-only database with time-travel queries.
- Append-only storage: All changes are immutable drift events
- Time travel: Query any historical state with
AS OFclauses - CRC-verified segments: Data integrity with checksummed records
- Secondary indexes: Fast lookups on indexed columns
- Snapshots & compaction: Optimize query performance
- Crash-safe: Atomic writes with fsync on segment boundaries
- Simple query language: DriftQL for easy interaction
# Clone and build
git clone https://github.com/driftdb/driftdb
cd driftdb
make build# Run the full demo (creates sample data and runs queries)
make demo# Initialize a database
driftdb init ./mydata
# Create a table
driftdb sql -d ./mydata -e 'CREATE TABLE users (pk=id, INDEX(email, status))'
# Insert data
driftdb sql -d ./mydata -e 'INSERT INTO users {"id": "user1", "email": "[email protected]", "status": "active"}'
# Query data
driftdb select -d ./mydata --table users --where 'status="active"'
# Time travel query
driftdb select -d ./mydata --table users --as-of "@seq:100"
# Show row history
driftdb drift -d ./mydata --table users --key "user1"
# Create snapshot for faster queries
driftdb snapshot -d ./mydata --table users
# Compact storage
driftdb compact -d ./mydata --table users-- Create table with indexes
CREATE TABLE orders (pk=id, INDEX(status, customer_id))
-- Insert full document
INSERT INTO orders {"id": "order1", "status": "pending", "amount": 100}
-- Partial update
PATCH orders KEY "order1" SET {"status": "paid"}
-- Soft delete (data remains for audit)
SOFT DELETE FROM orders KEY "order1"
-- Query with time travel
SELECT * FROM orders WHERE status="paid" AS OF "2025-01-01T00:00:00Z"
SELECT * FROM orders WHERE customer_id="cust1" AS OF "@seq:1000"
-- Show drift history
SHOW DRIFT orders KEY "order1"
-- Maintenance
SNAPSHOT orders
COMPACT ordersdata/
tables/<table>/
schema.yaml # Table schema definition
segments/ # Append-only event logs
00000001.seg
00000002.seg
snapshots/ # Materialized states
00000100.snap
indexes/ # Secondary B-tree indexes
status.idx
customer_id.idx
meta.json # Table metadata
- INSERT: Add new row with full document
- PATCH: Partial update by primary key
- SOFT_DELETE: Mark row as deleted (audit trail preserved)
[u32 length][u32 crc32][varint seq][u64 unix_ms][u8 event_type][msgpack payload]
- Single writer: Process-global lock prevents concurrent writes
- Multi-reader: Readers never block, use snapshots + replay
- Crash recovery: CRC verification + truncate corrupt tail
- Atomic operations: fsync on segment rotation and snapshots
# Run tests
make test
# Run benchmarks
make bench
# Format code
make fmt
# Run linter
make clippy
# Full CI checks
make ci- Snapshot every 100k events or 128MB (configurable)
- Compaction rewrites minimal segments with latest state
- Indexed columns use B-tree for O(log n) lookups
- Zstd compression for snapshots
- MessagePack for efficient event serialization
MIT
Experimental MVP - suitable for development and testing, not production use.