ONE DATABASE
A platform inside a single engine.
EzraDB is one database — but inside it are engines that usually take five systems to assemble: a transactional store, a columnar warehouse, a time-series/quant engine, a streaming layer, a Redis-compatible cache, search, and in-database compute. One copy of your data. One wire protocol. One thing to operate.
*In development — not yet production-ready.
Transactional engine
A serious OLTP core — MVCC, WAL, real Postgres semantics.
- RocksDB-backed LSM row store
- MVCC with snapshot + serializable (SSI) isolation
- WAL group-commit + tested crash recovery
- Savepoints, sequences, FK CASCADE, JSONB
- Secondary indexes: hash, GIN, GiST, SP-GiST
- RANGE / LIST / HASH partitioning
Analytical engine
A vectorized columnar warehouse living in the same database.
- Arrow / Parquet column store with zone maps & bloom filters
- SIMD codecs: Gorilla, Delta, Double-Delta, Dictionary, RLE, FOR
- PREWHERE + runtime filters, morsel-driven NUMA parallelism
- Cascades cost-based optimizer — Rust-native, no JVM
- 3–5× Postgres / Timescale / ClickHouse on TPC-H SF1; finishes all 22
HTAP fabric
Both engines over one MVCC timeline. Pick per table, keep one copy.
- CREATE TABLE … USING row | column | hybrid
- ALTER TABLE … SET ACCESS METHOD — convert in place
- Freshness-routed hybrid reads (designed sub-second row→column)*
- Row-vs-column parity gate: all 22 TPC-H queries byte-identical
Time-series & quant
A kdb-class time engine that speaks SQL.
- ASOF JOIN / LEFT ASOF JOIN / WITHIN INTERVAL
- AS OF SYSTEM TIME time-travel (row engine; column/hybrid next)
- candlestick_agg — OHLC, volume, VWAP
- Deterministic BACKTEST — walk-forward, Sharpe / Sortino / Calmar / max-DD
- CREATE TICKS TABLE … RETENTION · LATEST ON
- Market-data parsers: Polygon.io, IEX DEEP, Databento*
- SQL finance functions: SMA / EMA / RSI / MACD / Bollinger / ATR*
Streaming & CDC
Data in and out, continuously.
- CREATE STREAMING PIPELINE — Kafka, Kinesis, NATS
- Exactly-once ingest with DEDUP KEY, backpressure, dead-lettering
- Arrow Flight SQL endpoint — queries + bulk ingest (port 5434)
- CREATE CHANGEFEED — Debezium-style JSON envelopes*
- PostgreSQL logical-replication wire + binary COPY
Redis-compatible KV & queues
A cache and job-queue engine inside your database.
- RESP2/RESP3 listener — strings, hashes, lists, sets, sorted sets
- Streams with consumer groups (XADD / XREADGROUP / XACK)
- MULTI/EXEC transactions, pub/sub, blocking ops
- Lua scripting — EVAL / EVALSHA
- Runs unmodified BullMQ — drop-in for job queues
- Tiered persistence for KV data*
Search & AI
Full-text and vectors without another system.
- Full-text search — tsvector / tsquery / ts_rank, GIN-indexed
- pgvector-compatible vector type — <-> L2, <#> inner, <=> cosine
- KNN queries: ORDER BY embedding <-> $1 LIMIT k
- HNSW / IVF approximate-nearest-neighbor index*
In-database compute
Bring the code to the data — JavaScript today, distributed & GPU next.
- PL/pgSQL — cursors, triggers, exceptions, SECURITY DEFINER
- PLV8 JavaScript (QuickJS) — plv8.execute / prepare / subtransaction
- Per-function memory & timeout limits; trigger + SETOF support
- WASM UDFs (wasmtime, LANGUAGE plwasm)*
- Distributed UDF execution — fan-out by volatility / parallel-safety*
- GPU-accelerated aggregation kernels*
Resilience & scale
Spanner-class ambitions, engineered in the open of our roadmap.
- Instant clones — CREATE CLONE, sub-second copy-on-write at any size
- pg_dump round-trip + physical backup
- Hot / warm / cold storage tiering (S3 cold tier in preview)
- Per-range Raft consensus with automatic leader failover*
- Horizontal scale-out by adding nodes*
- Multi-region: REGIONAL BY ROW, GLOBAL tables, follower reads*
- Point-in-time recovery via WAL replay*
- Scale-to-zero serverless*
Security & operations
Run it like Postgres. Watch it like a modern system.
- SCRAM-SHA-256 auth, TLS, reads your postgresql.conf / pg_hba.conf
- Row-level security (CREATE POLICY) + column-level GRANT
- BLAKE3 hash-chained, tamper-evident audit log
- Prometheus /metrics, OpenTelemetry tracing, slow-query log
- Built-in admin console — cluster, nodes, ranges, Raft (MVP)
- Self-tuning index & layout advisor*
- KMS envelope encryption, FIPS profiles, PII tokenization*
DROP-IN
It speaks fluent Postgres.
PostgreSQL-18 wire protocol, system catalogs, and your existing postgresql.conf. Full ORM test suites run against EzraDB and stock Postgres 18 — same behavior, verified.
# EzraDB speaks the PostgreSQL wire protocol — any PG client connects
psql -h your-host -p 5432 -U ezra -d ezra -c 'SELECT version();'
-- PostgreSQL 18.0 (EzraDB 0.1)
ONE PROCESS, MANY DOORS
Judge it by the numbers.
The benchmark methodology is fully disclosed and reproducible with one command — and the roadmap tells you honestly what's next.
*In development — not yet production-ready.