EzraDB
v0.1 · pre-release

Where EzraDB is, and where it’s going.

EzraDB is pre-1.0. Here’s exactly what’s shipped today, what’s next, and what we’re still building — so you can decide with real information, not marketing.

Pre-1.0, and honest about it. Formally model-checked core, reproducible benchmarks, a public roadmap.

Nowv0.1Nextv0.2LaterplannedExploringresearch

This is a direction, not a commitment. Scope and order change. We ship a thing when it’s correct, not when a date says so.

Now
v0.1

Shipped · single-node

  • PostgreSQL-18 wire, SCRAM/TLS, PL/pgSQL, system catalogs
  • Row + column + hybrid engines (USING …), MVCC (SI + SSI), WAL + crash recovery
  • Vectorized columnar OLAP — Cascades optimizer, SIMD codecs, PREWHERE, zone maps
  • ASOF joins, AS OF SYSTEM TIME (row engine), candlestick / VWAP
  • BACKTEST — walk-forward, deterministic, Sharpe / Sortino / Calmar / max-DD
  • Streaming ingest pipelines (Kafka / Kinesis / NATS)
  • Vector KNN (brute-force), full-text search, PLV8 (JavaScript) UDFs
  • Materialized + incremental materialized views (restart-durable)
  • Instant clones — sub-second CREATE CLONE, copy-on-write, any size
  • Redis-compatible KV — Streams, Lua, pub/sub (unmodified BullMQ passes)
  • Arrow Flight SQL endpoint · pg_dump round-trip · binary COPY
  • RLS, column-level GRANT/REVOKE, hash-chained audit log
  • Prometheus / OpenTelemetry, built-in admin console (MVP)
Next
v0.2

In progress

  • Distributed cluster mode — Raft per range, automatic failover
  • Multi-region geo-partitioning — REGIONAL BY ROW, GLOBAL, leaseholder pinning
  • AS OF SYSTEM TIME on column / hybrid tables
  • Point-in-time recovery (physical WAL replay)
  • Async row→column replicas · point-in-time & multi-node clones
Later
planned

Real in the repo, not yet surfaced

  • Vector ANN index (HNSW / IVF)
  • Finance window functions in SQL — SMA / EMA / RSI / MACD / Bollinger / ATR
  • Distributed UDF execution — fan-out by volatility & parallel-safety
  • Approximate analytics — APPROX_COUNT_DISTINCT / PERCENTILE / TOP_K
  • WASM UDFs (LANGUAGE plwasm)
  • Production Kubernetes operator
  • KMS envelope encryption, FIPS, PII-FPE, dual-control
  • Benchmark scale ladder — TPC-H SF10 → SF100
Exploring
research

Under consideration

  • GPU-accelerated aggregation kernels
  • Serverless scale-to-zero
  • CDC sinks (Kafka / Webhook / S3)
  • Additional streaming sources (PubSub)
  • Timescale-style sugar (SAMPLE BY, create_hypertable, policies)

PRE-1.0, PLAINLY

We’d rather tell you the edges than have you find them.

What “pre-1.0” means

  • Single-node today — distributed cluster mode is v0.2.
  • No security SLA yet, and we don’t recommend running it on production data.

What is solid

  • The transaction core is formally model-checked.
  • Every TPC-H query is byte-identical across engines.
  • You can reproduce our benchmarks with one command.

Watching this list?

We're heads-down building and not taking signups yet — this page is the honest record of what ships and when.

Explore the product →