YugabyteDB — Distributed SQL That Is Fully PostgreSQL Compatible

Introduction

YugabyteDB takes the PostgreSQL query layer — indexing, extensions, JSONB, stored procedures, triggers — and puts it on top of a distributed storage engine (DocDB, a fork of RocksDB). The result is a database that feels exactly like Postgres to your ORM, but scales horizontally and survives zone/region failures.

With over 10,000 GitHub stars, YugabyteDB is used by General Motors, Kroger, Fidelity, and hundreds of companies that need Postgres semantics at cloud scale. It's a practical alternative to Spanner/Cosmos when you want PostgreSQL.

What YugabyteDB Does

YugabyteDB runs two APIs: YSQL (PostgreSQL wire protocol, 99% syntax compatible) and YCQL (Cassandra wire protocol for semi-structured). Data is sharded across nodes via consistent hashing and replicated via Raft. Distributed ACID transactions use the two-phase commit protocol with hybrid logical clocks.

Architecture Overview

Clients -> YSQL (Postgres wire) or YCQL (Cassandra wire)
              |
   [Query Layer — reuses Postgres PG code]
              |
   [DocDB Storage Engine]
     Sharded tablets (Raft-replicated)
     Based on RocksDB + Raft + Hybrid Logical Clocks
              |
   +----+----+----+----+
   |    |    |    |    |
 [Node A] [Node B] [Node C] ... (tablet peers)
              |
   [Coordinator master processes]
     Metadata, leader elections, load balancing

Self-Hosting & Configuration

-- Geo-distributed table: user rows co-located with user's region
CREATE TABLE user_profiles (
  user_id BIGINT,
  region  TEXT,
  data    JSONB,
  PRIMARY KEY ((region), user_id)
) SPLIT INTO 3 TABLETS;

-- Tablespaces for placement policies
CREATE TABLESPACE us_east WITH (
  replica_placement = '{"num_replicas": 3, "placement_blocks": [
    {"cloud":"aws","region":"us-east-1","zone":"us-east-1a","min_num_replicas":1},
    {"cloud":"aws","region":"us-east-1","zone":"us-east-1b","min_num_replicas":1},
    {"cloud":"aws","region":"us-east-1","zone":"us-east-1c","min_num_replicas":1}
  ]}'
);

ALTER TABLE user_profiles SET TABLESPACE us_east;

Key Features

• PostgreSQL compatible — YSQL runs real Postgres query code
• Distributed transactions — ACID across rows, tablets, regions
• Horizontal scale — add nodes, tablets rebalance automatically
• Raft replication — 3 or 5 replicas per tablet for HA
• Multi-region — synchronous or asynchronous replication modes
• YCQL — Cassandra wire protocol for key-value/wide-column use cases
• Postgres extensions — PostGIS, pgvector, pg_cron, etc. supported
• YugabyteDB Aeon — managed service on AWS/GCP/Azure

Comparison with Similar Tools

FAQ

Q: YugabyteDB vs CockroachDB? A: Both aim for distributed SQL. Yugabyte reuses actual Postgres code for YSQL — the richest Postgres compatibility. Cockroach has Postgres-like syntax but a custom parser. For apps that rely on Postgres extensions (pgvector, PostGIS), Yugabyte wins.

Q: Can I just run Postgres with logical replication instead? A: Yes, if you're OK with one primary + read replicas and asynchronous replication. Yugabyte gives you active-active writes across nodes/regions with ACID guarantees — a different scale/resilience tier.

Q: Is Yugabyte open source? A: Yes, Apache-2.0 core. Some advanced management features (xCluster, security) are in the Enterprise/Aeon tiers.

Q: How is it for workloads with complex joins? A: Joins across tablets incur network shuffles. Co-locate related rows (hash partition by customer_id) to keep joins fast. Yugabyte's planner handles distributed joins, but design matters for multi-region deployments.

Sources

• GitHub: https://github.com/yugabyte/yugabyte-db
• Docs: https://docs.yugabyte.com
• Company: Yugabyte
• License: Apache-2.0