CockroachDB¶

Distributed SQL database with global consistency — PostgreSQL-compatible, survive anything, scale horizontally.

Overview¶

CockroachDB is a distributed SQL database that provides serializable ACID transactions, horizontal scaling, and multi-region survivability — all behind a PostgreSQL-compatible wire protocol. Data is automatically sharded into ranges, replicated via Raft consensus, and rebalanced across nodes. It survives disk, node, rack, and even datacenter failures without manual intervention.

Key Facts¶

Attribute	Detail
Website	cockroachlabs.com
Stars	~31k+ ⭐
Latest Version	v26.1 (2026)
Language	Go
License	BSL 1.1 (core) / CCL (enterprise features)
Company	Cockroach Labs
Wire Protocol	PostgreSQL-compatible

Evaluation¶

Pros	Cons
Survive datacenter failures automatically	BSL 1.1 license
PostgreSQL wire protocol compatible	Higher latency than single-node PG
Serializable isolation (strongest ACID)	Complex to tune for specific workloads
Horizontal scaling, automatic rebalancing	Expensive (especially managed CockroachDB Cloud)
Data locality / geo-partitioning	Single-row writes slower than PG
Built-in vector search (v26.1)
AI/agent-ready governance

Architecture¶

flowchart TB
    subgraph Cluster["CockroachDB Cluster"]
        subgraph Node1["Node 1"]
            SQL1["SQL Layer\n(PG wire protocol)"]
            KV1["Distributed KV"]
            Ranges1["Ranges\n(R1, R4, R7)"]
        end
        subgraph Node2["Node 2"]
            SQL2["SQL Layer"]
            KV2["Distributed KV"]
            Ranges2["Ranges\n(R2, R5, R8)"]
        end
        subgraph Node3["Node 3"]
            SQL3["SQL Layer"]
            KV3["Distributed KV"]
            Ranges3["Ranges\n(R3, R6, R9)"]
        end
    end

    Client["App (PG driver)"] --> SQL1
    Ranges1 <-->|"Raft replication"| Ranges2
    Ranges2 <-->|"Raft replication"| Ranges3

    style Cluster fill:#6933ff,color:#fff

Key Features (v26.1)¶

Feature	Detail
Raft consensus	Strong consistency, automatic leader election
Range-based sharding	Automatic split/merge at 64MB boundaries
Geo-partitioning	Pin data to regions for sovereignty/latency
Vector search	Built-in similarity search (no separate vector DB)
CDC	Streaming to Kafka, cloud storage
Zero-downtime DDL	Online schema changes, no maintenance windows
AI agent governance	Zero-trust access for AI workloads
Multi-region	Active-active across datacenters