CockroachDB

View Current Viewing Revision #12 from 11/11/2019 10:07 a.m.

CockroachDB is a scalable, fault-tolerant, SQL database built on a transactional and strongly-consistent key-value store. It is backed by RocksDB and uses the Raft distributed consensus algorithm to ensure consistency.

History

CockroachDB is inspired by Google’s Spanner and F1 to address fault tolerance and distributed consistency.

The backend storage in CockroachDB is RocksDB. It is disk-oriented storage.

Data Model

Relational

Concurrency Control

Multi-version Concurrency Control (MVCC)

Transactions in CockroachDB is implemented

Isolation Levels

Serializable Snapshot Isolation

CockroachDB supports serializable snapshot isolation (SSI). It is implemented using RocksDB snapshot capability.

Storage Model

Custom

CockroachDB stores its data in the distributed key-value store.

Storage Organization

Log-structured

Indexes

Log-Structured Merge Tree

CockroachDB's indexes are stored in RocksDB's LSM tree implementation.

Query Compilation

Not Supported

Foreign Keys

Supported

CockroachDB supports foreign keys.

System Architecture

Shared-Nothing

CockroachDB has two layers, the SQL layer and the storage layer. The SQL layer sits on top of the transactional and strongly-consistent distributed key-value store. In the key-value store, the key ranges are divided and stored in RocksDB and replicated across cluster. It exports structured data API of relational concepts. It exports standard SQL interface at the SQL layer. The SQL layer translates SQL statements into calls to structured data API.

Query Execution

Tuple-at-a-Time Model Vectorized Model

CockroachDB supports distributed SQL execution. The gateway query node is responsible for creating a DAG of SQL processors and distributing them across the cluster, like in F1. The individual SQL processors operate in either a tuple-at-a-time mode, which supports all SQL queries in CockroachDB, or a vectorized mode, which supports a subset of SQL, depending on query cardinality.

Stored Procedures

Not Supported

Joins

Nested Loop Join Hash Join Sort-Merge Join Shuffle Join Semi Join Index Nested Loop Join

CockroachDB supports all standard SQL joins. Distributed joins are performed via a shuffle by hash strategy, or a batched nested loop strategy. Broadcast joins are not supported.