PolarDB:Enterprise editions

Cluster edition

Cluster Edition uses a compute and storage decoupling architecture to provide rapid elastic scaling, larger storage capacity, and low latency between the primary and read-only nodes.

• Single writer, multiple readers

  PolarDB uses a distributed cluster architecture. A Cluster Edition cluster contains one primary node and up to 15 read-only nodes, or just a single primary node. These database nodes form the database engine layer. The primary node processes read and write requests, while read-only nodes process only read requests. The primary and read-only nodes use Active-Active failover to provide a high-availability database service.

• Shared distributed storage (PolarStore)

  Multiple compute nodes share a single copy of data instead of each node storing its own copy. This design significantly reduces your storage costs. Based on a distributed block store and file system, storage can be scaled smoothly online, independent of single-server capacity limits. This allows the cluster to support datasets up to several hundred terabytes.

• Compute and storage decoupling

  The compute and storage decoupling architecture enables elastic cluster scaling in public cloud environments. The database compute nodes store only metadata. Storage nodes store data files and redo logs in the shared distributed storage (PolarStore). Compute nodes only need to synchronize redo log metadata, which greatly reduces replication latency between the primary and read-only nodes. If the primary node fails, the system can quickly promote a read-only node to become the new primary.

• Read/write splitting

  The read/write splitting feature of PolarProxy is a transparent, highly available, and adaptive load balancing capability that is provided by default with the Cluster Edition. Through the cluster endpoint, SQL requests are automatically forwarded to the various database nodes of the Cluster Edition, which provides aggregated, high-throughput concurrent SQL processing. For more information, see What is read/write splitting.

• High-speed interconnection

  A high-speed network interconnects the compute and storage nodes of the database. These nodes use the Remote Direct Memory Access (RDMA) protocol to transfer data. This eliminates I/O performance bottlenecks.

• Data reliability and consistency

  Storage nodes replicate data across multiple copies to ensure reliability, and the system uses the Parallel-Raft protocol to guarantee data consistency.

Architecture

The following figure shows the architecture of a Cluster Edition cluster:

Supported kernel versions

Cluster Edition supports three kernel versions of PolarDB for MySQL: 5.6, 5.7, and 8.0.

Use cases

• Production databases for medium- and large-sized enterprises.

• Databases for industries such as Internet, IoT, retail, e-commerce, logistics, and gaming.

• Core databases for industries with high data security requirements, such as finance, securities, and insurance.

Node specifications and pricing

Cluster Edition supports dedicated specifications and general-purpose specifications. For more information, see Billable items.

Multi-master cluster (Limitless)

PolarDB for MySQL launches Multi-master Cluster (Limitless), which upgrades the architecture from single-writer, multi-reader to multi-writer, multi-reader and enhances the database's concurrent read and write capabilities.

Architecture

All data files in the cluster are stored in the shared distributed storage (PolarStore). The read/write nodes share these data files by using the distributed file system (PolarFileSystem). You can access the entire cluster through the cluster endpoint, and PolarProxy automatically routes SQL statements to the correct read/write node.

Supported kernel versions

Multi-master Cluster (Limitless) is available only with the 8.0 kernel version of PolarDB for MySQL.

Benefits

• Horizontal write scaling in seconds

  This architecture distributes concurrent writes across different compute nodes, supporting up to 32 nodes for simultaneous writes. You can dynamically schedule different databases across these nodes in seconds, which greatly improves overall concurrent read and write throughput.

• Multi-primary failover (no standby nodes required)

  If a read/write node fails, traffic fails over in seconds to another read/write node with low traffic. This model eliminates the need for idle hot standby resources, potentially reducing costs by half.

• Global read-only nodes

  A global read-only node provides a consolidated view of the data from all read/write nodes. This is useful for executing queries that require data consolidation.

Use cases

Node specifications and pricing

Multi-master Cluster (Limitless) supports dedicated specifications and general-purpose specifications. For more information, see Compute node specifications for Enterprise Edition.

For information about the billing for Multi-master Cluster (Limitless), see Billable items.