What is Multi-Master Cluster (Limitless)
PolarDB for MySQL Multi-master Cluster (Limitless) provides horizontally scalable read/write capacity through multiple read/write (RW) nodes. Based on how data is organized and routed, Multi-master Cluster (Limitless) comes in two product forms: database/table-level multi-master and multi-master cluster partitioned tables.
Product forms
Database/table-level multi-master
Database/table-level multi-master distributes different databases or data objects across different RW nodes. Each database or data object is served by one RW node for reads and writes at any given moment, and you can switch its serving node without migrating data. This form is suitable for multi-tenant SaaS, region-sharded gaming, and workloads that already use database/table sharding middleware.
For more information, see What is multi-master database/table-level, Usage, and FAQ.
Multi-master cluster partitioned tables
Multi-master cluster partitioned tables automatically distribute the data of a logical table across multiple sharded databases and RW nodes based on distribution keys and splitting rules. The system performs transparent routing, distributed queries, and distributed DDL. This form is suitable for horizontal splitting of large tables, high-concurrency transactions, and workloads that want to reduce reliance on database/table sharding middleware.
For more information, see What are multi-master cluster partitioned tables and Usage instructions.
How to choose
|
Comparison |
Database/table-level multi-master |
Multi-master cluster partitioned tables |
|
Data distribution granularity |
Database or data object |
Data shards of a logical table |
|
Write mode |
Different databases or objects can be written concurrently on different RW nodes |
Different shards of the same logical table can be written concurrently on multiple RW nodes |
|
Scheduling |
Specify or switch the RW access point of a database or object |
The system automatically routes by distribution key |
|
Core syntax |
|
|
|
Typical migration |
Keep the existing database/table layout and adjust node ownership |
Choose distribution keys and splitting rules for logical tables |
If your workloads are already isolated by database or table and your main goal is to quickly spread the existing database/table load across more RW nodes, choose database/table-level multi-master. If the data volume or access pressure of a single table has become the bottleneck and you want the database to perform horizontal splitting transparently, choose multi-master cluster partitioned tables.
General capabilities
Global read-only nodes can read data on multiple RW nodes. In the database/table-level multi-master form, it runs aggregation queries across databases or data objects. In the multi-master cluster partitioned table form, it aggregates data from all shards to run cross-shard queries.