AnalyticDB for MySQL clusters in elastic mode for Cluster Edition use a compute-storage separation architecture. Storage nodes handle data storage, writing, searching, and scanning. In workloads that combine high-throughput writes with large data scans—such as concurrent ingestion and analytics on the same cluster—storage nodes can become an I/O or CPU bottleneck, degrading read/write performance. To restore performance, scale out elastic I/O resources in the AnalyticDB for MySQL console. Scale in when workloads decrease to reduce costs.
An elastic I/O unit (EIU) is the unit that measures storage performance for these clusters. The following table lists the performance metrics per EIU.
| Metric | Value per EIU |
|---|---|
| CPU |
|
| Maximum hot data storage | 8 TB |
| IOPS | 16,800–50,000 |
| Throughput | 350 MB/s |
How EIU scale-out works
EIU scale-out uses an online shard migration approach: new capacity is provisioned and data is migrated before the original shards go offline. This keeps the cluster available throughout the process—no cluster restart is required, and data reads and writes continue uninterrupted. EIU resources can be scaled out and scaled in at any time based on business requirements.
When you add an EIU, the cluster migrates a subset of shards from existing nodes to the new node using a load balancing algorithm. The migration follows these steps:
-
Add a node. A new EIU node joins the cluster.
-
Migrate shards online. The cluster takes snapshots of the shards to migrate and transfers them to the new node. During migration, the shards on the original node remain online and continue serving reads and writes. The new node loads the snapshots and begins appending shard logs.
-
Activate the new node. When the shard logs on the new node catch up with the original node, the new node starts accepting new writes and queries.
-
Clean up the original node. The migrated shards on the original node are gracefully stopped. After existing queries against those shards finish, the shards go offline and are cleared.
Scale-in follows a similar process in reverse. Scale in EIUs when business workloads decrease to reduce resource costs.