This topic describes how to use AI clusters, LogService clusters, and object storage together to help you quickly get started with the deployment and management of shared storage architectures.
Architecture overview
OceanBase AI clusters adopt a shared storage architecture. The data plane is jointly supported by object storage and LogService, enabling more extreme elasticity.
Core components
Component |
Role |
Relationship |
Description |
|---|---|---|---|
| Object Storage | Data Persistence Layer | 1:N | Stores all business data (baseline data + incremental data) as the final destination of data. One object storage configuration can be used by multiple AI clusters, but the data of each cluster is isolated through different directory paths. |
| LogService cluster | Log Service Layer | 1:1 | Provides log persistence capabilities, decouples compute nodes from high-availability requirements, and supports cross-AZ disaster recovery. A LogService cluster can only be bound to one AI cluster. |
| AI Cluster | Compute Service Layer | - | Retain only the local cache disk to provide read and write services, with data pulled from object storage on demand. |
┌─────────────────────────────────────────────────────────┐
│ OCP Console │
├─────────────────────────────────────────────────────────┤
│ │
│ Object storage configuration │
│ │ │
│ ▼ │
│ Logservice cluster ◄─── Log persistence / Cross-AZ disaster recovery │
│ │ │
│ ▼ │
│ AI cluster (compute nodes) ◄─── Local cache disk │
│ │ │
│ └───────────┬───────────────┘ │
│ │ │
│ ▼ │
│ Object storage │
│ (Final data persistence location) │
│ │
└─────────────────────────────────────────────────────────┘
Note
Although multiple AI clusters can share one object storage configuration, each cluster's data is isolated by different directory paths and does not affect each other. It is recommended to specify an independent directory path for each cluster when creating it.
Procedure
Step 1: Create an object storage configuration
Before creating an AI cluster, you must configure object storage in the external integration module:
- Log in to OCP, choose Administration > External Integration > Object Storage.
- Click Create Object Storage Configuration.
- Specify the configuration information:
- Name: Enter a custom name for the configuration. It is recommended to use a name that can distinguish different businesses or environments.
- Protocol Type: Select S3, OSS, or OBS.
- Endpoint: The access address of the object storage.
- Region: The region where the object storage is located. If region information is not available, you can enter any value.
- AccessKey ID/Secret: The access credentials.
For more information, see Create an object storage configuration.
Step 2: Create a Logservice cluster
A Logservice cluster provides log service capabilities for an AI cluster and has a one-to-one relationship with the AI cluster.
- Log in to OCP, and choose Construction > Logservice Instance.
- Click Create Cluster.
- Specify the cluster information and you can select an application parameter template.
- Click Finish.
For more information, see Create a Logservice cluster.
Step 3: Create an AI cluster
When creating an AI cluster, associate it with an object storage and a Logservice:
- Log in to OCP, and choose Cluster > Create Cluster.
- Select the AI Cluster type.
- Configure the basic information (cluster name, password, version, etc.).
- Select the created object storage configuration.
- Select the created Logservice cluster.
- Click Finish.
For more information, see Create an AI cluster.
Other related operations
Rotate AccessKeys
When periodically updating AccessKey credentials, it is recommended to follow this process:
- Create an object storage configuration: Create a configuration using the new credentials.
- Modify cluster reference: Switch the object storage of the AI cluster to the new configuration.
- Wait for effect: Wait at least one hour to ensure all nodes have completed credential updates.
- Delete the old configuration: After confirming the cluster is normal, delete the old configuration.
Delete a cluster
Consider the following when you delete an AI cluster:
- Deleting an AI cluster does not delete the data on the object storage.
- The associated Logservice cluster is set to the DEPRECATED state.
- The object storage configuration is retained for use by other clusters.