FeatureStore supports offline and online data stores. Offline data stores are for daily calculated offline feature data, while online data stores are for offline features and real-time feature data required by online services. Before creating a FeatureStore project, you must configure offline and online data stores. This topic describes how to configure data stores.
Offline data stores support MaxCompute. It can store attribute tables of feature entities such as users and items, and data tables produced after statistics and derivation, such as user offline feature tables, product feature tables, and advertisement feature tables. Offline feature data is typically not directly called by online services. Instead, FeatureStore's data synchronization feature can write it into online data stores for invocation.
Online data stores include high-performance storage services such as Hologres, and Tablestore, which support millisecond-level invocation by online services.
Prerequisites
You have prepared a project or an instance based on the type of data store.
Offline data store
If the data store type is MaxCompute, you must create a MaxCompute project.
Online data store
If the data store type is FeatureDB, you must activate FeatureDB as prompted when creating the data store.
If you want to configure VPC direct connection, you need a VPC, a vSwitch, and a security group. If not, create a VPC, a vSwitch and a security group.
If the data store type is Hologres, you must purchase a Hologres instance.
Note: For the VPC configuration of Hologres, see instance configurations.
You must create a database on the Hologres instance.
If the data store type is GraphCompute, you must create a GraphCompute instance.
If the data store type is Tablestore, you must create a Tablestore instance.
Procedure
Go to the FeatureStore page.
Log on to the PAI console. In the left-side navigation pane, choose Data Preparation > FeatureStore.
Select a workspace and click Enter FeatureStore.
On the Store tab, click Create Store. Configure the new offline or online data store.
Offline data store: MaxCompute
Configure parameters for the MaxCompute data store.
Type: MaxCompute
Name: Enter a custom name for the data store.
MaxCompute Project Name: Select the MaxCompute project you have created.
Click Submit.
Online data store: FeatureDB
Configure parameters for the FeatureStore data store.
Activate FeatureDB: If you are using FeatureDB for the first time, follow the prompts on the interface to activate it. FeatureDB is billed on a pay-as-you-go basis, determined by storage volume and the number of read/write operations. For billing rules, see Billing of FeatureStore.
Name: The name is fixed as feature_db.
Username: The username for the FeatureDB instance.
Password: The password for the FeatureDB instance.
(Optional) Complete the VPC Network High-speed Connectivity Configuration. Once configured successfully, you can utilize the FeatureStore SDK for direct access to FeatureDB via a private network in the VPC, which enhances data read/write performance and reduces access latency.
VPC: Choose the VPC associated with the online service that utilizes FeatureStore.
Zone and Switch: Choose the appropriate zone and switch, making sure to select the switch corresponding to the zone of your active service machine. For enhanced business continuity and stability, it is advisable to choose switches across a minimum of two zones to ensure high availability.
Security Group Name: Choose the security group you have created within the VPC.
Click Submit .
Online data store: Hologres
Configure parameters for the Hologres data store.
Type: Hologres
Name: Enter a custom name for the data store.
Instance ID: Select the Hologres instance ID you have created.
Database Name: Choose the database associated with the Hologres instance you created.
(Optional) If you are prompted that you do not have permissions to access the Hologres instance when configuring the Hologres data store, perform the following steps to complete the authorization:
Log on to the Hologres console. In the left-side navigation pane, click Instances. Select the desired Hologres instance. On the instance details page, click Account Management.
On the User Management page, click Add User. In the dialog box that appears, select the RAM user who will use FeatureStore. Then, select Examples of the Super Administrator (SuperUser) or Regular User. Then click OK.
(Optional) If you select Regular User, you must also authorize the RAM user to access the database.
Go to the instance details page and click Database Management.
Click Authorize User in the Actions column.
Click Grant Permissions in the upper right corner.
Select the RAM user using FeatureStore and the role of Developer. Click OK.
(Optional) If you are still prompted that you do not have permissions after authorization, check whether the Hologres instance version is lower than 2.0.18. If so, upgrade the instance to the latest version.
Click Submit.
Online data store: GraphCompute
Configure parameters for the GraphCompute data store.
Type: GraphCompute
Name: Enter a custom name for the data store.
Instance ID: Select the GraphCompute instance ID you have created.
Store Address, Username, and Password: Set to the Endpoint, Username and Password of your GraphCompute instance.
Click Submit.
What to do next
After configuring the data store, you can create a FeatureStore project.
References
For more information about FeatureDB, see FeatureDB overview.