Metrics Platform

Step 1: Prepare an AnalyticDB for MySQL instance

1. You have created an Enterprise Edition or Basic Edition cluster.

   Important

   Model inference is computationally intensive. When creating an instance, we recommend the following:

   • Use the Subscription billing method for cost-effectiveness.

   • If you have already created a model inference service, set the Enable AI Node option to No.

   • If you have not yet created a model inference service, we recommend enabling AI nodes in the Ask. Select the ADB.MLTensor.5 specification and set the quantity to 8. When you create the inference service later, the system automatically applies the prepaid resources if you select the same AI node specification.

   • If you do not purchase AI node resources when you create the instance, the model inference service defaults to the pay-as-you-go billing method.

2. You have created a database account and configured an IP whitelist.

Step 2: Create a model inference service

1. On your instance details page, go to AI Applications > Inference Service.

2. Create a new inference service and configure it as follows:

   • Model Selection: Select Metrics Platform Model (Platform + Knowledge Base).

   • Resource Configuration: Select the appropriate AI node specification and quantity.

     • Platform (base) - Worker configuration: We recommend the resource specification ADB.MLTensor.5 with 8 resource cards.

     • Knowledge base (emb) - Worker configuration: We recommend the resource specification ADB.MLTensor.2 with 1 resource card.

     • Remove the storage mount.

3. After the service is created, view the Invocation Information in the Service Call column. Record the Endpoint URL and API Key for later use.

Step 3: Create the Metrics Platform

1. On the instance details page, go to AI Applications > Metrics Analysis.

2. On the Metrics Platform list page, click Create Metrics Analysis Platform. In the Create Metrics Analysis Platform dialog box, enter the model inference service information obtained in the previous step: in the Base Model Endpoint field, enter the Endpoint URL recorded in Step 2. Then click OK.

   

Scenario 1: Same-VPC deployment

• Description: The model inference service and Metrics Platform are deployed in an AnalyticDB for MySQL instance within the same VPC.

• Network path: Metrics Platform -> VPC internal network -> model inference service.

• Procedure: When you create the Metrics Platform, use the model access endpoint within the VPC. No IP whitelist configuration is required.

Scenario 2: Different VPC deployment (public network access)

• Description: The Metrics Platform needs to access the model inference service, which is deployed in another VPC, over the public network.

• Network path: Metrics Platform -> user VPC -> public NAT gateway (SNAT) -> public network -> model inference service.

• Procedure:

  1. Create a public NAT gateway for the Metrics Platform's VPC and configure SNAT rules to enable public network access. You can skip this step if public network access is already enabled.

  2. Obtain the outbound public IP address of the public NAT gateway.

  3. In the security group or IP whitelist of the AnalyticDB for MySQL instance that hosts the model inference service, add the public IP address obtained in the previous step.

Scenario 3: Different VPC deployment (GA acceleration)

• Description: Cross-region deployment that uses Global Accelerator (GA) to accelerate public network access.

• Network path: Metrics Platform -> user VPC -> public NAT gateway -> Global Accelerator (GA) instance -> model inference service.

• Procedure:

  1. Configure public network access for the Metrics Platform's VPC (see Step 1 in Scenario 2).

  2. Create a Global Accelerator (GA) instance.

  3. In the access control settings of the GA instance, add the IP address of the public NAT gateway for the Metrics Platform.

  4. In the IP whitelist of the AnalyticDB for MySQL instance that hosts the model inference service, add the service IP address provided by the GA instance.

First-time login

1. On the Metrics Analysis List page of your instance, click the target Platform service address.

   

2. Log in with the default system credentials. The default username and password for the administrator account are both admin. For security, change your password immediately after your first login.

3. After you log in, the Ordinary User page appears.

Chat Assistant

The Creator is the core interface for all intelligent analysis. A successful analysis typically involves the following key elements:

• A clear analysis request: In the conversation box, describe your needs as precisely as possible, including:

  • Data source: Which table or file do you want to analyze?

  • Data scope: Do you want to analyze all data or a subset that meets specific conditions?

  • Analysis action: Do you want to see trends, find the best/worst performers, or get optimization suggestions?

  • Business terms: If your request includes domain-specific terms, abbreviations, or formulas, ensure they are already in the knowledge base.

• Understand the core modes: Before making a request, adjust the conversation mode as needed.

  Mode type	Option	Description
  Conversation mode	Data Retrieval and Analysis (Default)	Automatically retrieves data from the database, generates code or SQL, and returns a report based on your request.
  	Open conversation	Allows simple, context-aware conversations, similar to a large language model. You can also use it to further interpret generated analysis reports.
  Execution mode	Ask (Default)	The platform executes the request directly without pre-generating an execution plan. This lightweight interaction is ideal for exploratory, drill-down queries. In this mode, you cannot view plan details or save the session as a scenario template.
  	Plan	The platform first generates a global execution plan based on your request. You can preview and edit the plan before executing it. This provides a controllable analysis process, suitable for generating stable metrics or reports.
  Execution method	Automatic (Default)	After you submit a request, the analysis proceeds automatically step-by-step. The platform controls the entire process, and you cannot intervene except to stop the execution.
  	Manual	After you submit a request, you execute each step manually. In Plan mode, you can also roll back steps.

• Use auxiliary functions:

  • Specify Database: Temporarily specify a data source for analysis in the current conversation.

  • Local Upload: Upload local files such as .xlsx and .csv for supplementary analysis.

  • Online Search: Allows the platform to search the public internet for additional context when its knowledge is insufficient. This is disabled by default.

Advanced features