zhaohai666 opened a new pull request, #693:
URL: https://github.com/apache/rocketmq-dashboard/pull/693
# PR: Refactor LlmGateway to Interface\-Based Provider Pattern
## Overview
Refactor the original monolithic single\-implementation `LlmGateway` and
reconstruct it into a standard interface\-based provider pattern architecture,
improving system scalability, testability and maintainability for
multi\-LLM\-provider access scenarios\.
## Core Changes
1. **Decouple LLM Gateway Logic**
Split the original integrated LLM gateway business logic into a unified
abstract interface and two independent concrete implementations:
- `LlmGatewayImpl`: Official HTTP client implementation for production
environment API requests
- `LlmGatewayStub`: Mock stub implementation for local frontend
development and unit testing
2. **Unify Configuration Parsing Logic**
Introduce `LlmSettingsResolver` to take charge of unified LLM configuration
parsing\. It replaces the standalone `LlmConfigService` and scattered
configuration parsing logic in `LlmController`\.
3. **Clean Up Redundant Legacy Components**
Simplify the implementation of `AiController` and `AiService` layers, and
remove obsolete redundant components including `OpenAiCompatibleLlmClient` and
`OpenAiCompatibleLlmGateway`\.
4. **Optimize MCP Tool Registration**
Clean up the entire `tool/` subpackage \(contains 10 independent tool
handler files\)\. Inline all tool definition logic directly into
`McpServerImpl` to simplify package structure\.
## Architecture Design Details
### 1\. Standardized Core LlmGateway Interface
The abstract `LlmGateway` interface defines three universal standardized
methods to unify the interaction specification of all LLM providers:
- `chat()`: Process and execute LLM chat completion requests
- `testConnection()`: Verify network connectivity and availability of the
target LLM provider endpoint
- `listModels()`: Fetch and return the list of available models from the LLM
provider
### 2\. Production Implementation \(LlmGatewayImpl\)
Built based on Spring WebClient, supporting multi\-provider standard
specifications natively\. Compatible with mainstream LLM services including
OpenAI, Azure OpenAI, DeepSeek, Tongyi Qwen, Ollama and AWS Bedrock, realizing
unified adaptation of multiple LLM providers\.
### 3\. Mock Implementation \(LlmGatewayStub\)
Provides static predefined mock response data\. It supports offline local
development and independent unit testing without relying on real LLM service
resources, greatly improving development and testing efficiency\.
### 4\. Unified Configuration Resolution
The newly added `LlmSettingsResolver` loads and parses all LLM\-related
configurations through `SettingsService`, abandoning the independent
`LlmConfigService`, realizing centralized management of configuration logic and
reducing code redundancy\.
### 5\. Simplified MCP Tool Architecture
Removed all independent standalone `ToolHandler` classes\. All MCP tool
definition and registration logic is uniformly embedded in `McpServerImpl`,
optimizing code cohesion and eliminating scattered tool management logic.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
