[Proposal] Build observability of spring-ai-alibaba.

[Cirilla-zmh]

Backstory

So far, spring-ai-alibaba has achieved basic observability for some key scenarios (partly thanks to the foundational work of spring-ai), but more effort is needed to make it as comprehensive as possible. This proposal is mainly aimed at discussing the future evolution plan for the observability of this project, and you are welcome to freely comment and join our discussion.

Principles

The observability of spring-ai-alibaba should be built upon the foundation of spring-ai, with potentially some additional observability data being recorded. When a similar observability capability is being developed by spring-ai, we should avoid duplicating efforts as much as possible, unless it is very important or urgent. We would also strive to integrate existing observability capabilities into the spring-ai project, making them more widely accepted by users.

More importantly, the observability of spring-ai-alibaba needs to meet to the OpenTelemetry semconv as closely as possible.

RoadMap

Basic Observability

• dashscope chat model #224
• dashscope embedding model #225
• dashscope image model
• dashscope audio model
• chat client (implemented by spring-ai)
• advisor (implemented by spring-ai)

RAG Components

RAG components may need extended attributes. However, there is no specification provided by either opentelemetry or spring-ai, so we need to set a convention in advance.

• rag convention
• retriever
• reranker

Function Calling

spring-ai is already planning to do something similar. See this issue for further information.

Documents for Best Practices

• how to export observability data

Debug Tool

TODO cc @chickenlj

Aliyun Application Real-Time Monitoring Service (ARMS) Integration

• integration of aliyun java llm sdk
• zero-code integration with aliyun java agent