refactor: 客服agent的切片结构调整重构

rag/embedding.py

@@ -3,11 +3,13 @@ from __future__ import annotations
 
 import logging
 
+from config.settings import settings
 from tool.llm import llm
 
 
 logger = logging.getLogger("rag.embedding")
-EMBEDDING_DIMENSION = 768
+# 单一来源：.env 的 LLM_EMBED_DIMENSIONS；Milvus 建表、入库校验、embeddings 请求参数均以此为准
+EMBEDDING_DIMENSION = settings.llm.embed_dimensions
 
 
 class EmbeddingError(RuntimeError):

rag/milvus_collections.py

@@ -4,10 +4,10 @@ from __future__ import annotations
 from pymilvus import AsyncMilvusClient, DataType
 
 from config.database.milvus import client as configured_milvus_client
+from rag.embedding import EMBEDDING_DIMENSION
 
 
 KNOWLEDGE_COLLECTIONS = ("fin_faq", "fin_fund_doc", "fin_policy")
-EMBEDDING_DIMENSION = 768
 
 
 def build_knowledge_schema():
@@ -44,3 +44,15 @@ async def ensure_collections(milvus_client: AsyncMilvusClient | None = None) ->
                 schema=schema,
                 index_params=index_params,
             )
+            continue
+        # 已存在的集合维度必须与当前 embedding 配置一致，否则入库/检索会在运行时失败
+        desc = await client.describe_collection(collection_name)
+        for field in desc.get("fields", []):
+            if field.get("name") != "vector":
+                continue
+            dim = field.get("params", {}).get("dim")
+            if dim is not None and int(dim) != EMBEDDING_DIMENSION:
+                raise RuntimeError(
+                    f"Milvus collection {collection_name!r} vector dim={dim}, "
+                    f"expected {EMBEDDING_DIMENSION}; drop and recreate it"
+                )