fix: honor dashscope embedding batch limit

backend/src/kbqa/indexing/worker.py

@@ -119,7 +119,7 @@ class IndexWorker:
             await self._mark_failed(document_id, "INDEXING_FAILED", "文档索引失败，请重试")
 
     async def _embed_and_upsert(self, document_id: str, chunks: list[PreparedChunk]) -> bool:
-        batch_size = 20
+        batch_size = 10
         for start in range(0, len(chunks), batch_size):
             batch = chunks[start : start + batch_size]
             vectors = await self.embeddings.embed_documents([chunk.content for chunk in batch])

backend/src/kbqa/rag/embeddings.py

@@ -1,8 +1,12 @@
+import logging
+
 from langchain_openai import OpenAIEmbeddings
 
 from kbqa.api.errors import AppError
 from kbqa.config import Settings, validate_live_settings
 
+logger = logging.getLogger(__name__)
+
 
 class EmbeddingProvider:
     def __init__(self, settings: Settings) -> None:
@@ -17,7 +21,7 @@ class EmbeddingProvider:
                 base_url=self.settings.dashscope_base_url,
                 model=self.settings.embedding_model,
                 dimensions=self.settings.embedding_dim,
-                chunk_size=20,
+                chunk_size=10,
                 max_retries=2,
                 timeout=60.0,
                 check_embedding_ctx_length=False,
@@ -40,6 +44,7 @@ class EmbeddingProvider:
         except AppError:
             raise
         except Exception as exc:
+            logger.exception("Embedding document request failed error_type=%s", type(exc).__name__)
             raise AppError(
                 "EMBEDDING_UPSTREAM_ERROR",
                 "文档向量化失败，请稍后重试",
@@ -54,6 +59,7 @@ class EmbeddingProvider:
         except AppError:
             raise
         except Exception as exc:
+            logger.exception("Embedding query request failed error_type=%s", type(exc).__name__)
             raise AppError(
                 "EMBEDDING_UPSTREAM_ERROR",
                 "查询向量化失败，请稍后重试",