Azure Cognitive Search删除向量条目报错：缺少必填batch参数

我正在开发一个Azure Function，当Azure Cosmos DB容器中的文档被标记为已删除时触发。触发后，需要删除Cosmos DB中关联的消息，同时移除Azure Cognitive Search中对应的向量条目。以下是用于Azure Cognitive Search删除操作的相关代码：

import azure.functions as func
from azure.cosmos import CosmosClient, exceptions
from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient
import os
import logging

chat_deletion_trigger = func.Blueprint()

@chat_deletion_trigger.cosmos_db_trigger(
    connection="CosmosDBConnectionString",
    database_name="ChatDatabase",
    container_name="Chats",
    lease_container_name="leases",
    create_lease_container_if_not_exists=True,
    arg_name="documents"
)
def ChatDeletionTrigger(documents: func.DocumentList) -> None:
    logging.info("ChatDeletionTrigger function started processing.")

    cosmos_client = CosmosClient.from_connection_string(os.environ['CosmosDBConnectionString'])
    database_name = 'ChatDatabase'
    database = cosmos_client.get_database_client(database_name)
    messages_container = database.get_container_client('Messages')

    search_client = SearchClient(
        endpoint=os.getenv('SearchServiceEndpoint'),
        index_name="azureblob-index",
        credential=AzureKeyCredential(os.getenv('SearchServiceKey'))
    )

    for document in documents:
        chat_id = document.get('id')
        is_deleted = document.get('isDeleted', False)

        if not chat_id:
            logging.error("Document does not have an 'id' field. Skipping document.")
            continue

        if not is_deleted:
            logging.info(f"Chat with id: {chat_id} is not marked as deleted. Skipping document.")
            continue

        logging.info(f"Processing deletion of messages and vector entries associated with chat id: {chat_id}")

        try:
            messages_container.scripts.execute_stored_procedure(
                sproc="deleteMessagesByChatId",
                params=[chat_id],
                partition_key=chat_id
            )
            logging.info(f"Deleted messages associated with chatId {chat_id} using stored procedure.")

        except exceptions.CosmosHttpResponseError as e:
            logging.error(f"Error deleting messages for chat id {chat_id}: {str(e)}")

        try:
            results = search_client.search(search_text="", filter=f"chatId eq '{chat_id}'")

            document_ids = [doc['documentId'] for doc in results]

            if document_ids:
                delete_actions = [{"@search.action": "delete", "documentId": doc_id} for doc_id in document_ids]
                batch = {"value": delete_actions}
                search_client.index_documents(batch=batch)
                logging.info(f"Deleted vector entries for chatId: {chat_id}")

            else:
                logging.info(f"No vector entries found for chatId: {chat_id}")

        except Exception as e:
            logging.error(f"Error deleting vector entries for chat id {chat_id}: {str(e)}")

    logging.info("ChatDeletionTrigger function completed processing.")

执行删除操作时收到错误：
Error deleting vector entries for chat id {chat_id}: SearchClient.index_documents() missing 1 required positional argument: 'batch'

请问如何正确向Azure Cognitive Search的index_documents方法传递删除操作的批量参数？是批量参数结构错误，还是有其他遗漏？

错误源于index_documents方法的调用方式不正确：该方法不需要传入batch关键字参数，也不需要手动将操作列表包装成{"value": ...}的结构，直接传递操作对象列表即可。

修改后的代码片段 #

try:
    results = search_client.search(search_text="", filter=f"chatId eq '{chat_id}'")

    document_ids = [doc['documentId'] for doc in results]

    if document_ids:
        delete_actions = [{"@search.action": "delete", "documentId": doc_id} for doc_id in document_ids]
        # 直接传入操作列表，无需外层batch字典
        search_client.index_documents(delete_actions)
        logging.info(f"Deleted vector entries for chatId: {chat_id}")

    else:
        logging.info(f"No vector entries found for chatId: {chat_id}")

except Exception as e:
    logging.error(f"Error deleting vector entries for chat id {chat_id}: {str(e)}")

关键说明 #

• Azure Cognitive Search Python SDK的index_documents方法接受的参数是操作对象的列表，每个对象需包含@search.action（这里为delete）和文档的键字段（你的索引中是documentId）。
• SDK会自动处理请求格式的转换，无需手动添加{"value": ...}外层结构。
• 额外检查：确认documentId是你搜索索引中定义的键字段，如果键字段名称不同，需替换为对应名称，否则删除操作会因找不到目标文档而失效。

内容的提问来源于stack exchange，提问作者Shravan K Subrahmanya