feat: implement retry mechanism for model requests in anthropic and openai providers

Added instructions for deploying AstrBot using AstrBot Launcher.

README.md

@@ -81,6 +81,10 @@ uv tool install astrbot
 astrbot
 ```
 
+#### 启动器一键部署（AstrBot Launcher）
+
+进入 [AstrBot Launcher](https://github.com/Raven95676/astrbot-launcher) 仓库，在 Releases 页最新版本下找到对应的系统安装包安装即可。
+
 #### 宝塔面板部署
 
 AstrBot 与宝塔面板合作，已上架至宝塔面板。

astrbot/cli/__init__.py

@@ -1 +1 @@
-__version__ = "4.17.0"
+__version__ = "4.17.2"

astrbot/core/astr_main_agent.py

@@ -42,6 +42,7 @@ from astrbot.core.message.components import File, Image, Reply
 from astrbot.core.platform.astr_message_event import AstrMessageEvent
 from astrbot.core.provider import Provider
 from astrbot.core.provider.entities import ProviderRequest
+from astrbot.core.provider.manager import llm_tools
 from astrbot.core.skills.skill_manager import SkillManager, build_skills_prompt
 from astrbot.core.star.context import Context
 from astrbot.core.star.star_handler import star_map
@@ -769,6 +770,14 @@ def _plugin_tool_fix(event: AstrMessageEvent, req: ProviderRequest) -> None:
             if plugin.name in event.plugins_name or plugin.reserved:
                 new_tool_set.add_tool(tool)
         req.func_tool = new_tool_set
+    else:
+        # mcp tools
+        tool_set = req.func_tool
+        if not tool_set:
+            tool_set = ToolSet()
+        for tool in llm_tools.func_list:
+            if isinstance(tool, MCPTool):
+                tool_set.add_tool(tool)
 
 
 async def _handle_webchat(

astrbot/core/config/default.py

@@ -5,7 +5,7 @@ from typing import Any, TypedDict
 
 from astrbot.core.utils.astrbot_path import get_astrbot_data_path
 
-VERSION = "4.17.0"
+VERSION = "4.17.2"
 DB_PATH = os.path.join(get_astrbot_data_path(), "data_v4.db")
 
 WEBHOOK_SUPPORTED_PLATFORMS = [

astrbot/core/log.py

@@ -299,7 +299,9 @@ class LogManager:
     ) -> int:
         os.makedirs(os.path.dirname(file_path) or ".", exist_ok=True)
         rotation = f"{max_mb} MB" if max_mb and max_mb > 0 else None
-        retention = f"{backup_count} files" if rotation else None
+        retention = (
+            backup_count if rotation and backup_count and backup_count > 0 else None
+        )
         if trace:
             return _loguru.add(
                 file_path,
@@ -363,13 +365,16 @@ class LogManager:
         if not enable_file:
             return
 
-        cls._file_sink_id = cls._add_file_sink(
-            file_path=cls._resolve_log_path(file_path),
-            level=logger.level,
-            max_mb=max_mb,
-            backup_count=3,
-            trace=False,
-        )
+        try:
+            cls._file_sink_id = cls._add_file_sink(
+                file_path=cls._resolve_log_path(file_path),
+                level=logger.level,
+                max_mb=max_mb,
+                backup_count=3,
+                trace=False,
+            )
+        except Exception as e:
+            logger.error(f"Failed to add file sink: {e}")
 
     @classmethod
     def configure_trace_logger(cls, config: dict | None) -> None:

astrbot/core/provider/sources/anthropic_source.py

@@ -22,6 +22,7 @@ from astrbot.core.utils.network_utils import (
 )
 
 from ..register import register_provider_adapter
+from .default import with_model_request_retry
 
 
 @register_provider_adapter(
@@ -204,6 +205,7 @@ class ProviderAnthropic(Provider):
         if usage.output_tokens is not None:
             token_usage.output = usage.output_tokens
 
+    @with_model_request_retry()
     async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
         if tools:
             if tool_list := tools.get_func_desc_anthropic_style():
@@ -265,6 +267,10 @@ class ProviderAnthropic(Provider):
 
         return llm_response
 
+    @with_model_request_retry()
+    async def _create_message_stream(self, payloads: dict, extra_body: dict):
+        return self.client.messages.stream(**payloads, extra_body=extra_body)
+
     async def _query_stream(
         self,
         payloads: dict,
@@ -293,9 +299,8 @@ class ProviderAnthropic(Provider):
                 "type": "enabled",
             }
 
-        async with self.client.messages.stream(
-            **payloads, extra_body=extra_body
-        ) as stream:
+        stream_ctx = await self._create_message_stream(payloads, extra_body)
+        async with stream_ctx as stream:
             assert isinstance(stream, anthropic.AsyncMessageStream)
             async for event in stream:
                 if event.type == "message_start":

astrbot/core/provider/sources/default.py

@@ -0,0 +1,38 @@
+from tenacity import (
+    AsyncRetrying,
+    retry,
+    retry_if_exception_type,
+    stop_after_attempt,
+    wait_exponential,
+)
+
+MODEL_REQUEST_RETRY_ATTEMPTS = 5
+MODEL_REQUEST_RETRY_WAIT_MAX_SECONDS = 15
+MODEL_REQUEST_RETRY_WAIT_MIN_SECONDS = 1
+MODEL_REQUEST_RETRY_WAIT_MULTIPLIER = 1
+
+
+def with_model_request_retry():
+    return retry(
+        retry=retry_if_exception_type(Exception),
+        stop=stop_after_attempt(MODEL_REQUEST_RETRY_ATTEMPTS),
+        wait=wait_exponential(
+            multiplier=MODEL_REQUEST_RETRY_WAIT_MULTIPLIER,
+            min=MODEL_REQUEST_RETRY_WAIT_MIN_SECONDS,
+            max=MODEL_REQUEST_RETRY_WAIT_MAX_SECONDS,
+        ),
+        reraise=True,
+    )
+
+
+def get_model_request_async_retrying() -> AsyncRetrying:
+    return AsyncRetrying(
+        retry=retry_if_exception_type(Exception),
+        stop=stop_after_attempt(MODEL_REQUEST_RETRY_ATTEMPTS),
+        wait=wait_exponential(
+            multiplier=MODEL_REQUEST_RETRY_WAIT_MULTIPLIER,
+            min=MODEL_REQUEST_RETRY_WAIT_MIN_SECONDS,
+            max=MODEL_REQUEST_RETRY_WAIT_MAX_SECONDS,
+        ),
+        reraise=True,
+    )

astrbot/core/provider/sources/gemini_source.py

@@ -21,6 +21,7 @@ from astrbot.core.utils.io import download_image_by_url
 from astrbot.core.utils.network_utils import is_connection_error, log_connection_failure
 
 from ..register import register_provider_adapter
+from .default import get_model_request_async_retrying, with_model_request_retry
 
 
 class SuppressNonTextPartsWarning(logging.Filter):
@@ -513,6 +514,7 @@ class ProviderGoogleGenAI(Provider):
                 llm_response.reasoning_signature = base64.b64encode(ts).decode("utf-8")
         return MessageChain(chain=chain)
 
+    @with_model_request_retry()
     async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
         """非流式请求 Gemini API"""
         system_instruction = next(
@@ -601,6 +603,17 @@ class ProviderGoogleGenAI(Provider):
         self,
         payloads: dict,
         tools: ToolSet | None,
+    ) -> AsyncGenerator[LLMResponse, None]:
+        async for attempt in get_model_request_async_retrying():
+            with attempt:
+                async for response in self._query_stream_once(payloads, tools):
+                    yield response
+                return
+
+    async def _query_stream_once(
+        self,
+        payloads: dict,
+        tools: ToolSet | None,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式请求 Gemini API"""
         system_instruction = next(
@@ -759,18 +772,7 @@ class ProviderGoogleGenAI(Provider):
 
         payloads = {"messages": context_query, "model": model}
 
-        retry = 10
-        keys = self.api_keys.copy()
-
-        for _ in range(retry):
-            try:
-                return await self._query(payloads, func_tool)
-            except APIError as e:
-                if await self._handle_api_error(e, keys):
-                    continue
-                break
-
-        raise Exception("请求失败。")
+        return await self._query(payloads, func_tool)
 
     async def text_chat_stream(
         self,
@@ -814,18 +816,8 @@ class ProviderGoogleGenAI(Provider):
 
         payloads = {"messages": context_query, "model": model}
 
-        retry = 10
-        keys = self.api_keys.copy()
-
-        for _ in range(retry):
-            try:
-                async for response in self._query_stream(payloads, func_tool):
-                    yield response
-                break
-            except APIError as e:
-                if await self._handle_api_error(e, keys):
-                    continue
-                break
+        async for response in self._query_stream(payloads, func_tool):
+            yield response
 
     async def get_models(self):
         try:

astrbot/core/provider/sources/openai_source.py

@@ -31,6 +31,7 @@ from astrbot.core.utils.network_utils import (
 from astrbot.core.utils.string_utils import normalize_and_dedupe_strings
 
 from ..register import register_provider_adapter
+from .default import get_model_request_async_retrying, with_model_request_retry
 
 
 @register_provider_adapter(
@@ -221,6 +222,7 @@ class ProviderOpenAIOfficial(Provider):
         except NotFoundError as e:
             raise Exception(f"获取模型列表失败：{e}")
 
+    @with_model_request_retry()
     async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
         if tools:
             model = payloads.get("model", "").lower()
@@ -246,8 +248,6 @@ class ProviderOpenAIOfficial(Provider):
         if isinstance(custom_extra_body, dict):
             extra_body.update(custom_extra_body)
 
-        model = payloads.get("model", "").lower()
-
         completion = await self.client.chat.completions.create(
             **payloads,
             stream=False,
@@ -269,6 +269,17 @@ class ProviderOpenAIOfficial(Provider):
         self,
         payloads: dict,
         tools: ToolSet | None,
+    ) -> AsyncGenerator[LLMResponse, None]:
+        async for attempt in get_model_request_async_retrying():
+            with attempt:
+                async for response in self._query_stream_once(payloads, tools):
+                    yield response
+                return
+
+    async def _query_stream_once(
+        self,
+        payloads: dict,
+        tools: ToolSet | None,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式查询API，逐步返回结果"""
         if tools:
@@ -323,7 +334,8 @@ class ProviderOpenAIOfficial(Provider):
                 llm_response.reasoning_content = reasoning
                 _y = True
             if delta.content:
-                completion_text = delta.content
+                # Don't strip streaming chunks to preserve spaces between words
+                completion_text = self._normalize_content(delta.content, strip=False)
                 llm_response.result_chain = MessageChain(
                     chain=[Comp.Plain(completion_text)],
                 )
@@ -371,6 +383,86 @@ class ProviderOpenAIOfficial(Provider):
             output=completion_tokens,
         )
 
+    @staticmethod
+    def _normalize_content(raw_content: Any, strip: bool = True) -> str:
+        """Normalize content from various formats to plain string.
+
+        Some LLM providers return content as list[dict] format
+        like [{'type': 'text', 'text': '...'}] instead of
+        plain string. This method handles both formats.
+
+        Args:
+            raw_content: The raw content from LLM response, can be str, list, or other.
+            strip: Whether to strip whitespace from the result. Set to False for
+                   streaming chunks to preserve spaces between words.
+
+        Returns:
+            Normalized plain text string.
+        """
+        if isinstance(raw_content, list):
+            # Check if this looks like OpenAI content-part format
+            # Only process if at least one item has {'type': 'text', 'text': ...} structure
+            has_content_part = any(
+                isinstance(part, dict) and part.get("type") == "text"
+                for part in raw_content
+            )
+            if has_content_part:
+                text_parts = []
+                for part in raw_content:
+                    if isinstance(part, dict) and part.get("type") == "text":
+                        text_val = part.get("text", "")
+                        # Coerce to str in case text is null or non-string
+                        text_parts.append(str(text_val) if text_val is not None else "")
+                return "".join(text_parts)
+            # Not content-part format, return string representation
+            return str(raw_content)
+
+        if isinstance(raw_content, str):
+            content = raw_content.strip() if strip else raw_content
+            # Check if the string is a JSON-encoded list (e.g., "[{'type': 'text', ...}]")
+            # This can happen when streaming concatenates content that was originally list format
+            # Only check if it looks like a complete JSON array (requires strip for check)
+            check_content = raw_content.strip()
+            if (
+                check_content.startswith("[")
+                and check_content.endswith("]")
+                and len(check_content) < 8192
+            ):
+                try:
+                    # First try standard JSON parsing
+                    parsed = json.loads(check_content)
+                except json.JSONDecodeError:
+                    # If that fails, try parsing as Python literal (handles single quotes)
+                    # This is safer than blind replace("'", '"') which corrupts apostrophes
+                    try:
+                        import ast
+
+                        parsed = ast.literal_eval(check_content)
+                    except (ValueError, SyntaxError):
+                        parsed = None
+
+                if isinstance(parsed, list):
+                    # Only convert if it matches OpenAI content-part schema
+                    # i.e., at least one item has {'type': 'text', 'text': ...}
+                    has_content_part = any(
+                        isinstance(part, dict) and part.get("type") == "text"
+                        for part in parsed
+                    )
+                    if has_content_part:
+                        text_parts = []
+                        for part in parsed:
+                            if isinstance(part, dict) and part.get("type") == "text":
+                                text_val = part.get("text", "")
+                                # Coerce to str in case text is null or non-string
+                                text_parts.append(
+                                    str(text_val) if text_val is not None else ""
+                                )
+                        if text_parts:
+                            return "".join(text_parts)
+            return content
+
+        return str(raw_content)
+
     async def _parse_openai_completion(
         self, completion: ChatCompletion, tools: ToolSet | None
     ) -> LLMResponse:
@@ -383,8 +475,7 @@ class ProviderOpenAIOfficial(Provider):
 
         # parse the text completion
         if choice.message.content is not None:
-            # text completion
-            completion_text = str(choice.message.content).strip()
+            completion_text = self._normalize_content(choice.message.content)
             # specially, some providers may set <think> tags around reasoning content in the completion text,
             # we use regex to remove them, and store then in reasoning_content field
             reasoning_pattern = re.compile(r"<think>(.*?)</think>", re.DOTALL)
@@ -394,6 +485,8 @@ class ProviderOpenAIOfficial(Provider):
                     [match.strip() for match in matches],
                 )
                 completion_text = reasoning_pattern.sub("", completion_text).strip()
+            # Also clean up orphan </think> tags that may leak from some models
+            completion_text = re.sub(r"</think>\s*$", "", completion_text).strip()
             llm_response.result_chain = MessageChain().message(completion_text)
 
         # parse the reasoning content if any
@@ -634,7 +727,7 @@ class ProviderOpenAIOfficial(Provider):
         extra_user_content_parts=None,
         **kwargs,
     ) -> LLMResponse:
-        payloads, context_query = await self._prepare_chat_payload(
+        payloads, _ = await self._prepare_chat_payload(
             prompt,
             image_urls,
             contexts,
@@ -646,47 +739,9 @@ class ProviderOpenAIOfficial(Provider):
         )
 
         llm_response = None
-        max_retries = 10
-        available_api_keys = self.api_keys.copy()
-        chosen_key = random.choice(available_api_keys)
-        image_fallback_used = False
-
-        last_exception = None
-        retry_cnt = 0
-        for retry_cnt in range(max_retries):
-            try:
-                self.client.api_key = chosen_key
-                llm_response = await self._query(payloads, func_tool)
-                break
-            except Exception as e:
-                last_exception = e
-                (
-                    success,
-                    chosen_key,
-                    available_api_keys,
-                    payloads,
-                    context_query,
-                    func_tool,
-                    image_fallback_used,
-                ) = await self._handle_api_error(
-                    e,
-                    payloads,
-                    context_query,
-                    func_tool,
-                    chosen_key,
-                    available_api_keys,
-                    retry_cnt,
-                    max_retries,
-                    image_fallback_used=image_fallback_used,
-                )
-                if success:
-                    break
-
-        if retry_cnt == max_retries - 1 or llm_response is None:
-            logger.error(f"API 调用失败，重试 {max_retries} 次仍然失败。")
-            if last_exception is None:
-                raise Exception("未知错误")
-            raise last_exception
+        if self.api_keys:
+            self.client.api_key = random.choice(self.api_keys)
+        llm_response = await self._query(payloads, func_tool)
         return llm_response
 
     async def text_chat_stream(
@@ -702,7 +757,7 @@ class ProviderOpenAIOfficial(Provider):
         **kwargs,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式对话，与服务商交互并逐步返回结果"""
-        payloads, context_query = await self._prepare_chat_payload(
+        payloads, _ = await self._prepare_chat_payload(
             prompt,
             image_urls,
             contexts,
@@ -712,48 +767,10 @@ class ProviderOpenAIOfficial(Provider):
             **kwargs,
         )
 
-        max_retries = 10
-        available_api_keys = self.api_keys.copy()
-        chosen_key = random.choice(available_api_keys)
-        image_fallback_used = False
-
-        last_exception = None
-        retry_cnt = 0
-        for retry_cnt in range(max_retries):
-            try:
-                self.client.api_key = chosen_key
-                async for response in self._query_stream(payloads, func_tool):
-                    yield response
-                break
-            except Exception as e:
-                last_exception = e
-                (
-                    success,
-                    chosen_key,
-                    available_api_keys,
-                    payloads,
-                    context_query,
-                    func_tool,
-                    image_fallback_used,
-                ) = await self._handle_api_error(
-                    e,
-                    payloads,
-                    context_query,
-                    func_tool,
-                    chosen_key,
-                    available_api_keys,
-                    retry_cnt,
-                    max_retries,
-                    image_fallback_used=image_fallback_used,
-                )
-                if success:
-                    break
-
-        if retry_cnt == max_retries - 1:
-            logger.error(f"API 调用失败，重试 {max_retries} 次仍然失败。")
-            if last_exception is None:
-                raise Exception("未知错误")
-            raise last_exception
+        if self.api_keys:
+            self.client.api_key = random.choice(self.api_keys)
+        async for response in self._query_stream(payloads, func_tool):
+            yield response
 
     async def _remove_image_from_context(self, contexts: list):
         """从上下文中删除所有带有 image 的记录"""

changelogs/v4.17.1.md

@@ -0,0 +1,34 @@
+## What's Changed
+
+hotfix of 4.17.0
+
+- 修复：当开启了 “启用文件日志” 后，无法启动 AstrBot，报错 `ValueError: Invalid unit value while parsing duration: 'files'`。这是由于日志轮转设置中保留配置错误导致的，已通过根据备份数量正确设置保留参数进行修复。
+- fix: When "Enable file logging" is turned on, AstrBot fails to start with error `ValueError: Invalid unit value while parsing duration: 'files'`. This is due to an incorrect retention configuration in the log rotation setup, which has been fixed by properly setting the retention parameter based on backup count. 
+
+### 新增
+- 新增 LINE 平台适配器与相关配置支持 ([#5085](https://github.com/AstrBotDevs/AstrBot/issues/5085))
+- 新增备用回退聊天模型列表，当主模型报错时自动切换到备用模型 ([#5109](https://github.com/AstrBotDevs/AstrBot/issues/5109))
+- 新增插件加载失败后的热重载支持，便于插件修复后快速恢复 ([#5043](https://github.com/AstrBotDevs/AstrBot/issues/5043))
+- WebUI 新增 SSL 配置选项并同步更新相关日志行为 ([#5117](https://github.com/AstrBotDevs/AstrBot/issues/5117))
+
+### 修复
+- 修复 Dockerfile 中依赖导出流程，增加 `uv lock` 步骤并移除不必要的 `--frozen` 参数，提升构建稳定性 ([#5091](https://github.com/AstrBotDevs/AstrBot/issues/5091), [#5089](https://github.com/AstrBotDevs/AstrBot/issues/5089))
+- 修复首次启动公告 `FIRST_NOTICE.md` 的本地化路径解析问题，补充兼容路径处理 ([#5083](https://github.com/AstrBotDevs/AstrBot/issues/5083), [#5082](https://github.com/AstrBotDevs/AstrBot/issues/5082))
+
+### 优化
+- 日志系统由 `colorlog` 切换为 `loguru`，增强日志输出与展示能力 ([#5115](https://github.com/AstrBotDevs/AstrBot/issues/5115))
+
+## What's Changed (EN)
+
+### New Features
+- Added LINE platform adapter support with related configuration options ([#5085](https://github.com/AstrBotDevs/AstrBot/issues/5085))
+- Added fallback chat model chain support in tool loop runner, with corresponding config and improved provider selection display ([#5109](https://github.com/AstrBotDevs/AstrBot/issues/5109))
+- Added hot reload support after plugin load failure for faster recovery during plugin development and maintenance ([#5043](https://github.com/AstrBotDevs/AstrBot/issues/5043))
+- Added SSL configuration options for WebUI and updated related logging behavior ([#5117](https://github.com/AstrBotDevs/AstrBot/issues/5117))
+
+### Fixes
+- Fixed Dockerfile dependency export flow by adding a `uv lock` step and removing unnecessary `--frozen` flag to improve build stability ([#5091](https://github.com/AstrBotDevs/AstrBot/issues/5091), [#5089](https://github.com/AstrBotDevs/AstrBot/issues/5089))
+- Fixed locale path resolution for `FIRST_NOTICE.md` and added compatible fallback handling ([#5083](https://github.com/AstrBotDevs/AstrBot/issues/5083), [#5082](https://github.com/AstrBotDevs/AstrBot/issues/5082))
+
+### Improvements
+- Replaced `colorlog` with `loguru` to improve logging capabilities and console display ([#5115](https://github.com/AstrBotDevs/AstrBot/issues/5115))

changelogs/v4.17.2.md

@@ -0,0 +1,8 @@
+## What's Changed
+
+hotfix of 4.17.0
+
+- 修复：MCP 服务器的 Tools 没有被正确添加到上下文中。
+- 修复：Electron 桌面应用部署时，系统自带插件未被正确加载的问题。
+- fix: Tools from MCP server were not properly added to context.
+- fix: built-in plugins were not properly loaded in Electron desktop application deployment.

desktop/package.json

@@ -1,6 +1,6 @@
 {
   "name": "astrbot-desktop",
-  "version": "4.17.0",
+  "version": "4.17.2",
   "description": "AstrBot desktop wrapper",
   "private": true,
   "main": "main.js",

desktop/scripts/build-backend.mjs

@@ -16,6 +16,8 @@ const kbStopwordsSrc = path.join(
   'hit_stopwords.txt',
 );
 const kbStopwordsDest = 'astrbot/core/knowledge_base/retrieval';
+const builtinStarsSrc = path.join(rootDir, 'astrbot', 'builtin_stars');
+const builtinStarsDest = 'astrbot/builtin_stars';
 
 const args = [
   'run',
@@ -35,9 +37,13 @@ const args = [
   'pip',
   '--collect-submodules',
   'astrbot.api',
+  '--collect-submodules',
+  'astrbot.builtin_stars',
   '--collect-data',
   'certifi',
   '--add-data',
+  `${builtinStarsSrc}${dataSeparator}${builtinStarsDest}`,
+  '--add-data',
   `${kbStopwordsSrc}${dataSeparator}${kbStopwordsDest}`,
   '--distpath',
   outputDir,

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "AstrBot"
-version = "4.17.0"
+version = "4.17.2"
 description = "Easy-to-use multi-platform LLM chatbot and development framework"
 readme = "README.md"
 requires-python = ">=3.12"