optimize error stack of module (#897)

Author: wzh1994

.github/workflows/main.yml

@@ -144,7 +144,7 @@ jobs:
         uses: ./.github/actions/run_tests
         with:
           working_directory: ${{ env.CI_PATH }}
-          changed_files: ${{ needs.Clone.outputs.changed_files }}
+          changed_files: ${{ needs.clone.outputs.changed_files }}
           markers: "not skip_on_linux"
           tests_dir: "tests/basic_tests"
           launcher: "empty"
@@ -163,13 +163,15 @@ jobs:
         uses: ./.github/actions/run_tests
         with:
           working_directory: ${{ env.CI_PATH }}
-          changed_files: ${{ needs.Clone.outputs.changed_files }}
+          changed_files: ${{ needs.clone.outputs.changed_files }}
           markers: "not skip_on_linux"
           tests_dir: "tests/advanced_tests"
 
   engine_tests:
     runs-on: tps_sco_nv
     needs: [ clone ]
+    if: |
+      contains(needs.clone.outputs.changed_files, 'lazyllm/engine/')
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -180,7 +182,7 @@ jobs:
         uses: ./.github/actions/run_tests
         with:
           working_directory: ${{ env.CI_PATH }}
-          changed_files: ${{ needs.Clone.outputs.changed_files }}
+          changed_files: ${{ needs.clone.outputs.changed_files }}
           markers: "not skip_on_linux"
           tests_dir: "tests/engine_tests"
 
@@ -230,7 +232,7 @@ jobs:
         uses: ./.github/actions/run_tests
         with:
           working_directory: ${{ env.CI_PATH }}
-          changed_files: ${{ needs.Clone.outputs.changed_files }}
+          changed_files: ${{ needs.clone.outputs.changed_files }}
           markers: "not skip_on_linux"
           tests_dir: "tests/charge_tests"
 
@@ -454,7 +456,7 @@ jobs:
               tests/charge_tests
           fi
         env:
-          CHANGED_FILES: ${{ needs.Clone.outputs.changed_files }}
+          CHANGED_FILES: ${{ needs.clone.outputs.changed_files }}
           LAZYLLM_KIMI_API_KEY: ${{ secrets.LAZYLLM_KIMI_API_KEY }}
           LAZYLLM_GLM_API_KEY: ${{ secrets.LAZYLLM_GLM_API_KEY }}
           LAZYLLM_GLM_MODEL_NAME: ${{ secrets.LAZYLLM_GLM_MODEL_NAME }}
@@ -580,7 +582,7 @@ jobs:
               --override-ini=cache_dir=tests/charge_tests/.pytest_cache tests/charge_tests
          fi
         env:
-          CHANGED_FILES: ${{ needs.Clone.outputs.changed_files }}
+          CHANGED_FILES: ${{ needs.clone.outputs.changed_files }}
           LAZYLLM_KIMI_API_KEY: ${{ secrets.LAZYLLM_KIMI_API_KEY }}
           LAZYLLM_GLM_API_KEY: ${{ secrets.LAZYLLM_GLM_API_KEY }}
           LAZYLLM_GLM_MODEL_NAME: ${{ secrets.LAZYLLM_GLM_MODEL_NAME }}

lazyllm/common/__init__.py

@@ -3,6 +3,8 @@
 from .common import FlatList, Identity, ResultCollector, ArgsDict, CaseInsensitiveDict
 from .common import ReprRule, make_repr, modify_repr, is_valid_url, is_valid_path, SingletonMeta, SingletonABCMeta
 from .common import once_flag, call_once, once_wrapper, singleton, reset_on_pickle, Finalizer
+from .inspection import _get_callsite
+from .exception import _trim_traceback, _register_trim_module, HandledException, _change_exception_type
 from .text import Color, colored_text
 from .option import Option, OptionIter
 from .threading import Thread, ThreadPoolExecutor
@@ -22,6 +24,9 @@
     '_get_base_cls_from_registry',
     'Register',
 
+    # inspection
+    '_get_callsite',
+
     # utils
     'FlatList',
     'ReadOnlyWrapper',
@@ -50,6 +55,12 @@
     'Finalizer',
     'redis_client',
 
+    # exception
+    '_trim_traceback',
+    '_register_trim_module',
+    'HandledException',
+    '_change_exception_type',
+
     # arg praser
     'LazyLLMCMD',
     'package',

lazyllm/common/exception.py

@@ -0,0 +1,61 @@
+import sys
+import types
+
+
+_expected_trip_modules = {}
+_expected_trip_continuous_modules = {}
+
+def _register_trim_module(expected: dict, continuous: bool = False):
+    expected = {k: [v] if isinstance(v, str) else v for k, v in expected.items()}
+    if continuous:
+        global _expected_trip_continuous_modules
+        _expected_trip_continuous_modules.update(expected)
+    else:
+        global _expected_trip_modules
+        _expected_trip_modules.update(expected)
+
+def _is_lazyllm_internal_frame(frame, continuous: bool = False):
+    global _expected_trip_modules, _expected_trip_continuous_modules
+    expected = _expected_trip_continuous_modules if continuous else _expected_trip_modules
+    mod = frame.f_globals.get('__name__', '')
+    if not mod.startswith('lazyllm.') and 'lazyllm/components/deploy/relay' not in frame.f_code.co_filename: return False
+    if mod in expected and frame.f_code.co_name in expected[mod]: return True
+    return False
+
+def _trim_traceback(tb):
+    kept, keep_call = [], True
+    global _expected_trip_modules, _expected_trip_continuous_modules
+    while tb:
+        if not _is_lazyllm_internal_frame(tb.tb_frame):
+            if _is_lazyllm_internal_frame(tb.tb_frame, continuous=True):
+                if keep_call:
+                    kept.append(tb)
+                    keep_call = False
+            else:
+                kept.append(tb)
+                keep_call = True
+        tb = tb.tb_next
+
+    new_tb = None
+    for tb in reversed(kept):
+        new_tb = types.TracebackType(tb_next=new_tb, tb_frame=tb.tb_frame, tb_lasti=tb.tb_lasti, tb_lineno=tb.tb_lineno)
+    return new_tb
+
+_original_excepthook = sys.excepthook
+
+def _lazyllm_excepthook(exc_type, exc_value, tb):
+    _original_excepthook(exc_type, exc_value, _trim_traceback(tb))
+
+sys.excepthook = _lazyllm_excepthook
+
+
+class HandledException(Exception):
+    ''' Builtin class. Exception that is intended to be handled exceptions. '''
+    pass
+
+def _change_exception_type(e, new_type):
+    new_exc = new_type(str(e)).with_traceback(e.__traceback__)
+    new_exc.__cause__ = e.__cause__
+    new_exc.__context__ = e.__context__
+    new_exc.__suppress_context__ = e.__suppress_context__
+    return new_exc

lazyllm/common/inspection.py

@@ -0,0 +1,14 @@
+import os
+import inspect
+
+def _get_callsite(depth: int = 1):
+    try:
+        frame = inspect.currentframe()
+        for _ in range(depth): frame = frame.f_back
+        if frame is None: return None
+        else:
+            while frame.f_code.co_name == '__setattr__' and frame.f_globals.get('__name__', '') == 'lazyllm.common.bind':
+                frame = frame.f_back
+        return f'"file: {os.path.abspath(frame.f_code.co_filename)}", line {frame.f_lineno}'
+    except Exception:
+        return None

lazyllm/components/deploy/relay/base.py

@@ -18,7 +18,7 @@ class RelayServer(LazyLLMDeployBase):
 
     def __init__(self, port=None, *, func=None, pre_func=None, post_func=None, pythonpath=None,
                  log_path=None, cls=None, launcher=launchers.remote(sync=False), num_replicas: int = 1,  # noqa B008
-                 security_key: Optional[str] = None):
+                 security_key: Optional[str] = None, defined_pos: Optional[str] = None):
         # func must dump in __call__ to wait for dependancies.
         self._func = func
         self._pre = dump_obj(pre_func)
@@ -27,6 +27,7 @@ def __init__(self, port=None, *, func=None, pre_func=None, post_func=None, pytho
         self._pythonpath = pythonpath
         self._num_replicas = num_replicas
         self._security_key = security_key
+        self._defined_pos = defined_pos
         super().__init__(launcher=launcher)
         self.temp_folder = make_log_dir(log_path, cls or 'relay') if log_path else None
 
@@ -49,12 +50,14 @@ def impl():
                 cmd += f'--num_replicas={self._num_replicas}'
             if self._security_key:
                 cmd += f'--security_key="{self._security_key}" '
+            if self._defined_pos:
+                cmd += '--defined_pos="{}" '.format(self._defined_pos.replace('"', r'\"'))
             if self.temp_folder: cmd += f' 2>&1 | tee {get_log_path(self.temp_folder)}'
             return cmd
 
         return LazyLLMCMD(cmd=impl, return_value=self.geturl,
                           checkf=verify_ray_func if config['use_ray'] else verify_fastapi_func,
-                          no_displays=['function', 'before_function', 'after_function', 'security_key'])
+                          no_displays=['function', 'before_function', 'after_function', 'security_key', 'defined_pos'])
 
     def geturl(self, job=None):
         if job is None:

lazyllm/components/deploy/relay/server.py

@@ -9,6 +9,7 @@
 import lazyllm
 from lazyllm import kwargs, package, load_obj
 from lazyllm import FastapiApp, globals
+from lazyllm.common import _trim_traceback, _register_trim_module
 import time
 import pickle
 import codecs
@@ -38,6 +39,7 @@
 parser.add_argument('--pythonpath')
 parser.add_argument('--num_replicas', type=int, default=1, help='num of ray replicas')
 parser.add_argument('--security_key', type=str, default=None, help='security key')
+parser.add_argument('--defined_pos', type=str, default=None, help='user defined positional')
 args = parser.parse_args()
 
 if args.pythonpath:
@@ -50,6 +52,12 @@
     after_func = load_obj(args.after_function)
 
 
+_register_trim_module({'__main__': ['async_wrapper', 'impl']})
+_err_msg = ('service of ServerModule execuate failed.\n\nThe above exception was the direct cause '
+            'of the following exception in service of ServerModule')
+_err_msg += (f' defined at `{args.defined_pos}`' if args.defined_pos else '') + ':\n'
+
+
 app = FastAPI()
 FastapiApp.update()
 
@@ -82,8 +90,10 @@ async def lazyllm_call(request: Request):
         return Response(content=codecs.encode(pickle.dumps(r), 'base64'))
     except requests.RequestException as e:
         return Response(content=f'{str(e)}', status_code=500)
-    except Exception as e:
-        return Response(content=f'{e}\n--- traceback ---\n{traceback.format_exc()}', status_code=500)
+    except Exception:
+        exc_type, exc_value, exc_tb = sys.exc_info()
+        formatted = ''.join(traceback.format_exception(exc_type, exc_value, _trim_traceback(exc_tb)))
+        return Response(content=f'{_err_msg}\n{formatted}', status_code=500)
 
 @app.post('/generate')
 @security_check
@@ -141,8 +151,10 @@ def generate_stream():
         return Response(content=impl(output))
     except requests.RequestException as e:
         return Response(content=f'{str(e)}', status_code=500)
-    except Exception as e:
-        return Response(content=f'{str(e)}\n--- traceback ---\n{traceback.format_exc()}', status_code=500)
+    except Exception:
+        exc_type, exc_value, exc_tb = sys.exc_info()
+        formatted = ''.join(traceback.format_exception(exc_type, exc_value, _trim_traceback(exc_tb)))
+        return Response(content=f'{_err_msg}\n{formatted}', status_code=500)
     finally:
         globals.clear()
 

lazyllm/flow/flow.py

@@ -2,7 +2,8 @@
 import builtins
 from lazyllm import config
 from lazyllm.common import LazyLLMRegisterMetaClass, package, kwargs, arguments, bind, root
-from lazyllm.common import ReadOnlyWrapper, LOG, globals, locals
+from lazyllm.common import ReadOnlyWrapper, LOG, globals, locals, _get_callsite
+from lazyllm.common import _register_trim_module, HandledException, _change_exception_type
 from lazyllm.common.bind import _MetaBind
 from functools import partial
 from contextlib import contextmanager
@@ -21,44 +22,10 @@
 from itertools import repeat
 
 
-class FlowException(Exception):
+class FlowException(HandledException):
     pass
 
 
-def _is_lazyllm_internal_frame(frame, expected: Optional[dict] = None):
-    expected = expected or {'lazyllm.flow.flow': ['_run', 'invoke'], 'lazyllm.common.bind': ['__call__']}
-    mod = frame.f_globals.get('__name__', '')
-    if not mod.startswith('lazyllm.'): return False
-    if mod in expected and frame.f_code.co_name in expected[mod]: return True
-    return False
-
-_is_lazyllm_call_frame = bind(_is_lazyllm_internal_frame, expected={'lazyllm.flow.flow': '__call__'})
-
-def _trim_traceback(tb):
-    kept, keep_call = [], True
-    while tb:
-        if not _is_lazyllm_internal_frame(tb.tb_frame):
-            if _is_lazyllm_call_frame(tb.tb_frame):
-                if keep_call:
-                    kept.append(tb)
-                    keep_call = False
-            else:
-                kept.append(tb)
-                keep_call = True
-        tb = tb.tb_next
-
-    new_tb = None
-    for tb in reversed(kept):
-        new_tb = types.TracebackType(tb_next=new_tb, tb_frame=tb.tb_frame, tb_lasti=tb.tb_lasti, tb_lineno=tb.tb_lineno)
-    return new_tb
-
-_original_excepthook = sys.excepthook
-
-def _lazyllm_excepthook(exc_type, exc_value, tb):
-    _original_excepthook(exc_type, exc_value, _trim_traceback(tb))
-
-sys.excepthook = _lazyllm_excepthook
-
 class _FuncWrap(object):
     def __init__(self, f):
         self._f = f._f if isinstance(f, _FuncWrap) else f
@@ -76,19 +43,6 @@ def __getattr__(self, __key):
             return getattr(self._f, __key)
         return super(__class__, self).__getattr__(__key)
 
-
-def _get_callsite(depth=1):
-    try:
-        frame = inspect.currentframe()
-        for _ in range(depth): frame = frame.f_back
-        if frame is None: return None
-        else:
-            while frame.f_code.co_name == '__setattr__' and frame.f_globals.get('__name__', '') == 'lazyllm.common.bind':
-                frame = frame.f_back
-        return f'"file: {os.path.abspath(frame.f_code.co_filename)}", line {frame.f_lineno}'
-    except Exception:
-        return None
-
 _oldins = isinstance
 def new_ins(obj, cls):
     if _oldins(obj, _FuncWrap) and os.getenv('LAZYLLM_ON_CLOUDPICKLE', None) != 'ON':
@@ -105,6 +59,9 @@ def _is_function(f):
 _flow_stack = threading.local()
 _flow_stack.value = []
 
+_register_trim_module({'lazyllm.flow.flow': ['__call__']}, continuous=True)
+_register_trim_module({'lazyllm.flow.flow': ['_run', 'invoke'], 'lazyllm.common.bind': ['__call__']})
+
 class FlowBase(metaclass=_MetaBind):
     def __init__(self, *items, item_names=None, auto_capture=False) -> None:
         self._father = None
@@ -315,8 +272,7 @@ def invoke(self, it, __input, *, bind_args_source=None, **kw):
                 return it(*__input, **kw) if isinstance(__input, package) else it(**__input, **kw)
             else:
                 return it(__input, **kw)
-        except FlowException as e:
-            raise e
+        except HandledException as e: raise e
         except Exception as e:
             try:
                 pos = self._item_pos[self._items.index(it)]
@@ -329,7 +285,7 @@ def invoke(self, it, __input, *, bind_args_source=None, **kw):
             LOG.error(err_msg)
             LOG.debug(f'Error type: {type(e).__name__}, Error message: {str(e)}\n'
                       f'Traceback: {"".join(traceback.format_exception(*sys.exc_info()))}')
-            raise FlowException(err_msg) from e
+            raise _change_exception_type(e, FlowException) from None
 
     def bind(self, *args, **kw):
         return bind(self, *args, **kw)

lazyllm/module/llms/onlinemodule/supplier/qwen.py

@@ -406,6 +406,8 @@ def _forward(self, input: str = None, negative_prompt: str = None, n: int = 1, p
         if self._api_key: call_params['api_key'] = self._api_key
         if seed: call_params['seed'] = seed
         task_response = dashscope.ImageSynthesis.async_call(**call_params)
+        if not task_response.output:
+            raise RuntimeError('Failed to create image synthesis task, may due to insufficient balance on your account.')
         response = dashscope.ImageSynthesis.wait(task=task_response.output.task_id, api_key=self._api_key)
         if response.status_code == HTTPStatus.OK:
             return encode_query_with_filepaths(None, bytes_to_file([requests.get(result.url).content