refactor: option to log exception

Signed-off-by: thxCode <thxcode0824@gmail.com>

Author: thxCode

gpustack_runtime/deployer/docker.py

@@ -317,7 +317,10 @@ def is_supported() -> bool:
             try:
                 supported = client.ping()
             except docker.errors.APIError:
-                if logger.isEnabledFor(logging.DEBUG):
+                if (
+                    logger.isEnabledFor(logging.DEBUG)
+                    and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+                ):
                     logger.exception("Failed to connect to Docker API server")
 
         return supported
@@ -344,7 +347,10 @@ def _get_client() -> docker.DockerClient | None:
                 else:
                     client = docker.from_env()
         except docker.errors.DockerException:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to get Docker client")
 
         return client
@@ -1175,14 +1181,16 @@ def _prepare_create(self):
                 self_container.id,
             )
             self_image = self_container.image
-        except docker.errors.APIError as e:
-            output_log = logger.warning
-            if logger.isEnabledFor(logging.DEBUG):
-                output_log = logger.exception
-            output_log(
-                "Mirrored deployment enabled, but failed to get self Container %s, skipping: %s",
+        except docker.errors.APIError:
+            output_logger = logger.warning
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
+                output_logger = logger.exception
+            output_logger(
+                "Mirrored deployment enabled, but failed to get self Container %s, skipping",
                 self_container_id,
-                e,
             )
             return
 

gpustack_runtime/deployer/kuberentes.py

@@ -315,7 +315,10 @@ def is_supported() -> bool:
                 urllib3.exceptions.MaxRetryError,
                 kubernetes.client.exceptions.ApiException,
             ):
-                if logger.isEnabledFor(logging.DEBUG):
+                if (
+                    logger.isEnabledFor(logging.DEBUG)
+                    and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+                ):
                     logger.exception("Failed to connect to Kubernetes API server")
 
         return supported
@@ -341,7 +344,10 @@ def _get_client() -> kubernetes.client.ApiClient | None:
                 client = kubernetes.client.ApiClient()
                 client.user_agent = "gpustack/runtime"
         except kubernetes.config.config_exception.ConfigException:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to get Kubernetes client")
 
         return client
@@ -1101,10 +1107,15 @@ def _prepare_create(self):
             )
         except kubernetes.client.exceptions.ApiException:
             output_log = logger.warning
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 output_log = logger.exception
             output_log(
-                f"Mirrored deployment enabled, but failed to get self Pod {self_pod_namespace}/{self_pod_name}, skipping",
+                "Mirrored deployment enabled, but failed to get self Pod %s/%s, skipping",
+                self_pod_namespace,
+                self_pod_name,
             )
             return
         ## - Get the first Container, or the Container named "default" if exists.

gpustack_runtime/detector/__init__.py

@@ -96,7 +96,10 @@ def detect_devices(fast: bool = True) -> Devices:
             detect_target = envs.GPUSTACK_RUNTIME_DETECT.lower()
             if detect_target == det.name:
                 raise
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to detect devices for %s", det.name)
 
     return devices

gpustack_runtime/detector/amd.py

@@ -47,7 +47,10 @@ def is_supported() -> bool:
             pyamdsmi.amdsmi_shut_down()
             supported = True
         except pyamdsmi.AmdSmiException:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to initialize AMD SMI")
 
         return supported
@@ -187,11 +190,17 @@ def detect(self) -> Devices | None:
                     ),
                 )
         except pyamdsmi.AmdSmiException:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to fetch devices")
             raise
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
         finally:

gpustack_runtime/detector/ascend.py

@@ -42,7 +42,10 @@ def is_supported() -> bool:
             pydcmi.dcmi_init()
             supported = True
         except pydcmi.DCMIError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to initialize DCMI")
 
         return supported
@@ -188,11 +191,17 @@ def detect(self) -> Devices | None:
                         ),
                     )
         except pydcmi.DCMIError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to fetch devices")
             raise
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
 
@@ -262,7 +271,7 @@ def _get_device_roce_network_info(
             pydcmi.DCMI_PORT_TYPE_ROCE_PORT,
         )
     except pydcmi.DCMIError:
-        if slogger.isEnabledFor(logging.DEBUG):
+        if slogger.isEnabledFor(logging.DEBUG) and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION:
             slogger.exception("Failed to get device roce network info")
 
     return ip, mask, gateway
@@ -289,7 +298,7 @@ def _get_device_virtual_info(
             c_vdev_query_stru,
         )
     except pydcmi.DCMIError:
-        if slogger.isEnabledFor(logging.DEBUG):
+        if slogger.isEnabledFor(logging.DEBUG) and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION:
             slogger.exception("Failed to get device virtual info")
     else:
         return c_vdev_query_stru

gpustack_runtime/detector/cambricon.py

@@ -123,7 +123,10 @@ def detect(self) -> Devices | None:
                 )
 
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
 

gpustack_runtime/detector/hygon.py

@@ -47,7 +47,10 @@ def is_supported() -> bool:
             pyrocmsmi.rsmi_init()
             supported = True
         except pyrocmsmi.ROCMSMIError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to initialize ROCM SMI")
 
         return supported
@@ -154,11 +157,17 @@ def detect(self) -> Devices | None:
                     ),
                 )
         except pyrocmsmi.ROCMSMIError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to fetch devices")
             raise
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
 

gpustack_runtime/detector/iluvatar.py

@@ -165,7 +165,10 @@ def detect(self) -> Devices | None:
                 )
 
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
 

gpustack_runtime/detector/metax.py

@@ -46,7 +46,10 @@ def is_supported() -> bool:
             pymxsml.mxSmlInit()
             supported = True
         except pymxsml.MXSMLError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to initialize MXSML")
 
         return supported
@@ -159,11 +162,17 @@ def detect(self) -> Devices | None:
                 )
 
         except pymxsml.MXSMLError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to fetch devices")
             raise
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
 

gpustack_runtime/detector/mthreads.py

@@ -41,7 +41,10 @@ def is_supported() -> bool:
             pymtml.mtmlLibraryShutDown()
             supported = True
         except pymtml.MTMLError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to initialize MTML")
 
         return supported
@@ -152,11 +155,17 @@ def detect(self) -> Devices | None:
                 )
 
         except pymtml.MTMLError:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to fetch devices")
             raise
         except Exception:
-            if logger.isEnabledFor(logging.DEBUG):
+            if (
+                logger.isEnabledFor(logging.DEBUG)
+                and envs.GPUSTACK_RUNTIME_LOG_EXCEPTION
+            ):
                 logger.exception("Failed to process devices fetching")
             raise
         finally: