@@ -21,34 +21,37 @@ def teardown_bridge():
2121 unregister_cuda_error_log ()
2222
2323
24- # 使用方式1: register(logger) 后用 caplog 捕获 driver log
24+ # Usage 1: show driver error log interactively
2525def test_cuda_driver_log_captured_via_caplog (caplog ):
2626 driver .cuInit (0 )
2727
2828 logger_name = "cuda.driver.caplog_test"
2929 logger_obj = logging .getLogger (logger_name )
3030 register_cuda_error_log (logger_obj )
3131
32- # cuDeviceGet(9999) 越界,驱动写入两条 ERROR 日志
32+ # cuDeviceGet(9999) is out of range; driver writes two ERROR logs
3333 with caplog .at_level (logging .ERROR , logger = logger_name ):
3434 driver .cuDeviceGet (9999 )
3535
3636 messages = [r .message for r in caplog .records if r .name == logger_name ]
3737
38- assert any ( "Parameter ordinal must be between 0 and 1" in m for m in messages ), (
39- f"Expected ordinal error in caplog, got: \n { messages } "
40- )
38+ assert len ( messages ) == 2 , f"Expected exactly 2 driver log messages, got: \n { messages } "
39+ assert "Parameter ordinal must be between 0 and 1" in messages [ 0 ]
40+ assert "CUDA_ERROR_INVALID_DEVICE" in messages [ 1 ] and "cuDeviceGet" in messages [ 1 ]
4141
42- assert any ("CUDA_ERROR_INVALID_DEVICE" in m and "cuDeviceGet" in m for m in messages ), (
43- f"Expected cuDeviceGet return error in caplog, got:\n { messages } "
44- )
42+ # After unregistering, driver logs are no longer forwarded
43+ unregister_cuda_error_log ()
44+ caplog .clear ()
45+ with caplog .at_level (logging .ERROR , logger = logger_name ):
46+ driver .cuDeviceGet (9999 )
47+ assert not any (r .name == logger_name for r in caplog .records )
48+ assert "Parameter ordinal must be between 0 and 1" not in caplog .text
4549
4650
47- # 使用方式2: 通过 FileHandler 把 driver log 写到文件
51+ # Usage 2: redirect driver log to a file
4852def test_cuda_driver_log_written_to_file (tmp_path ):
4953 driver .cuInit (0 )
5054
51- # 独立 logger name,避免污染 root logger / 影响其他用例
5255 log_file = tmp_path / "cuda_driver.log"
5356 logger_name = "cuda.driver.file_test"
5457 logger_obj = logging .getLogger (logger_name )
@@ -63,22 +66,16 @@ def test_cuda_driver_log_written_to_file(tmp_path):
6366 register_cuda_error_log (logger_obj )
6467 driver .cuDeviceGet (9999 )
6568 finally :
66- # close() flush 落盘;removeHandler 防止跨用例累加 handler
6769 logger_obj .removeHandler (file_handler )
6870 file_handler .close ()
6971
7072 content = log_file .read_text (encoding = "utf-8" )
7173
72- assert "Parameter ordinal must be between 0 and 1" in content , (
73- f"Expected ordinal error in log file, got:\n { content } "
74- )
75-
76- assert "CUDA_ERROR_INVALID_DEVICE" in content and "cuDeviceGet" in content , (
77- f"Expected cuDeviceGet return error in log file, got:\n { content } "
78- )
74+ assert "Parameter ordinal must be between 0 and 1" in content
75+ assert "CUDA_ERROR_INVALID_DEVICE" in content and "cuDeviceGet" in content
7976
8077
81- # 使用方式3 : register -> unregister -> 换绑另一个 logger
78+ # Usage 3 : register -> unregister -> rebind to another logger
8279def test_cuda_driver_log_rebind_logger (caplog ):
8380 driver .cuInit (0 )
8481
@@ -89,7 +86,7 @@ def test_cuda_driver_log_rebind_logger(caplog):
8986 driver .cuDeviceGet (9999 )
9087 assert any (r .name == name_a for r in caplog .records )
9188
92- # 必须先 unregister 释放驱动侧 slot,再绑新 logger
89+ # Must unregister first to free the driver-side slot, then bind the new logger
9390 unregister_cuda_error_log ()
9491 caplog .clear ()
9592 register_cuda_error_log (logging .getLogger (name_b ))
@@ -100,25 +97,25 @@ def test_cuda_driver_log_rebind_logger(caplog):
10097 assert any (r .name == name_b for r in caplog .records )
10198
10299
103- # 使用方式4: logger 的 level 阈值对驱动日志同样生效
100+ # Usage 4: the logger's level threshold applies to driver logs too
104101def test_cuda_driver_log_level_filtering (caplog ):
105102 driver .cuInit (0 )
106103
107104 logger_name = "cuda.driver.level_test"
108105 register_cuda_error_log (logging .getLogger (logger_name ))
109106
110- # CRITICAL 阈值过滤掉 ERROR 级别的驱动日志
107+ # CRITICAL threshold filters out ERROR-level driver logs
111108 with caplog .at_level (logging .CRITICAL , logger = logger_name ):
112109 driver .cuDeviceGet (9999 )
113110 assert not any (r .name == logger_name for r in caplog .records )
114111
115- # 降到 ERROR 阈值就能捕获到
112+ # Lowering to ERROR threshold lets them through
116113 with caplog .at_level (logging .ERROR , logger = logger_name ):
117114 driver .cuDeviceGet (9999 )
118115 assert any (r .name == logger_name for r in caplog .records )
119116
120117
121- # 使用方式5: 自定义 Handler 统计错误次数,接入监控/告警
118+ # Usage 5: custom Handler counts errors, for monitoring/alerting integration
122119def test_cuda_driver_log_alert_counter ():
123120 driver .cuInit (0 )
124121
@@ -140,58 +137,40 @@ def emit(self, record):
140137 finally :
141138 logger_obj .removeHandler (counter_handler )
142139
143- assert counter_handler .count >= 1 , "Expected at least one ERROR-level driver log to be counted"
140+ assert counter_handler .count == 2 , "Expected exactly 2 ERROR-level driver logs to be counted"
144141
145142
146- # 使用方式6: 用 callback 把 driver log 接进已有的 logger
143+ # Usage 6: use a custom python callback function to change log level and log format
147144def test_cuda_driver_log_custom_callback_overrides_default (caplog ):
148145 driver .cuInit (0 )
149146
150147 logger_name = "my_library.cuda-driver"
151148 logger_obj = logging .getLogger (logger_name )
152149
153- # 按 raw CUlogLevel 分发;ERROR 记成 DEBUG
150+ # Log everything as DEBUG, prefixed with the logger name instead of "[CUDA Driver] "
154151 def my_callback (cu_log_level , message ):
155152 msg = message .decode ("utf-8" , errors = "replace" ).rstrip ("\n " )
156- if cu_log_level == driver .CUlogLevel .CU_LOG_LEVEL_WARNING :
157- logger_obj .warning ("%s: %s" , logger_obj .name , msg )
158- elif cu_log_level == driver .CUlogLevel .CU_LOG_LEVEL_ERROR :
159- logger_obj .debug ("%s: %s" , logger_obj .name , msg )
153+ logger_obj .debug ("%s: %s" , logger_obj .name , msg )
160154
161155 register_cuda_error_log (logger_obj , callback = my_callback )
162156
163- # 默认 WARNING 阈值下, DEBUG 级别的驱动日志被过滤掉
157+ # Default WARNING threshold filters out the DEBUG-level driver logs
164158 with caplog .at_level (logging .WARNING , logger = logger_name ):
165159 driver .cuDeviceGet (9999 )
166- assert not any (r .name == logger_name for r in caplog .records ), (
167- "DEBUG-level driver log should be filtered out at the default WARNING level"
168- )
160+ assert not caplog .records
169161
170162 caplog .clear ()
171163
172- # 调到 DEBUG 阈值才能收到
164+ # Lowering the threshold to DEBUG lets them through, with the custom level/format
173165 with caplog .at_level (logging .DEBUG , logger = logger_name ):
174166 driver .cuDeviceGet (9999 )
175167
176- records = [r for r in caplog .records if r .name == logger_name ]
177- assert records , "Expected driver log records forwarded via my_callback once the threshold is DEBUG"
178-
179- # my_callback 生效:级别统一改成了 DEBUG
180- assert all (r .levelno == logging .DEBUG for r in records ), (
181- f"Expected every record logged as DEBUG via my_callback, got levels: { [r .levelname for r in records ]} "
182- )
183-
184- # 消息前缀是 logger name,不是默认的 "[CUDA Driver] "
185- messages = [r .message for r in records ]
186- assert any (m .startswith (f"{ logger_name } : " ) for m in messages ), (
187- f"Expected messages prefixed with '{ logger_name } : ', got:\n { messages } "
188- )
189- assert not any (m .startswith ("[CUDA Driver] " ) for m in messages ), (
190- f"Default '[CUDA Driver] ' forwarding should be overridden by my_callback, got:\n { messages } "
191- )
168+ assert len (caplog .records ) == 2
169+ assert all (r .levelno == logging .DEBUG for r in caplog .records )
170+ assert all (r .message .startswith (f"{ logger_name } : " ) for r in caplog .records )
192171
193172
194- # callback 必须可调用或 None,否则应尽早报错
173+ # callback must be callable or None, else fail fast
195174def test_cuda_driver_log_callback_must_be_callable_or_none ():
196175 with pytest .raises (TypeError ):
197176 register_cuda_error_log (logging .getLogger ("cuda.driver.bad_callback_test" ), callback = "not-callable" )
0 commit comments