3 """Helpers for unittests. Note that when you import this we
4 automatically wrap unittest.main() with a call to
5 bootstrap.initialize so that we getLogger config, commandline args,
6 logging control, etc... this works fine but it's a little hacky so
10 from abc import ABC, abstractmethod
21 from typing import Callable, Dict, List
30 import sqlalchemy as sa
33 logger = logging.getLogger(__name__)
34 cfg = config.add_commandline_args(
35 f'Logging ({__file__})', 'Args related to function decorators'
38 '--unittests_ignore_perf',
41 help='Ignore unittest perf regression in @check_method_for_perf_regressions',
44 '--unittests_num_perf_samples',
47 help='The count of perf timing samples we need to see before blocking slow runs on perf grounds',
50 '--unittests_drop_perf_traces',
54 help='The identifier (i.e. file!test_fixture) for which we should drop all perf data',
57 '--unittests_persistance_strategy',
58 choices=['FILE', 'DATABASE'],
60 help='Should we persist perf data in a file or db?',
63 '--unittests_perfdb_filename',
66 default=f'{os.environ["HOME"]}/.python_unittest_performance_db',
67 help='File in which to store perf data (iff --unittests_persistance_strategy is FILE)',
70 '--unittests_perfdb_spec',
73 default='mariadb+pymysql://python_unittest:<PASSWORD>@db.house:3306/python_unittest_performance',
74 help='Db connection spec for perf data (iff --unittest_persistance_strategy is DATABASE)',
77 # >>> This is the hacky business, FYI. <<<
78 unittest.main = bootstrap.initialize(unittest.main)
81 class PerfRegressionDataPersister(ABC):
86 def load_performance_data(self) -> Dict[str, List[float]]:
90 def save_performance_data(self, method_id: str, data: Dict[str, List[float]]):
94 def delete_performance_data(self, method_id: str):
98 class FileBasedPerfRegressionDataPersister(PerfRegressionDataPersister):
99 def __init__(self, filename: str):
100 self.filename = filename
101 self.traces_to_delete = []
103 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
104 with open(self.filename, 'rb') as f:
105 return pickle.load(f)
107 def save_performance_data(self, method_id: str, data: Dict[str, List[float]]):
108 for trace in self.traces_to_delete:
112 with open(self.filename, 'wb') as f:
113 pickle.dump(data, f, pickle.HIGHEST_PROTOCOL)
115 def delete_performance_data(self, method_id: str):
116 self.traces_to_delete.append(method_id)
119 class DatabasePerfRegressionDataPersister(PerfRegressionDataPersister):
120 def __init__(self, dbspec: str):
122 self.engine = sa.create_engine(self.dbspec)
123 self.conn = self.engine.connect()
125 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
126 results = self.conn.execute(
128 f'SELECT * FROM runtimes_by_function WHERE function = "{method_id}";'
131 ret = {method_id: []}
132 for result in results.all():
133 ret[method_id].append(result['runtime'])
137 def save_performance_data(self, method_id: str, data: Dict[str, List[float]]):
138 self.delete_performance_data(method_id)
139 for (method_id, perf_data) in data.items():
140 sql = 'INSERT INTO runtimes_by_function (function, runtime) VALUES '
141 for perf in perf_data:
142 self.conn.execute(sql + f'("{method_id}", {perf});')
144 def delete_performance_data(self, method_id: str):
145 sql = f'DELETE FROM runtimes_by_function WHERE function = "{method_id}"'
146 self.conn.execute(sql)
149 def check_method_for_perf_regressions(func: Callable) -> Callable:
151 This is meant to be used on a method in a class that subclasses
152 unittest.TestCase. When thus decorated it will time the execution
153 of the code in the method, compare it with a database of
154 historical perfmance, and fail the test with a perf-related
155 message if it has become too slow.
159 @functools.wraps(func)
160 def wrapper_perf_monitor(*args, **kwargs):
161 if config.config['unittests_persistance_strategy'] == 'FILE':
162 filename = config.config['unittests_perfdb_filename']
163 helper = FileBasedPerfRegressionDataPersister(filename)
164 elif config.config['unittests_persistance_strategy'] == 'DATABASE':
165 dbspec = config.config['unittests_perfdb_spec']
166 dbspec = dbspec.replace(
167 '<PASSWORD>', scott_secrets.MARIADB_UNITTEST_PERF_PASSWORD
169 helper = DatabasePerfRegressionDataPersister(dbspec)
171 raise Exception('Unknown/unexpected --unittests_persistance_strategy value')
173 func_id = function_utils.function_identifier(func)
174 func_name = func.__name__
175 logger.debug(f'Watching {func_name}\'s performance...')
176 logger.debug(f'Canonical function identifier = {func_id}')
179 perfdb = helper.load_performance_data(func_id)
180 except Exception as e:
182 msg = 'Unable to load perfdb; skipping it...'
187 # cmdline arg to forget perf traces for function
188 drop_id = config.config['unittests_drop_perf_traces']
189 if drop_id is not None:
190 helper.delete_performance_data(drop_id)
192 # Run the wrapped test paying attention to latency.
193 start_time = time.perf_counter()
194 value = func(*args, **kwargs)
195 end_time = time.perf_counter()
196 run_time = end_time - start_time
198 # See if it was unexpectedly slow.
199 hist = perfdb.get(func_id, [])
200 if len(hist) < config.config['unittests_num_perf_samples']:
201 hist.append(run_time)
202 logger.debug(f'Still establishing a perf baseline for {func_name}')
204 stdev = statistics.stdev(hist)
205 logger.debug(f'For {func_name}, performance stdev={stdev}')
207 logger.debug(f'For {func_name}, slowest perf on record is {slowest:f}s')
208 limit = slowest + stdev * 4
209 logger.debug(f'For {func_name}, max acceptable runtime is {limit:f}s')
210 logger.debug(f'For {func_name}, actual observed runtime was {run_time:f}s')
211 if run_time > limit and not config.config['unittests_ignore_perf']:
212 msg = f'''{func_id} performance has regressed unacceptably.
213 {slowest:f}s is the slowest runtime on record in {len(hist)} perf samples.
214 It just ran in {run_time:f}s which is 4+ stdevs slower than the slowest.
215 Here is the current, full db perf timing distribution:
221 slf = args[0] # Peek at the wrapped function's self ref.
222 slf.fail(msg) # ...to fail the testcase.
224 hist.append(run_time)
226 # Don't spam the database with samples; just pick a random
227 # sample from what we have and store that back.
228 n = min(config.config['unittests_num_perf_samples'], len(hist))
229 hist = random.sample(hist, n)
231 perfdb[func_id] = hist
232 helper.save_performance_data(func_id, perfdb)
235 return wrapper_perf_monitor
238 def check_all_methods_for_perf_regressions(prefix='test_'):
239 """Decorate unittests with this to pay attention to the perf of the
240 testcode and flag perf regressions. e.g.
242 import unittest_utils as uu
244 @uu.check_all_methods_for_perf_regressions()
245 class TestMyClass(unittest.TestCase):
247 def test_some_part_of_my_class(self):
252 def decorate_the_testcase(cls):
253 if issubclass(cls, unittest.TestCase):
254 for name, m in inspect.getmembers(cls, inspect.isfunction):
255 if name.startswith(prefix):
256 setattr(cls, name, check_method_for_perf_regressions(m))
257 logger.debug(f'Wrapping {cls.__name__}:{name}.')
260 return decorate_the_testcase
264 """Hard code a breakpoint somewhere; drop into pdb."""
270 class RecordStdout(object):
272 Record what is emitted to stdout.
274 >>> with RecordStdout() as record:
275 ... print("This is a test!")
276 >>> print({record().readline()})
277 {'This is a test!\\n'}
281 def __init__(self) -> None:
282 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
285 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
286 self.recorder = contextlib.redirect_stdout(self.destination)
287 self.recorder.__enter__()
288 return lambda: self.destination
290 def __exit__(self, *args) -> bool:
291 self.recorder.__exit__(*args)
292 self.destination.seek(0)
296 class RecordStderr(object):
298 Record what is emitted to stderr.
301 >>> with RecordStderr() as record:
302 ... print("This is a test!", file=sys.stderr)
303 >>> print({record().readline()})
304 {'This is a test!\\n'}
308 def __init__(self) -> None:
309 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
312 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
313 self.recorder = contextlib.redirect_stderr(self.destination)
314 self.recorder.__enter__()
315 return lambda: self.destination
317 def __exit__(self, *args) -> bool:
318 self.recorder.__exit__(*args)
319 self.destination.seek(0)
323 class RecordMultipleStreams(object):
325 Record the output to more than one stream.
328 def __init__(self, *files) -> None:
329 self.files = [*files]
330 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
331 self.saved_writes = []
333 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
335 self.saved_writes.append(f.write)
336 f.write = self.destination.write
337 return lambda: self.destination
339 def __exit__(self, *args) -> bool:
341 f.write = self.saved_writes.pop()
342 self.destination.seek(0)
345 if __name__ == '__main__':