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
22 from abc import ABC, abstractmethod
23 from typing import Any, Callable, Dict, List, Optional
25 import sqlalchemy as sa
32 logger = logging.getLogger(__name__)
33 cfg = config.add_commandline_args(f'Logging ({__file__})', 'Args related to function decorators')
35 '--unittests_ignore_perf',
38 help='Ignore unittest perf regression in @check_method_for_perf_regressions',
41 '--unittests_num_perf_samples',
44 help='The count of perf timing samples we need to see before blocking slow runs on perf grounds',
47 '--unittests_drop_perf_traces',
51 help='The identifier (i.e. file!test_fixture) for which we should drop all perf data',
54 '--unittests_persistance_strategy',
55 choices=['FILE', 'DATABASE'],
57 help='Should we persist perf data in a file or db?',
60 '--unittests_perfdb_filename',
63 default=f'{os.environ["HOME"]}/.python_unittest_performance_db',
64 help='File in which to store perf data (iff --unittests_persistance_strategy is FILE)',
67 '--unittests_perfdb_spec',
70 default='mariadb+pymysql://python_unittest:<PASSWORD>@db.house:3306/python_unittest_performance',
71 help='Db connection spec for perf data (iff --unittest_persistance_strategy is DATABASE)',
74 # >>> This is the hacky business, FYI. <<<
75 unittest.main = bootstrap.initialize(unittest.main)
78 class PerfRegressionDataPersister(ABC):
79 """A base class for a signature dealing with persisting perf
86 def load_performance_data(self, method_id: str) -> 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 """A perf regression data persister that uses files."""
101 def __init__(self, filename: str):
103 self.filename = filename
104 self.traces_to_delete: List[str] = []
106 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
107 with open(self.filename, 'rb') as f:
108 return pickle.load(f)
110 def save_performance_data(self, method_id: str, data: Dict[str, List[float]]):
111 for trace in self.traces_to_delete:
115 with open(self.filename, 'wb') as f:
116 pickle.dump(data, f, pickle.HIGHEST_PROTOCOL)
118 def delete_performance_data(self, method_id: str):
119 self.traces_to_delete.append(method_id)
122 class DatabasePerfRegressionDataPersister(PerfRegressionDataPersister):
123 """A perf regression data persister that uses a database backend."""
125 def __init__(self, dbspec: str):
128 self.engine = sa.create_engine(self.dbspec)
129 self.conn = self.engine.connect()
131 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
132 results = self.conn.execute(
133 sa.text(f'SELECT * FROM runtimes_by_function WHERE function = "{method_id}";')
135 ret: Dict[str, List[float]] = {method_id: []}
136 for result in results.all():
137 ret[method_id].append(result['runtime'])
141 def save_performance_data(self, method_id: str, data: Dict[str, List[float]]):
142 self.delete_performance_data(method_id)
143 for (mid, perf_data) in data.items():
144 sql = 'INSERT INTO runtimes_by_function (function, runtime) VALUES '
145 for perf in perf_data:
146 self.conn.execute(sql + f'("{mid}", {perf});')
148 def delete_performance_data(self, method_id: str):
149 sql = f'DELETE FROM runtimes_by_function WHERE function = "{method_id}"'
150 self.conn.execute(sql)
153 def check_method_for_perf_regressions(func: Callable) -> Callable:
155 This is meant to be used on a method in a class that subclasses
156 unittest.TestCase. When thus decorated it will time the execution
157 of the code in the method, compare it with a database of
158 historical perfmance, and fail the test with a perf-related
159 message if it has become too slow.
163 @functools.wraps(func)
164 def wrapper_perf_monitor(*args, **kwargs):
165 if config.config['unittests_ignore_perf']:
166 return func(*args, **kwargs)
168 if config.config['unittests_persistance_strategy'] == 'FILE':
169 filename = config.config['unittests_perfdb_filename']
170 helper = FileBasedPerfRegressionDataPersister(filename)
171 elif config.config['unittests_persistance_strategy'] == 'DATABASE':
172 dbspec = config.config['unittests_perfdb_spec']
173 dbspec = dbspec.replace('<PASSWORD>', scott_secrets.MARIADB_UNITTEST_PERF_PASSWORD)
174 helper = DatabasePerfRegressionDataPersister(dbspec)
176 raise Exception('Unknown/unexpected --unittests_persistance_strategy value')
178 func_id = function_utils.function_identifier(func)
179 func_name = func.__name__
180 logger.debug('Watching %s\'s performance...', func_name)
181 logger.debug('Canonical function identifier = "%s"', func_id)
184 perfdb = helper.load_performance_data(func_id)
185 except Exception as e:
187 msg = 'Unable to load perfdb; skipping it...'
192 # cmdline arg to forget perf traces for function
193 drop_id = config.config['unittests_drop_perf_traces']
194 if drop_id is not None:
195 helper.delete_performance_data(drop_id)
197 # Run the wrapped test paying attention to latency.
198 start_time = time.perf_counter()
199 value = func(*args, **kwargs)
200 end_time = time.perf_counter()
201 run_time = end_time - start_time
203 # See if it was unexpectedly slow.
204 hist = perfdb.get(func_id, [])
205 if len(hist) < config.config['unittests_num_perf_samples']:
206 hist.append(run_time)
207 logger.debug('Still establishing a perf baseline for %s', func_name)
209 stdev = statistics.stdev(hist)
210 logger.debug('For %s, performance stdev=%.2f', func_name, stdev)
212 logger.debug('For %s, slowest perf on record is %.2fs', func_name, slowest)
213 limit = slowest + stdev * 4
214 logger.debug('For %s, max acceptable runtime is %.2fs', func_name, limit)
215 logger.debug('For %s, actual observed runtime was %.2fs', func_name, run_time)
217 msg = f'''{func_id} performance has regressed unacceptably.
218 {slowest:f}s is the slowest runtime on record in {len(hist)} perf samples.
219 It just ran in {run_time:f}s which is 4+ stdevs slower than the slowest.
220 Here is the current, full db perf timing distribution:
226 slf = args[0] # Peek at the wrapped function's self ref.
227 slf.fail(msg) # ...to fail the testcase.
229 hist.append(run_time)
231 # Don't spam the database with samples; just pick a random
232 # sample from what we have and store that back.
233 n = min(config.config['unittests_num_perf_samples'], len(hist))
234 hist = random.sample(hist, n)
236 perfdb[func_id] = hist
237 helper.save_performance_data(func_id, perfdb)
240 return wrapper_perf_monitor
243 def check_all_methods_for_perf_regressions(prefix='test_'):
244 """Decorate unittests with this to pay attention to the perf of the
245 testcode and flag perf regressions. e.g.
247 import unittest_utils as uu
249 @uu.check_all_methods_for_perf_regressions()
250 class TestMyClass(unittest.TestCase):
252 def test_some_part_of_my_class(self):
257 def decorate_the_testcase(cls):
258 if issubclass(cls, unittest.TestCase):
259 for name, m in inspect.getmembers(cls, inspect.isfunction):
260 if name.startswith(prefix):
261 setattr(cls, name, check_method_for_perf_regressions(m))
262 logger.debug('Wrapping %s:%s.', cls.__name__, name)
265 return decorate_the_testcase
269 """Hard code a breakpoint somewhere; drop into pdb."""
275 class RecordStdout(object):
277 Record what is emitted to stdout.
279 >>> with RecordStdout() as record:
280 ... print("This is a test!")
281 >>> print({record().readline()})
282 {'This is a test!\\n'}
286 def __init__(self) -> None:
287 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
288 self.recorder: Optional[contextlib.redirect_stdout] = None
290 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
291 self.recorder = contextlib.redirect_stdout(self.destination)
292 assert self.recorder is not None
293 self.recorder.__enter__()
294 return lambda: self.destination
296 def __exit__(self, *args) -> Optional[bool]:
297 assert self.recorder is not None
298 self.recorder.__exit__(*args)
299 self.destination.seek(0)
303 class RecordStderr(object):
305 Record what is emitted to stderr.
308 >>> with RecordStderr() as record:
309 ... print("This is a test!", file=sys.stderr)
310 >>> print({record().readline()})
311 {'This is a test!\\n'}
315 def __init__(self) -> None:
316 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
317 self.recorder: Optional[contextlib.redirect_stdout[Any]] = None
319 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
320 self.recorder = contextlib.redirect_stderr(self.destination) # type: ignore
321 assert self.recorder is not None
322 self.recorder.__enter__()
323 return lambda: self.destination
325 def __exit__(self, *args) -> Optional[bool]:
326 assert self.recorder is not None
327 self.recorder.__exit__(*args)
328 self.destination.seek(0)
332 class RecordMultipleStreams(object):
334 Record the output to more than one stream.
337 def __init__(self, *files) -> None:
338 self.files = [*files]
339 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
340 self.saved_writes: List[Callable[..., Any]] = []
342 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
344 self.saved_writes.append(f.write)
345 f.write = self.destination.write
346 return lambda: self.destination
348 def __exit__(self, *args) -> Optional[bool]:
350 f.write = self.saved_writes.pop()
351 self.destination.seek(0)
355 if __name__ == '__main__':