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(
91 self, method_id: str, data: Dict[str, List[float]]
96 def delete_performance_data(self, method_id: str):
100 class FileBasedPerfRegressionDataPersister(PerfRegressionDataPersister):
101 def __init__(self, filename: str):
102 self.filename = filename
103 self.traces_to_delete = []
105 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
106 with open(self.filename, 'rb') as f:
107 return pickle.load(f)
109 def save_performance_data(
110 self, method_id: str, data: Dict[str, List[float]]
112 for trace in self.traces_to_delete:
116 with open(self.filename, 'wb') as f:
117 pickle.dump(data, f, pickle.HIGHEST_PROTOCOL)
119 def delete_performance_data(self, method_id: str):
120 self.traces_to_delete.append(method_id)
123 class DatabasePerfRegressionDataPersister(PerfRegressionDataPersister):
124 def __init__(self, dbspec: str):
126 self.engine = sa.create_engine(self.dbspec)
127 self.conn = self.engine.connect()
129 def load_performance_data(self, method_id: str) -> Dict[str, List[float]]:
130 results = self.conn.execute(
132 f'SELECT * FROM runtimes_by_function WHERE function = "{method_id}";'
135 ret = {method_id: []}
136 for result in results.all():
137 ret[method_id].append(result['runtime'])
141 def save_performance_data(
142 self, method_id: str, data: Dict[str, List[float]]
144 self.delete_performance_data(method_id)
145 for (method_id, perf_data) in data.items():
146 sql = 'INSERT INTO runtimes_by_function (function, runtime) VALUES '
147 for perf in perf_data:
148 self.conn.execute(sql + f'("{method_id}", {perf});')
150 def delete_performance_data(self, method_id: str):
151 sql = f'DELETE FROM runtimes_by_function WHERE function = "{method_id}"'
152 self.conn.execute(sql)
155 def check_method_for_perf_regressions(func: Callable) -> Callable:
157 This is meant to be used on a method in a class that subclasses
158 unittest.TestCase. When thus decorated it will time the execution
159 of the code in the method, compare it with a database of
160 historical perfmance, and fail the test with a perf-related
161 message if it has become too slow.
165 @functools.wraps(func)
166 def wrapper_perf_monitor(*args, **kwargs):
167 if config.config['unittests_persistance_strategy'] == 'FILE':
168 filename = config.config['unittests_perfdb_filename']
169 helper = FileBasedPerfRegressionDataPersister(filename)
170 elif config.config['unittests_persistance_strategy'] == 'DATABASE':
171 dbspec = config.config['unittests_perfdb_spec']
172 dbspec = dbspec.replace(
173 '<PASSWORD>', scott_secrets.MARIADB_UNITTEST_PERF_PASSWORD
175 helper = DatabasePerfRegressionDataPersister(dbspec)
178 'Unknown/unexpected --unittests_persistance_strategy value'
181 func_id = function_utils.function_identifier(func)
182 func_name = func.__name__
183 logger.debug(f'Watching {func_name}\'s performance...')
184 logger.debug(f'Canonical function identifier = {func_id}')
187 perfdb = helper.load_performance_data(func_id)
188 except Exception as e:
190 msg = 'Unable to load perfdb; skipping it...'
195 # cmdline arg to forget perf traces for function
196 drop_id = config.config['unittests_drop_perf_traces']
197 if drop_id is not None:
198 helper.delete_performance_data(drop_id)
200 # Run the wrapped test paying attention to latency.
201 start_time = time.perf_counter()
202 value = func(*args, **kwargs)
203 end_time = time.perf_counter()
204 run_time = end_time - start_time
206 # See if it was unexpectedly slow.
207 hist = perfdb.get(func_id, [])
208 if len(hist) < config.config['unittests_num_perf_samples']:
209 hist.append(run_time)
210 logger.debug(f'Still establishing a perf baseline for {func_name}')
212 stdev = statistics.stdev(hist)
213 logger.debug(f'For {func_name}, performance stdev={stdev}')
216 f'For {func_name}, slowest perf on record is {slowest:f}s'
218 limit = slowest + stdev * 4
220 f'For {func_name}, max acceptable runtime is {limit:f}s'
223 f'For {func_name}, actual observed runtime was {run_time:f}s'
225 if run_time > limit and not config.config['unittests_ignore_perf']:
226 msg = f'''{func_id} performance has regressed unacceptably.
227 {slowest:f}s is the slowest runtime on record in {len(hist)} perf samples.
228 It just ran in {run_time:f}s which is 4+ stdevs slower than the slowest.
229 Here is the current, full db perf timing distribution:
235 slf = args[0] # Peek at the wrapped function's self ref.
236 slf.fail(msg) # ...to fail the testcase.
238 hist.append(run_time)
240 # Don't spam the database with samples; just pick a random
241 # sample from what we have and store that back.
242 n = min(config.config['unittests_num_perf_samples'], len(hist))
243 hist = random.sample(hist, n)
245 perfdb[func_id] = hist
246 helper.save_performance_data(func_id, perfdb)
249 return wrapper_perf_monitor
252 def check_all_methods_for_perf_regressions(prefix='test_'):
253 """Decorate unittests with this to pay attention to the perf of the
254 testcode and flag perf regressions. e.g.
256 import unittest_utils as uu
258 @uu.check_all_methods_for_perf_regressions()
259 class TestMyClass(unittest.TestCase):
261 def test_some_part_of_my_class(self):
266 def decorate_the_testcase(cls):
267 if issubclass(cls, unittest.TestCase):
268 for name, m in inspect.getmembers(cls, inspect.isfunction):
269 if name.startswith(prefix):
270 setattr(cls, name, check_method_for_perf_regressions(m))
271 logger.debug(f'Wrapping {cls.__name__}:{name}.')
274 return decorate_the_testcase
278 """Hard code a breakpoint somewhere; drop into pdb."""
284 class RecordStdout(object):
286 Record what is emitted to stdout.
288 >>> with RecordStdout() as record:
289 ... print("This is a test!")
290 >>> print({record().readline()})
291 {'This is a test!\\n'}
294 def __init__(self) -> None:
295 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
298 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
299 self.recorder = contextlib.redirect_stdout(self.destination)
300 self.recorder.__enter__()
301 return lambda: self.destination
303 def __exit__(self, *args) -> bool:
304 self.recorder.__exit__(*args)
305 self.destination.seek(0)
309 class RecordStderr(object):
311 Record what is emitted to stderr.
314 >>> with RecordStderr() as record:
315 ... print("This is a test!", file=sys.stderr)
316 >>> print({record().readline()})
317 {'This is a test!\\n'}
320 def __init__(self) -> None:
321 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
324 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
325 self.recorder = contextlib.redirect_stderr(self.destination)
326 self.recorder.__enter__()
327 return lambda: self.destination
329 def __exit__(self, *args) -> bool:
330 self.recorder.__exit__(*args)
331 self.destination.seek(0)
335 class RecordMultipleStreams(object):
337 Record the output to more than one stream.
340 def __init__(self, *files) -> None:
341 self.files = [*files]
342 self.destination = tempfile.SpooledTemporaryFile(mode='r+')
343 self.saved_writes = []
345 def __enter__(self) -> Callable[[], tempfile.SpooledTemporaryFile]:
347 self.saved_writes.append(f.write)
348 f.write = self.destination.write
349 return lambda: self.destination
351 def __exit__(self, *args) -> bool:
353 f.write = self.saved_writes.pop()
354 self.destination.seek(0)
357 if __name__ == '__main__':