class TestCPULimitReachedDuringProfiling:
@before
def before(self):
self.timer = Timer()
self.profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
environment_override={
"timer": self.timer,
"cpu_limit_percentage": 40,
"sampling_interval": timedelta(seconds=0.01),
'agent_metadata':
AgentMetadata(fleet_info=DefaultFleetInfo())
},
)
yield
self.profiler.stop()
def test_profiler_terminates(self):
self.profiler.start()
assert self.profiler.is_running()
# With sampling_interval to be 0.01 seconds, having runProfiler as 0.5 seconds should breach
# the cpu limit. We need to sample 20 times before we check the CPU limit
for i in range(20):
self.timer.record('sampleAndAggregate', 0.5)
assert wait_for(lambda: not self.profiler.is_running(),
timeout_seconds=5)
class TestWhenAnotherInstanceAlreadyStarted:
@pytest.fixture(autouse=True)
def around(self):
self.profiler_runner = Mock(spec_set=ProfilerRunner)
self.profiler_runner.start = Mock(return_value=True)
self.profiler_runner.is_running = Mock(return_value=False)
self.first_profiler = Profiler(
profiling_group_name="pg_name",
environment_override={
"profiler_runner_factory":
mock_profiler_runner_factory(self.profiler_runner)
})
self.first_profiler.start()
self.second_profiler = Profiler(
profiling_group_name="test-application",
environment_override={
"profiler_runner_factory":
mock_profiler_runner_factory(self.profiler_runner)
})
yield
self.first_profiler.stop()
self.second_profiler.stop()
def test_it_fails_to_start_a_second_profiler(self):
assert (not self.second_profiler.start())
class TestKillSwitchActivatesDuringExecution:
@before
def before(self):
temporary_directory = tempfile.mkdtemp()
self.temp_filepath = str(
Path(temporary_directory,
'test_profiler_stops_after_killswitch_was_detected'))
self.profiler = Profiler(
profiling_group_name="test-application",
environment_override={
"cpu_limit_percentage": None,
"killswitch_filepath": self.temp_filepath,
"sampling_interval": timedelta(seconds=0.1),
'agent_metadata':
AgentMetadata(fleet_info=DefaultFleetInfo())
})
yield
shutil.rmtree(temporary_directory)
self.profiler.stop()
def test_profiler_stops_after_killswitch_was_detected(self):
self.profiler.start()
assert self.profiler.is_running() is True
Path(self.temp_filepath).touch()
# Force the killswitch check happens immediately
self.profiler._profiler_runner_instance.profiler_disabler.killswitch.last_check_for_file_time = None
assert (wait_for(lambda: not self.profiler.is_running(),
timeout_seconds=5))
def test_profiler_does_not_start(self):
with tempfile.NamedTemporaryFile() as killswitch_file:
profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
environment_override={
"killswitch_filepath": killswitch_file.name
})
try:
assert profiler.start() is False
finally:
profiler.stop()
class TestEndToEndProfiling:
@before
def before(self):
self.profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
environment_override={
'initial_sampling_interval': timedelta(),
'agent_metadata': AgentMetadata(fleet_info=DefaultFleetInfo())
})
self.client_stubber = Stubber(
self.profiler._profiler_runner.collector.reporter.
codeguru_client_builder.codeguru_client)
report_expected_params = {
'agentProfile': ANY,
'contentType': 'application/json',
'profilingGroupName': DUMMY_TEST_PROFILING_GROUP_NAME
}
self.client_stubber.add_response(
'configure_agent',
{'configuration': {
'shouldProfile': True,
'periodInSeconds': 100
}})
self.client_stubber.add_response('post_agent_profile', {},
report_expected_params)
yield
self.profiler.stop()
def test_report_when_stopped(self):
with \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False), \
patch(
"codeguru_profiler_agent.sdk_reporter.sdk_reporter.SdkReporter.check_create_pg_called_during_submit_profile",
return_value=False):
with self.client_stubber:
self.profiler.start()
self.profiler.stop() # stop will trigger the report
# check that we have reported. assert_no_pending_responses will fail if post_agent_profile is not called
# with the expected parameters.
self.client_stubber.assert_no_pending_responses()
def test_report_via_invalid_reporting_mode(self):
self.profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
environment_override={
"reporting_mode": "invalid_reporting_mode",
})
self.profiler.start()
assert self.profiler.is_running() is False
class TestExceptionHandling:
@pytest.fixture(autouse=True)
def around(self):
self.profiler_runner = Mock(spec_set=ProfilerRunner)
self.profiler_runner.is_running = Mock(return_value=False)
self.profiler_runner.start = throw_exception
self.profiler = Profiler(
profiling_group_name="test-application",
environment_override={
"profiler_runner_factory":
mock_profiler_runner_factory(self.profiler_runner)
})
yield
self.profiler.stop()
def test_exceptions_are_caught_and_do_not_propagate(self):
assert (not self.profiler.start())
class TestWhenStartWasNotCalled:
@pytest.fixture(autouse=True)
def around(self):
self.profiler = Profiler(
profiling_group_name="test-application")
yield
def test_it_returns_true(self):
assert self.profiler.stop()
def test_it_samples_and_saves_a_profile_to_a_file(self):
with \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False):
file_prefix = str(Path(self.temporary_directory, FILE_PREFIX))
test_start_time = time_utils.current_milli_time()
profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
environment_override={
"initial_sampling_interval": timedelta(),
"reporting_mode": "file",
"file_prefix": file_prefix,
'agent_metadata':
AgentMetadata(fleet_info=DefaultFleetInfo())
})
try:
profiler.start()
finally:
profiler.stop()
test_end_time = time_utils.current_milli_time()
resulting_profile_path = str(
Path(self.temporary_directory,
os.listdir(self.temporary_directory)[0]))
with (open(resulting_profile_path)) as profiling_result_file:
resulting_json = json.loads(profiling_result_file.read())
self.assert_valid_agent_metadata(resulting_json["agentMetadata"])
assert test_start_time <= resulting_json[
"start"] <= resulting_json["end"] <= test_end_time
assert frames_in_callgraph_are_in_expected_order(
resulting_json["callgraph"],
"test.help_utils:HelperThreadRunner:dummy_parent_method",
"test.help_utils:HelperThreadRunner:dummy_method")
class TestWhenRunnerIsRunning:
@pytest.fixture(autouse=True)
def around(self):
self.profiler_runner = Mock(spec_set=ProfilerRunner)
self.profiler_runner.is_running = Mock(return_value=True)
self.profiler_runner.resume = Mock()
self.profiler = Profiler(
profiling_group_name="test-application",
environment_override={
"profiler_runner_factory":
mock_profiler_runner_factory(self.profiler_runner)
})
yield
self.profiler.stop()
def test_it_returns_true(self):
assert self.profiler.start()
def test_it_calls_resume(self):
self.profiler.start()
assert self.profiler_runner.resume.called
class TestPause:
@pytest.fixture(autouse=True)
def around(self):
self.profiler_runner = Mock(spec_set=ProfilerRunner)
self.profiler_runner.pause = Mock()
self.profiler = Profiler(profiling_group_name="test-application",
environment_override={
"profiler_runner_factory":
mock_profiler_runner_factory(
self.profiler_runner)
})
yield
self.profiler.stop()
def test_it_returns_true(self):
assert self.profiler.pause()
def test_it_calls_pause_on_runner(self):
self.profiler.pause()
assert self.profiler_runner.pause.called
def test_exceptions_are_caught_and_do_not_propagate(self):
self.profiler_runner.pause = throw_exception
assert (not self.profiler.pause())
def test_live_profiling(self):
with \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False), \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration._is_reporting_interval_smaller_than_minimum_allowed",
return_value=False):
profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
region_name='eu-west-2',
environment_override={
"initial_sampling_interval": timedelta(),
"sampling_interval": timedelta(seconds=1),
"reporting_interval": timedelta(seconds=2),
'agent_metadata':
AgentMetadata(fleet_info=DefaultFleetInfo())
})
client = profiler._profiler_runner.collector.reporter.codeguru_client_builder.codeguru_client
aggregator = profiler._profiler_runner.collector
assert AgentConfiguration.get().sampling_interval == timedelta(
seconds=1)
assert AgentConfiguration.get().reporting_interval == timedelta(
seconds=2)
with \
patch.object(client, "post_agent_profile",
wraps=client.post_agent_profile) as wrapped_post_agent_profile, \
patch.object(client, "configure_agent",
wraps=client.configure_agent) as wrapped_configure_agent, \
patch.object(aggregator, "add",
wraps=aggregator.add) as wrapped_add, \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False), \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration._is_reporting_interval_smaller_than_minimum_allowed",
return_value=False):
wrapped_configure_agent.return_value = {
"configuration": {
"agentParameters": {
"SamplingIntervalInMilliseconds": "100",
"MinimumTimeForReportingInMilliseconds": "1000",
"MaxStackDepth": "1000",
"MemoryUsageLimitPercent": "29"
},
"periodInSeconds": 2,
"shouldProfile": True
}
}
try:
start_status = profiler.start()
assert start_status
assert profiler.is_running()
time.sleep(3)
finally:
profiler.stop()
assert wrapped_add.call_count >= 3
assert wrapped_post_agent_profile.call_count >= 1
assert wrapped_configure_agent.call_count >= 1
assert AgentConfiguration.get().sampling_interval == timedelta(
seconds=1)
assert AgentConfiguration.get(
).reporting_interval == timedelta(seconds=2)
def test_live_profiling(self):
with \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False), \
patch(
"codeguru_profiler_agent.sdk_reporter.sdk_reporter.SdkReporter.check_create_pg_called_during_submit_profile",
return_value=False), \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration._is_reporting_interval_smaller_than_minimum_allowed",
return_value=False):
profiler = Profiler(
profiling_group_name=DUMMY_TEST_PROFILING_GROUP_NAME,
region_name='eu-west-2',
environment_override={
"initial_sampling_interval": timedelta(),
"sampling_interval": timedelta(seconds=1),
"reporting_interval": timedelta(seconds=2),
'agent_metadata':
AgentMetadata(fleet_info=DefaultFleetInfo())
})
client = profiler._profiler_runner.collector.reporter.codeguru_client_builder.codeguru_client
aggregator = profiler._profiler_runner.collector
assert AgentConfiguration.get().sampling_interval == timedelta(
seconds=1)
assert AgentConfiguration.get().reporting_interval == timedelta(
seconds=2)
with \
patch.object(client, "post_agent_profile",
wraps=client.post_agent_profile) as wrapped_post_agent_profile, \
patch.object(client, "configure_agent",
wraps=client.configure_agent) as wrapped_configure_agent, \
patch.object(aggregator, "add",
wraps=aggregator.add) as wrapped_add, \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration.is_under_min_reporting_time",
return_value=False), \
patch(
"codeguru_profiler_agent.reporter.agent_configuration.AgentConfiguration._is_reporting_interval_smaller_than_minimum_allowed",
return_value=False):
wrapped_configure_agent.return_value = {
"configuration": {
"agentParameters": {
"SamplingIntervalInMilliseconds": "100",
"MinimumTimeForReportingInMilliseconds": "1000",
"MaxStackDepth": "1000",
"MemoryUsageLimitPercent": "29"
},
"periodInSeconds": 2,
"shouldProfile": True
}
}
try:
start_status = profiler.start()
assert start_status
assert profiler.is_running()
time.sleep(4)
finally:
profiler.stop()
# We should see at least 2 samples in 4 seconds as the sequence should happen in the order of
# initial delay (1 second)
# After 1 second, no flush -> sample
# After 2 seconds, it attempt to flush (possibly succeed) -> sample/ no sample
# After 3 seconds, it attempt to flush (must succeed if it did not flush before) -> no sample/ sample
# After 4 seconds, no flush -> sample (if profiler has not stopped yet)
assert wrapped_add.call_count >= 2
assert wrapped_post_agent_profile.call_count >= 1
assert wrapped_configure_agent.call_count >= 1
assert AgentConfiguration.get().sampling_interval == timedelta(
seconds=1)
assert AgentConfiguration.get(
).reporting_interval == timedelta(seconds=2)