def test_detect_changes_few_revisions_many_values():
'''
Tests that we correctly detect a regression with
a small number of revisions but a large number of values
'''
data = [RevisionDatum(0, 0, [0]*50+[1]*30),
RevisionDatum(1, 1, [0]*10+[1]*30),
RevisionDatum(1, 1, [0]*10+[1]*30)]
result = [(d.push_timestamp, d.change_detected) for d in
detect_changes(data, min_back_window=5, max_back_window=10,
fore_window=5, t_threshold=2)]
assert result == [(0, False),
(1, True),
(1, False)]
def test_detect_changes():
data = []
times = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
values = [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1]
for (t, v) in zip(times, values):
data.append(RevisionDatum(t, t, [float(v)]))
result = [
(d.push_timestamp, d.change_detected)
for d in detect_changes(
data, min_back_window=5, max_back_window=5, fore_window=5, t_threshold=2
)
]
assert result == [
(0, False),
(1, False),
(2, False),
(3, False),
(4, False),
(5, False),
(6, False),
(7, False),
(8, True),
(9, False),
(10, False),
(11, False),
(12, False),
(13, False),
(14, False),
(15, False),
]
def test_detect_changes_historical_data(filename, expected_timestamps):
"""Parse JSON produced by http://graphs.mozilla.org/api/test/runs"""
# Configuration for Analyzer
FORE_WINDOW = 12
MIN_BACK_WINDOW = 12
MAX_BACK_WINDOW = 24
THRESHOLD = 7
payload = SampleData.get_perf_data(os.path.join('graphs', filename))
runs = payload['test_runs']
data = [RevisionDatum(r[2], r[2], [r[3]]) for r in runs]
results = detect_changes(data,
min_back_window=MIN_BACK_WINDOW,
max_back_window=MAX_BACK_WINDOW,
fore_window=FORE_WINDOW,
t_threshold=THRESHOLD)
regression_timestamps = [d.push_timestamp for d in results if
d.change_detected]
assert regression_timestamps == expected_timestamps
def generate_new_alerts_in_series(signature):
# get series data starting from either:
# (1) the last alert, if there is one
# (2) the alerts max age
# (use whichever is newer)
max_alert_age = datetime.now() - settings.PERFHERDER_ALERTS_MAX_AGE
series = PerformanceDatum.objects.filter(signature=signature,
push_timestamp__gte=max_alert_age)
latest_alert_timestamp = (PerformanceAlert.objects.filter(
series_signature=signature).select_related('summary__push__time').
order_by('-summary__push__time').values_list(
'summary__push__time', flat=True)[:1])
if latest_alert_timestamp:
series = series.filter(push_timestamp__gt=latest_alert_timestamp[0])
revision_data = {}
for d in series:
if not revision_data.get(d.push_id):
revision_data[d.push_id] = RevisionDatum(
int(time.mktime(d.push_timestamp.timetuple())), d.push_id, [])
revision_data[d.push_id].values.append(d.value)
min_back_window = signature.min_back_window
if min_back_window is None:
min_back_window = settings.PERFHERDER_ALERTS_MIN_BACK_WINDOW
max_back_window = signature.max_back_window
if max_back_window is None:
max_back_window = settings.PERFHERDER_ALERTS_MAX_BACK_WINDOW
fore_window = signature.fore_window
if fore_window is None:
fore_window = settings.PERFHERDER_ALERTS_FORE_WINDOW
alert_threshold = signature.alert_threshold
if alert_threshold is None:
alert_threshold = settings.PERFHERDER_REGRESSION_THRESHOLD
analyzed_series = detect_changes(
revision_data.values(),
min_back_window=min_back_window,
max_back_window=max_back_window,
fore_window=fore_window,
)
with transaction.atomic():
for (prev, cur) in zip(analyzed_series, analyzed_series[1:]):
if cur.change_detected:
prev_value = cur.historical_stats['avg']
new_value = cur.forward_stats['avg']
alert_properties = get_alert_properties(
prev_value, new_value, signature.lower_is_better)
# ignore regressions below the configured regression
# threshold
if ((signature.alert_change_type is None
or signature.alert_change_type
== PerformanceSignature.ALERT_PCT)
and alert_properties.pct_change < alert_threshold) or (
signature.alert_change_type
== PerformanceSignature.ALERT_ABS
and alert_properties.delta < alert_threshold):
continue
summary, _ = PerformanceAlertSummary.objects.get_or_create(
repository=signature.repository,
framework=signature.framework,
push_id=cur.push_id,
prev_push_id=prev.push_id,
defaults={
'manually_created': False,
'created':
datetime.utcfromtimestamp(cur.push_timestamp),
},
)
# django/mysql doesn't understand "inf", so just use some
# arbitrarily high value for that case
t_value = cur.t
if t_value == float('inf'):
t_value = 1000
PerformanceAlert.objects.update_or_create(
summary=summary,
series_signature=signature,
defaults={
'is_regression': alert_properties.is_regression,
'amount_pct': alert_properties.pct_change,
'amount_abs': alert_properties.delta,
'prev_value': prev_value,
'new_value': new_value,
't_value': t_value,
},
)
def test_calc_t(old_data, new_data, expected):
assert calc_t([RevisionDatum(0, 0, old_data)], [RevisionDatum(1, 1, new_data)]) == expected
def test_analyze_fn(revision_data, weight_fn, expected):
data = [
RevisionDatum(i, i, values) for (i, values) in zip(range(len(revision_data)), revision_data)
]
assert analyze(data, weight_fn) == expected
def handle(self, *args, **options):
if not options['project']:
raise CommandError("Must specify at least one project with "
"--project")
pc = PerfherderClient(server_url=options['server'])
option_collection_hash = pc.get_option_collection_hash()
# print csv header
print(','.join([
"project",
"platform",
"signature",
"series",
"testrun_id",
"push_timestamp",
"change",
"percent change",
"t-value",
"revision",
]))
for project in options['project']:
if options['signature']:
signatures = [options['signature']]
signature_data = pc.get_performance_signatures(
project,
signatures=signatures,
interval=options['time_interval'])
else:
signature_data = pc.get_performance_signatures(
project, interval=options['time_interval'])
signatures = []
signatures_to_ignore = set()
# if doing everything, only handle summary series
for (signature, properties) in signature_data.items():
signatures.append(signature)
if 'subtest_signatures' in properties:
# Don't alert on subtests which have a summary
signatures_to_ignore.update(
properties['subtest_signatures'])
signatures = [
signature for signature in signatures
if signature not in signatures_to_ignore
]
for signature in signatures:
series = pc.get_performance_data(
project,
signatures=signature,
interval=options['time_interval'])[signature]
series_properties = signature_data.get(signature)
data = []
for (timestamp, value) in zip(series['push_timestamp'],
series['value']):
data.append(RevisionDatum(timestamp, value))
for r in detect_changes(data):
if r.state == 'regression':
pushes = pc.get_pushes(project, id=r.testrun_id)
revision = pushes[0]['revision'] if pushes else ''
initial_value = r.historical_stats['avg']
new_value = r.forward_stats['avg']
if initial_value != 0:
pct_change = (100.0 *
abs(new_value - initial_value) /
float(initial_value))
else:
pct_change = 0.0
delta = new_value - initial_value
print(','.join(
map(
str,
[
project,
series_properties['machine_platform'],
signature,
self._get_series_description(
option_collection_hash,
series_properties),
r.testrun_id,
r.push_timestamp,
delta,
pct_change,
r.t,
revision[0:12],
],
)))
def generate_new_alerts_in_series(signature):
# get series data starting from either:
# (1) the last alert, if there is one
# (2) the alerts max age
# (use whichever is newer)
max_alert_age = datetime.now() - settings.PERFHERDER_ALERTS_MAX_AGE
series = PerformanceDatum.objects.filter(signature=signature,
push_timestamp__gte=max_alert_age)
latest_alert_timestamp = (PerformanceAlert.objects.filter(
series_signature=signature).select_related('summary__push__time').
order_by('-summary__push__time').values_list(
'summary__push__time', flat=True)[:1])
if latest_alert_timestamp:
series = series.filter(push_timestamp__gt=latest_alert_timestamp[0])
revision_data = {}
for d in series:
if not revision_data.get(d.push_id):
revision_data[d.push_id] = RevisionDatum(
int(time.mktime(d.push_timestamp.timetuple())), d.push_id, [])
revision_data[d.push_id].values.append(d.value)
min_back_window = signature.min_back_window
if min_back_window is None:
min_back_window = settings.PERFHERDER_ALERTS_MIN_BACK_WINDOW
max_back_window = signature.max_back_window
if max_back_window is None:
max_back_window = settings.PERFHERDER_ALERTS_MAX_BACK_WINDOW
fore_window = signature.fore_window
if fore_window is None:
fore_window = settings.PERFHERDER_ALERTS_FORE_WINDOW
alert_threshold = signature.alert_threshold
if alert_threshold is None:
alert_threshold = settings.PERFHERDER_REGRESSION_THRESHOLD
data = revision_data.values()
analyzed_series = detect_changes(
data,
min_back_window=min_back_window,
max_back_window=max_back_window,
fore_window=fore_window,
)
with transaction.atomic():
for (prev, cur) in zip(analyzed_series, analyzed_series[1:]):
if cur.change_detected:
prev_value = cur.historical_stats['avg']
new_value = cur.forward_stats['avg']
alert_properties = get_alert_properties(
prev_value, new_value, signature.lower_is_better)
noise_profile = "N/A"
try:
# Gather all data up to the current data point that
# shows the regression and obtain a noise profile on it.
# This helps us to ignore this alert and others in the
# calculation that could influence the profile.
noise_data = []
for point in analyzed_series:
if point == cur:
break
noise_data.append(geomean(point.values))
noise_profile, _ = deviance(noise_data)
if not isinstance(noise_profile, str):
raise Exception(
"Expecting a string as a "
f"noise profile, got: {type(noise_profile)}")
except Exception:
# Fail without breaking the alert computation
newrelic.agent.record_exception()
logger.error("Failed to obtain a noise profile.")
# ignore regressions below the configured regression
# threshold
if ((signature.alert_change_type is None
or signature.alert_change_type
== PerformanceSignature.ALERT_PCT)
and alert_properties.pct_change < alert_threshold) or (
signature.alert_change_type
== PerformanceSignature.ALERT_ABS
and alert_properties.delta < alert_threshold):
continue
summary, _ = PerformanceAlertSummary.objects.get_or_create(
repository=signature.repository,
framework=signature.framework,
push_id=cur.push_id,
prev_push_id=prev.push_id,
defaults={
'manually_created': False,
'created':
datetime.utcfromtimestamp(cur.push_timestamp),
},
)
# django/mysql doesn't understand "inf", so just use some
# arbitrarily high value for that case
t_value = cur.t
if t_value == float('inf'):
t_value = 1000
PerformanceAlert.objects.update_or_create(
summary=summary,
series_signature=signature,
defaults={
'noise_profile': noise_profile,
'is_regression': alert_properties.is_regression,
'amount_pct': alert_properties.pct_change,
'amount_abs': alert_properties.delta,
'prev_value': prev_value,
'new_value': new_value,
't_value': t_value,
},
)