def test_cq_status_fetch_stats_new_fetch_from_last_build_run(self):
urlfetch_mock = mock.Mock()
urlfetch_mock.return_value.content = json.dumps({
'more': False, 'cursor': None, 'results': []})
FetchStatus(cursor='xxx', begin='1', end='2', done=True).put()
BuildRun(time_started=datetime.datetime(2015, 10, 30, 12, 17, 42),
time_finished=datetime.datetime(2015, 10, 30, 13, 17, 42),
buildnumber=0, result=0).put()
BuildRun(time_started=datetime.datetime(2015, 10, 30, 11, 17, 42),
time_finished=datetime.datetime(2015, 10, 30, 12, 17, 42),
buildnumber=0, result=0).put()
with mock.patch('google.appengine.api.urlfetch.fetch', urlfetch_mock):
cq_status.fetch_cq_status()
urlfetch_mock.assert_called_once_with(
'https://chromium-cq-status.appspot.com/query?'
'tags=action=verifier_jobs_update&begin=1446211062.0&end=1446214662.0&'
'count=10')
def _create_flake():
tf = datetime.datetime.utcnow()
ts = tf - datetime.timedelta(hours=1)
p = PatchsetBuilderRuns(issue=1,
patchset=1,
master='tryserver.bar',
builder='baz').put()
br_f0 = BuildRun(parent=p,
buildnumber=10,
result=2,
time_started=ts,
time_finished=tf).put()
br_f1 = BuildRun(parent=p,
buildnumber=20,
result=2,
time_started=ts,
time_finished=tf).put()
br_s0 = BuildRun(parent=p,
buildnumber=30,
result=0,
time_started=ts,
time_finished=tf).put()
occ1 = FlakyRun(failure_run=br_f0,
success_run=br_s0,
failure_run_time_started=ts,
failure_run_time_finished=tf,
flakes=[
FlakeOccurrence(name='step1', failure='testX'),
])
occ2 = FlakyRun(failure_run=br_f1,
success_run=br_s0,
failure_run_time_started=ts,
failure_run_time_finished=tf,
flakes=[
FlakeOccurrence(name='step2', failure='testX'),
FlakeOccurrence(name='step3', failure='step3'),
])
f = Flake(name='testX',
count_day=10,
occurrences=[occ1.put(), occ2.put()],
is_step=True,
issue_id=123456)
return f, [occ1, occ2]
def test_ignores_null_flaky_runs(self):
last_updated = datetime.datetime.now()
fake_build_key = BuildRun(buildnumber=1,
result=1,
time_finished=last_updated).put()
flake_run_key = FlakyRun(failure_run=fake_build_key,
success_run=fake_build_key,
failure_run_time_finished=last_updated,
flakes=[
FlakeOccurrence(name='fake_step',
failure='fake_test_name'),
FlakeOccurrence(name='fake_step2',
failure='fake_test_name')
]).put()
null_flake_run_key = ndb.Key('FlakyRun', 'fake-key')
Flake(issue_id=1,
is_step=False,
name='fake_test_name',
issue_last_updated=last_updated,
occurrences=[
flake_run_key,
null_flake_run_key,
]).put()
self.test_app.get('/migrate')
flake_types = FlakeType.query().fetch()
self.assertEqual(len(flake_types), 2)
flake_type_1 = flake_types[0]
self.assertEqual(flake_type_1.project, 'chromium')
self.assertEqual(flake_type_1.step_name, 'fake_step')
self.assertEqual(flake_type_1.test_name, 'fake_test_name')
self.assertIsNone(flake_type_1.config)
self.assertEqual(flake_type_1.last_updated, last_updated)
flake_type_2 = flake_types[1]
self.assertEqual(flake_type_2.project, 'chromium')
self.assertEqual(flake_type_2.step_name, 'fake_step2')
self.assertEqual(flake_type_2.test_name, 'fake_test_name')
self.assertIsNone(flake_type_2.config)
self.assertEqual(flake_type_2.last_updated, last_updated)
issues = Issue.query().fetch()
self.assertEqual(len(issues), 1)
issue = issues[0]
self.assertEqual(issue.issue_id, 1)
self.assertEqual(issue.project, 'chromium')
self.assertEqual(sorted(issue.flake_type_keys),
sorted(flake_type.key for flake_type in flake_types))
def _create_flake(self):
tf = datetime.datetime(2016, 8, 6, 10, 20, 30)
ts = tf - datetime.timedelta(hours=1)
tf2 = tf - datetime.timedelta(days=5)
ts2 = tf2 - datetime.timedelta(hours=1)
p = PatchsetBuilderRuns(issue=123456, patchset=1, master='tryserver.test',
builder='test-builder').put()
br_f0 = BuildRun(parent=p, buildnumber=0, result=2, time_started=ts2,
time_finished=tf2).put()
br_f1 = BuildRun(parent=p, buildnumber=1, result=2, time_started=ts,
time_finished=tf).put()
br_s1 = BuildRun(parent=p, buildnumber=2, result=0, time_started=ts,
time_finished=tf).put()
br_f2 = BuildRun(parent=p, buildnumber=3, result=4, time_started=ts,
time_finished=tf).put()
br_s2 = BuildRun(parent=p, buildnumber=4, result=0, time_started=ts,
time_finished=tf).put()
occ_key1 = FlakyRun(failure_run=br_f0, success_run=br_s2,
flakes=[
FlakeOccurrence(name='foo (x)', failure='foo.bar'),
FlakeOccurrence(name='foo (x)', failure='other')],
failure_run_time_started=ts2,
failure_run_time_finished=tf2).put()
occ_key2 = FlakyRun(failure_run=br_f1, success_run=br_s1,
flakes=[
FlakeOccurrence(name='bar (y)', failure='foo.bar')],
failure_run_time_started=ts,
failure_run_time_finished=tf).put()
occ_key3 = FlakyRun(failure_run=br_f2, success_run=br_s2,
flakes=[
FlakeOccurrence(
name='foo (x)', failure='foo.bar', issue_id=100),
FlakeOccurrence(
name='bar (y)', failure='foo.bar', issue_id=200)],
failure_run_time_started=ts,
failure_run_time_finished=tf).put()
return Flake(name='foo.bar', count_day=10, is_step=False,
occurrences=[occ_key1, occ_key2, occ_key3])
def post(self):
if (not self.request.get('failure_run_key') or
not self.request.get('success_run_key')):
self.response.set_status(400, 'Invalid request parameters')
return
failure_run = ndb.Key(urlsafe=self.request.get('failure_run_key')).get()
success_run = ndb.Key(urlsafe=self.request.get('success_run_key')).get()
flaky_run = FlakyRun(
failure_run=failure_run.key,
failure_run_time_started=failure_run.time_started,
failure_run_time_finished=failure_run.time_finished,
success_run=success_run.key)
failure_time = failure_run.time_finished
patchset_builder_runs = failure_run.key.parent().get()
master = BuildRun.removeMasterPrefix(patchset_builder_runs.master)
url = ('https://chrome-build-extract.appspot.com/p/' + master +
'/builders/' + patchset_builder_runs.builder +'/builds/' +
str(failure_run.buildnumber) + '?json=1')
urlfetch.set_default_fetch_deadline(60)
logging.info('get_flaky_run_reason ' + url)
response = urlfetch.fetch(url)
if response.status_code >= 400 and response.status_code <= 599:
logging.error('The request to %s has returned %d: %s', url,
response.status_code, response.content)
self.response.set_status(500, 'Failed to fetch build.')
return
json_result = json.loads(response.content)
steps = json_result['steps']
failed_steps = []
passed_steps = []
for step in steps:
result = step['results'][0]
if build_result.isResultSuccess(result):
passed_steps.append(step)
continue
if not build_result.isResultFailure(result):
continue
step_name = step['name']
step_text = ' '.join(step['text'])
if step_name in IGNORED_STEPS:
continue
# Custom (non-trivial) rules for ignoring flakes in certain steps:
# - [swarming] ...: summary step would also be red (do not double count)
# - Patch failure: ingore non-infra failures as they are typically due to
# changes in the code on HEAD
# - bot_update PATCH FAILED: Duplicates failure in 'Patch failure' step.
# - ... (retry summary): this is an artificial step to fail the build due
# to another step that has failed earlier (do not double count).
if (step_name.startswith('[swarming]') or
(step_name == 'Patch failure' and result != build_result.EXCEPTION) or
(step_name == 'bot_update' and 'PATCH FAILED' in step_text)):
continue
failed_steps.append(step)
steps_to_ignore = []
for step in failed_steps:
step_name = step['name']
if '(with patch)' in step_name:
# Ignore any steps from the same test suite, which is determined by the
# normalized step name. Additionally, if the step fails without patch,
# ignore the original step as well because tree is busted.
normalized_step_name = normalize_test_type(step_name, True)
for other_step in failed_steps:
if other_step == step:
continue
normalized_other_step_name = normalize_test_type(
other_step['name'], True)
if normalized_other_step_name == normalized_step_name:
steps_to_ignore.append(other_step['name'])
if '(without patch)' in other_step['name']:
steps_to_ignore.append(step['name'])
flakes_to_update = []
for step in failed_steps:
step_name = step['name']
if step_name in steps_to_ignore:
continue
flakes, is_step = self.get_flakes(
master, patchset_builder_runs.builder, failure_run.buildnumber, step)
for flake in flakes:
flake_occurrence = FlakeOccurrence(name=step_name, failure=flake)
flaky_run.flakes.append(flake_occurrence)
flakes_to_update.append((flake, is_step))
# Do not create FlakyRuns if all failed steps have been ignored.
if not flaky_run.flakes:
return
flaky_run_key = flaky_run.put()
for flake, is_step in flakes_to_update:
self.add_failure_to_flake(flake, flaky_run_key, failure_time, is_step)
self.flaky_runs.increment_by(1)
def fetch_cq_status():
"""Fetches data from chromium-cq-status app and saves new data.
Remembers old cursor and fetches new data.
"""
fetch_status = FetchStatus.query().get()
cursor = ''
begin = ''
end = ''
retry_count = 0
while True:
if fetch_status:
if fetch_status.done:
logging.info('historical fetching done so fetch latest...')
end = str(time_functions.timestamp.utcnow_ts())
last_build_run_seen = BuildRun.query().order(
-BuildRun.time_finished).fetch(1)
begin = str(time_functions.timestamp.utctimestamp(
last_build_run_seen[0].time_finished))
cursor = ''
else:
begin = fetch_status.begin
end = fetch_status.end
cursor = fetch_status.cursor
else:
logging.info('didnt find any historical information. fetching last week')
begin = str(time_functions.timestamp.utctimestamp(
datetime.datetime.utcnow() - datetime.timedelta(weeks=1)))
end = str(time_functions.timestamp.utcnow_ts())
if begin and end:
logging.info('fetching from ' +
str(datetime.datetime.fromtimestamp(float(begin))) + ' to ' +
str(datetime.datetime.fromtimestamp(float(end))) +
' cursor: ' + cursor)
else:
logging.info('fetching with no begin/end and cursor: ' + cursor)
url = "https://chromium-cq-status.appspot.com/query"
params = []
params.append('tags=action=verifier_jobs_update')
if cursor:
params.append('cursor=' + cursor)
if begin:
params.append('begin=' + begin)
if end:
params.append('end=' + end)
# Tried count of 200 or more but would get OOM or deadline errors. Even 50
# sometimes gives:
# "Values may not be more than 1000000 bytes in length; received 2118015
# bytes"
params.append('count=10')
url += '?' + '&'.join(params)
logging.info('fetching url: ' + url)
try:
urlfetch.set_default_fetch_deadline(60)
result = urlfetch.fetch(url).content
timestamp_str = '"timestamp":"'
logging_idx = result.find(timestamp_str)
if logging_idx != -1:
logging_idx += len(timestamp_str)
logging_idx2 = result.find('"', logging_idx)
logging.info(' current fetch has time of ' +
result[logging_idx:logging_idx2])
try:
json_result = json.loads(result)
more = json_result['more']
cursor = json_result['cursor']
try:
logging_output = parse_cq_data(json_result)
if logging_output:
logging.info('found flakes: ' + ' '.join(logging_output))
except DeadlineExceededError:
logging.info('got DeadlineExceededError during parse_cq_data, '
'catching to not show up as error')
return
except ValueError:
requests_metric.increment_by(1, fields={'status': 'parse_error'})
logging.exception('failed to parse CQ data from %s', url)
if 'DeadlineExceededError' in result:
logging.error('got deadline exceeded, trying again after 1s')
time.sleep(1)
continue
elif retry_count < 3:
retry_count += 1
logging.error('will retry after sleeping ' + str(retry_count))
time.sleep(retry_count)
continue
else:
logging.error('giving up and will count current fetch as done')
# Don't want to continue this as could be a bad cursor.
more = False
else:
requests_metric.increment_by(1, fields={'status': 'success'})
if not fetch_status:
fetch_status = FetchStatus()
fetch_status.done = not more
if fetch_status.done:
fetch_status.cursor = ''
fetch_status.begin = ''
fetch_status.end = ''
retry_count = 0
logging.info('finished fetching for current cursor')
else:
fetch_status.begin = begin
fetch_status.end = end
fetch_status.cursor = cursor
fetch_status.put()
if not more:
return # finish the cron job and wait for next iteration
except urllib2.URLError, e:
requests_metric.increment_by(1, fields={'status': 'fetch_error'})
logging.warning('Failed to fetch CQ status: %s', e.reason)
def parse_cq_data(json_data):
logging_output = []
for result in json_data.get('results', {}):
fields = result.get('fields', [])
if not 'action' in fields:
continue
action = fields.get('action')
if action != 'verifier_jobs_update':
continue
if fields.get('verifier') != 'try job':
continue
# At the moment, much of the parsing logic assumes this is a Chromium
# tryjob.
if fields.get('project') != 'chromium':
continue
job_states = fields.get('jobs', [])
for state in job_states:
# Just go by |result|.
#if state not in ['JOB_SUCCEEDED', 'JOB_FAILED', 'JOB_TIMED_OUT']:
# continue
for job in job_states[state]:
build_properties = job.get('build_properties')
if not build_properties:
continue
try:
master = job['master']
builder = job['builder']
result = job['result']
timestamp_tz = dateutil.parser.parse(job['timestamp'])
# We assume timestamps from chromium-cq-status are already in UTC.
timestamp = timestamp_tz.replace(tzinfo=None)
except KeyError:
continue
try:
buildnumber = get_int_value(build_properties, 'buildnumber')
issue = get_int_value(build_properties, 'issue')
patchset = get_int_value(build_properties, 'patchset')
attempt_start_ts = get_int_value(build_properties, 'attempt_start_ts')
time_started = datetime.datetime.utcfromtimestamp(
attempt_start_ts / 1000000)
except ValueError:
continue
if build_result.isResultPending(result):
continue
# At this point, only success or failure.
success = build_result.isResultSuccess(result)
patchset_builder_runs = get_patchset_builder_runs(issue=issue,
patchset=patchset,
master=master,
builder=builder)
build_run = BuildRun(parent=patchset_builder_runs.key,
buildnumber=buildnumber,
result=result,
time_started=time_started,
time_finished=timestamp)
previous_runs = BuildRun.query(
ancestor=patchset_builder_runs.key).fetch()
duplicate = False
for previous_run in previous_runs:
# We saw this build run already or there are multiple green runs,
# in which case we ignore subsequent ones to avoid showing failures
# multiple times.
if (previous_run.buildnumber == buildnumber) or \
(build_run.is_success and previous_run.is_success) :
duplicate = True
break
if duplicate:
continue
build_run.put()
for previous_run in previous_runs:
if previous_run.is_success == build_run.is_success:
continue
if success:
# We saw the flake and then the pass.
failure_run = previous_run
success_run = build_run
else:
# We saw the pass and then the failure. Could happen when fetching
# historical data, or for the bot_update step (patch can't be
# applied cleanly anymore).
failure_run = build_run
success_run = previous_run
logging_output.append(failure_run.key.parent().get().builder +
str(failure_run.buildnumber))
# Queue a task to fetch the error of this failure and create FlakyRun.
flakes_metric.increment_by(1)
taskqueue.add(
queue_name='issue-updates',
url='/issues/create_flaky_run',
params={'failure_run_key': failure_run.key.urlsafe(),
'success_run_key': success_run.key.urlsafe()})
return logging_output
def post(self):
if (not self.request.get('failure_run_key') or
not self.request.get('success_run_key')):
self.response.set_status(400, 'Invalid request parameters')
return
failure_run = ndb.Key(urlsafe=self.request.get('failure_run_key')).get()
success_run = ndb.Key(urlsafe=self.request.get('success_run_key')).get()
flaky_run = FlakyRun(
failure_run=failure_run.key,
failure_run_time_started=failure_run.time_started,
failure_run_time_finished=failure_run.time_finished,
success_run=success_run.key)
failure_time = failure_run.time_finished
patchset_builder_runs = failure_run.key.parent().get()
master = BuildRun.removeMasterPrefix(patchset_builder_runs.master)
url = ('https://luci-milo.appspot.com/'
'prpc/milo.Buildbot/GetBuildbotBuildJSON')
request = json.dumps({
'master': master,
'builder': patchset_builder_runs.builder,
'buildNum': failure_run.buildnumber,
})
headers = {
'Content-Type': 'application/json',
'Accept': 'application/json',
}
urlfetch.set_default_fetch_deadline(60)
logging.info('get_flaky_run_reason: %s, %s', url, request)
response = urlfetch.fetch(
url, payload=request, method=urlfetch.POST, headers=headers,
validate_certificate=True)
if response.status_code != 200:
logging.error('The request to %s has returned %d: %s', url,
response.status_code, response.content)
self.response.set_status(500, 'Failed to fetch build.')
return
content = response.content
if content.startswith(_MILO_RESPONSE_PREFIX):
content = content[len(_MILO_RESPONSE_PREFIX):]
data = json.loads(content)['data']
json_result = json.loads(base64.b64decode(data))
steps = json_result['steps']
failed_steps = []
passed_steps = []
for step in steps:
result = step['results'][0]
if build_result.isResultSuccess(result):
passed_steps.append(step)
continue
if not build_result.isResultFailure(result):
continue
# For Luci builds, some steps don't have step text anymore. Such steps
# include 'Failure reason', 'analyze', etc.
step_text = ' '.join(step['text'] or [])
step_name = step['name']
if step_name in IGNORED_STEPS:
continue
# Custom (non-trivial) rules for ignoring flakes in certain steps:
# - [swarming] ...: summary step would also be red (do not double count)
# - Patch failure: ingore non-infra failures as they are typically due to
# changes in the code on HEAD
# - bot_update PATCH FAILED: Duplicates failure in 'Patch failure' step.
# - ... (retry summary): this is an artificial step to fail the build due
# to another step that has failed earlier (do not double count).
if (step_name.startswith('[swarming]') or
(step_name == 'Patch failure' and result != build_result.EXCEPTION) or
(step_name == 'bot_update' and 'PATCH FAILED' in step_text)):
continue
failed_steps.append(step)
steps_to_ignore = []
for step in failed_steps:
step_name = step['name']
if '(with patch)' in step_name:
# Ignore any steps from the same test suite, which is determined by the
# normalized step name. Additionally, if the step fails without patch,
# ignore the original step as well because tree is busted.
normalized_step_name = normalize_test_type(step_name, True)
for other_step in failed_steps:
if other_step == step:
continue
normalized_other_step_name = normalize_test_type(
other_step['name'], True)
if normalized_other_step_name == normalized_step_name:
steps_to_ignore.append(other_step['name'])
if '(without patch)' in other_step['name']:
steps_to_ignore.append(step['name'])
flakes_to_update = []
for step in failed_steps:
step_name = step['name']
if step_name in steps_to_ignore:
continue
flakes, is_step = self.get_flakes(
master, patchset_builder_runs.builder, failure_run.buildnumber, step)
if is_step and not is_infra_step_flake(step_name):
continue # Ignore flakes of non-infra steps.
for flake in flakes:
flake_occurrence = FlakeOccurrence(name=step_name, failure=flake)
flaky_run.flakes.append(flake_occurrence)
flakes_to_update.append((flake, is_step))
# Do not create FlakyRuns if all failed steps have been ignored.
if not flaky_run.flakes:
return
flaky_run_key = flaky_run.put()
for flake, is_step in flakes_to_update:
if self.is_duplicate_occurrence(flake, flaky_run):
logging.info('Not adding duplicate occurrence for the same CL')
continue
self.add_failure_to_flake(flake, flaky_run_key, failure_time, is_step)
self.flaky_runs.increment_by(1)
def fetch_cq_status():
"""Fetches data from chromium-cq-status app and saves new data.
Remembers old cursor and fetches new data.
"""
fetch_status = FetchStatus.query().get()
cursor = ''
begin = ''
end = ''
retry_count = 0
while True:
if fetch_status:
if fetch_status.done:
logging.info('historical fetching done so fetch latest...')
end = str(time_functions.timestamp.utcnow_ts())
last_build_run_seen = BuildRun.query().order(
-BuildRun.time_finished).fetch(1)
begin = str(
time_functions.timestamp.utctimestamp(
last_build_run_seen[0].time_finished))
cursor = ''
else:
begin = fetch_status.begin
end = fetch_status.end
cursor = fetch_status.cursor
else:
logging.info(
'didnt find any historical information. fetching last week')
begin = str(
time_functions.timestamp.utctimestamp(
datetime.datetime.utcnow() - datetime.timedelta(weeks=1)))
end = str(time_functions.timestamp.utcnow_ts())
if begin and end:
logging.info('fetching from %s to %s cursor: %s',
str(datetime.datetime.utcfromtimestamp(float(begin))),
str(datetime.datetime.utcfromtimestamp(float(end))),
cursor)
else:
logging.info('fetching with no begin/end and cursor: ' + cursor)
url = "https://chromium-cq-status.appspot.com/query"
params = []
params.append('tags=action=verifier_jobs_update')
if cursor:
params.append('cursor=' + cursor)
if begin:
params.append('begin=' + begin)
if end:
params.append('end=' + end)
# Tried count of 200 or more but would get OOM or deadline errors. Even 50
# sometimes gives:
# "Values may not be more than 1000000 bytes in length; received 2118015
# bytes"
params.append('count=10')
url += '?' + '&'.join(params)
logging.info('fetching url: ' + url)
try:
urlfetch.set_default_fetch_deadline(60)
result = urlfetch.fetch(url).content
timestamp_str = '"timestamp":"'
logging_idx = result.find(timestamp_str)
if logging_idx != -1:
logging_idx += len(timestamp_str)
logging_idx2 = result.find('"', logging_idx)
logging.info(' current fetch has time of ' +
result[logging_idx:logging_idx2])
try:
json_result = json.loads(result)
more = json_result['more']
cursor = json_result['cursor']
try:
logging_output = parse_cq_data(json_result)
if logging_output:
logging.info('found flakes: ' +
' '.join(logging_output))
except DeadlineExceededError:
logging.info(
'got DeadlineExceededError during parse_cq_data, '
'catching to not show up as error')
return
except ValueError:
requests_metric.increment_by(1,
fields={'status': 'parse_error'})
logging.exception('failed to parse CQ data from %s', url)
if 'DeadlineExceededError' in result:
logging.error(
'got deadline exceeded, trying again after 1s')
time.sleep(1)
continue
elif retry_count < 3:
retry_count += 1
logging.error('will retry after sleeping ' +
str(retry_count))
time.sleep(retry_count)
continue
else:
logging.error(
'giving up and will count current fetch as done')
# Don't want to continue this as could be a bad cursor.
more = False
else:
requests_metric.increment_by(1, fields={'status': 'success'})
if not fetch_status:
fetch_status = FetchStatus()
fetch_status.done = not more
if fetch_status.done:
fetch_status.cursor = ''
fetch_status.begin = ''
fetch_status.end = ''
retry_count = 0
logging.info('finished fetching for current cursor')
else:
fetch_status.begin = begin
fetch_status.end = end
fetch_status.cursor = cursor
fetch_status.put()
if not more:
return # finish the cron job and wait for next iteration
except urllib2.URLError, e:
requests_metric.increment_by(1, fields={'status': 'fetch_error'})
logging.warning('Failed to fetch CQ status: %s', e.reason)
def parse_cq_data(json_data):
logging_output = []
for result in json_data.get('results', {}):
fields = result.get('fields', [])
if not 'action' in fields:
logging.warning('Missing field action in status record')
parsing_errors.increment_by(1)
continue
action = fields.get('action')
if action != 'verifier_jobs_update':
continue
if fields.get('verifier') != 'try job':
continue
# At the moment, much of the parsing logic assumes this is a Chromium
# tryjob.
project = fields.get('project')
if project != 'chromium/chromium/src':
logging.info('project not chromium: %s', project)
continue
job_states = fields.get('jobs', {})
for job in itertools.chain.from_iterable(job_states.values()):
try:
builder = job['builder']
result = job['result']
timestamp_tz = dateutil.parser.parse(
job.get('created_ts') or job['timestamp'])
# We assume timestamps from chromium-cq-status are already in UTC.
timestamp = timestamp_tz.replace(tzinfo=None)
except KeyError:
logging.warning('Failed to parse job details', exc_info=True)
parsing_errors.increment_by(1)
continue
if build_result.isResultPending(result):
continue
build_properties = job.get('build_properties')
if not build_properties:
logging.warning(
'Missing field build_properties in job details')
parsing_errors.increment_by(1)
continue
issue = -1
patchset = -1
time_started = 0
try:
buildnumber = get_int_value(build_properties, 'buildnumber')
if 'patch_issue' in build_properties:
issue = get_int_value(build_properties, 'patch_issue')
else: # pragma: no cover
logging.warning('no issue')
if 'patch_set' in build_properties:
patchset = get_int_value(build_properties, 'patch_set')
else: # pragma: no cover
logging.warning('no patchset')
if 'attempt_start_ts' in build_properties:
attempt_start_ts = get_int_value(build_properties,
'attempt_start_ts')
time_started = datetime.datetime.utcfromtimestamp(
attempt_start_ts / 1000000)
else: # pragma: no cover
logging.warning('no attempt_start_ts')
continue
# For builds through Buildbucket, job['master'] is actually the bucket
# name. For buildbot-based builds, it just happens to be the same as the
# master name. For Luci-based builds, it is different from the master
# name, and the master name is set as a build property instead.
# https://chromium.googlesource.com/chromium/src/+/infra/config/cr-buildbucket.cfg#115
# So in either case, the "real" master name is in the build properties.
master = build_properties['mastername']
except (ValueError, KeyError):
logging.warning('Failed to parse build properties',
exc_info=True)
parsing_errors.increment_by(1)
continue
# At this point, only success or failure.
success = build_result.isResultSuccess(result)
patchset_builder_runs = get_patchset_builder_runs(
issue=issue, patchset=patchset, master=master, builder=builder)
build_run = BuildRun(parent=patchset_builder_runs.key,
buildnumber=buildnumber,
result=result,
time_started=time_started,
time_finished=timestamp)
previous_runs = BuildRun.query(
ancestor=patchset_builder_runs.key).fetch()
duplicate = False
for previous_run in previous_runs:
# We saw this build run already or there are multiple green runs,
# in which case we ignore subsequent ones to avoid showing failures
# multiple times.
if (previous_run.buildnumber == buildnumber) or \
(build_run.is_success and previous_run.is_success) :
duplicate = True
break
if duplicate:
continue
build_run.put()
for previous_run in previous_runs:
if previous_run.is_success == build_run.is_success:
continue
if success:
# We saw the flake and then the pass.
failure_run = previous_run
success_run = build_run
else:
# We saw the pass and then the failure. Could happen when fetching
# historical data, or for the bot_update step (patch can't be
# applied cleanly anymore).
failure_run = build_run
success_run = previous_run
logging_output.append(failure_run.key.parent().get().builder +
str(failure_run.buildnumber))
# Queue a task to fetch the error of this failure and create FlakyRun.
flakes_metric.increment_by(1)
taskqueue.add(queue_name='issue-updates',
url='/issues/create_flaky_run',
params={
'failure_run_key': failure_run.key.urlsafe(),
'success_run_key': success_run.key.urlsafe()
})
return logging_output
def _create_flakes(ts, tf, ts2, tf2):
p = PatchsetBuilderRuns(issue=123456,
patchset=1,
master='tryserver.test',
builder='test-builder').put()
br_f0 = BuildRun(parent=p,
buildnumber=0,
result=2,
time_started=ts2,
time_finished=tf2).put()
br_f1 = BuildRun(parent=p,
buildnumber=1,
result=2,
time_started=ts,
time_finished=tf).put()
br_s1 = BuildRun(parent=p,
buildnumber=2,
result=0,
time_started=ts,
time_finished=tf).put()
br_f2 = BuildRun(parent=p,
buildnumber=3,
result=4,
time_started=ts,
time_finished=tf2).put()
br_s2 = BuildRun(parent=p,
buildnumber=4,
result=0,
time_started=ts,
time_finished=tf2).put()
occ_key1 = FlakyRun(failure_run=br_f0,
success_run=br_s2,
failure_run_time_started=ts2,
failure_run_time_finished=tf2).put()
occ_key2 = FlakyRun(failure_run=br_f1,
success_run=br_s1,
failure_run_time_started=ts,
failure_run_time_finished=tf).put()
occ_key3 = FlakyRun(failure_run=br_f2,
success_run=br_s2,
failure_run_time_started=ts,
failure_run_time_finished=tf).put()
Flake(name='foo',
last_hour=True,
last_day=True,
last_week=True,
last_month=True).put()
Flake(name='bar',
last_hour=True,
last_day=True,
last_week=True,
last_month=True,
occurrences=[occ_key1, occ_key2]).put()
Flake(name='baz',
last_hour=True,
last_day=True,
last_week=True,
last_month=True,
occurrences=[occ_key3]).put()
Flake(name='zee',
last_hour=False,
last_day=False,
last_week=True,
last_month=False).put()
def parse_cq_data(json_data):
logging_output = []
for result in json_data['results']:
fields = result['fields']
if not 'action' in fields:
continue
action = fields['action']
if action != 'verifier_jobs_update':
continue
if fields['verifier'] != 'try job':
continue
job_states = fields['jobs']
for state in job_states:
# Just go by |result|.
#if state not in ['JOB_SUCCEEDED', 'JOB_FAILED', 'JOB_TIMED_OUT']:
# continue
for job in job_states[state]:
build_properties = job['build_properties']
if not build_properties:
continue
master = job['master']
builder = job['builder']
result = job['result']
timestamp = datetime.datetime.strptime(job['timestamp'],
'%Y-%m-%d %H:%M:%S.%f')
try:
buildnumber = get_int_value(build_properties, 'buildnumber')
issue = get_int_value(build_properties, 'issue')
patchset = get_int_value(build_properties, 'patchset')
except ValueError as e:
continue
if build_result.isResultPending(result):
continue
# At this point, only success or failure.
success = build_result.isResultSuccess(result)
patchset_builder_runs = get_patchset_builder_runs(issue=issue,
patchset=patchset,
master=master,
builder=builder)
build_run = BuildRun(parent=patchset_builder_runs.key,
buildnumber=buildnumber,
result=result,
time_finished=timestamp)
previous_runs = BuildRun.query(
ancestor=patchset_builder_runs.key).fetch()
duplicate = False
for previous_run in previous_runs:
# We saw this build run already or there are multiple green runs,
# in which case we ignore subsequent ones to avoid showing failures
# multiple times.
if (previous_run.buildnumber == buildnumber) or \
(build_run.is_success and previous_run.is_success) :
duplicate = True
break
if duplicate:
continue
build_run.put()
for previous_run in previous_runs:
if previous_run.is_success == build_run.is_success:
continue
if success:
# We saw the flake and then the pass.
flaky_run = FlakyRun(
failure_run=previous_run.key,
failure_run_time_finished=previous_run.time_finished,
success_run=build_run.key)
flaky_run.put()
logging_output.append(previous_run.key.parent().get().builder +
str(previous_run.buildnumber))
else:
# We saw the pass and then the failure. Could happen when fetching
# historical data.
flaky_run = FlakyRun(
failure_run=build_run.key,
failure_run_time_finished=build_run.time_finished,
success_run=previous_run.key)
flaky_run.put()
logging_output.append(build_run.key.parent().get().builder +
str(build_run.buildnumber))
# Queue a task to fetch the error of this failure.
deferred.defer(get_flaky_run_reason, flaky_run.key)
return logging_output