def test_update_comment_empty(self):
"""Basic test on a testcase with empty comments."""
data_handler.update_testcase_comment(
self.testcase, data_types.TaskState.STARTED, 'message')
self.assertEqual(
'[2019-01-01 00:00:00] bot: Progression task started: message.\n',
self.testcase.comments)
def _testcase_reproduces_in_revision(testcase,
testcase_file_path,
job_type,
revision,
update_metadata=False):
"""Test to see if a test case reproduces in the specified revision."""
build_manager.setup_build(revision)
if not build_manager.check_app_path():
raise errors.BuildSetupError(revision, job_type)
if testcase_manager.check_for_bad_build(job_type, revision):
log_message = 'Bad build at r%d. Skipping' % revision
testcase = data_handler.get_testcase_by_id(testcase.key.id())
data_handler.update_testcase_comment(testcase,
data_types.TaskState.WIP,
log_message)
raise errors.BadBuildError(revision, job_type)
test_timeout = environment.get_value('TEST_TIMEOUT', 10)
result = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=testcase.http_flag)
_log_output(revision, result)
if update_metadata:
_update_issue_metadata(testcase)
return result
def validate_regression_range(testcase, testcase_file_path, job_type,
revision_list, min_index):
"""Ensure that we found the correct min revision by testing earlier ones."""
earlier_revisions = revision_list[
min_index - EARLIER_REVISIONS_TO_CONSIDER_FOR_VALIDATION:min_index]
revision_count = min(len(earlier_revisions),
REVISIONS_TO_TEST_FOR_VALIDATION)
revisions_to_test = random.sample(earlier_revisions, revision_count)
for revision in revisions_to_test:
try:
if _testcase_reproduces_in_revision(testcase, testcase_file_path,
job_type, revision):
testcase = data_handler.get_testcase_by_id(testcase.key.id())
testcase.regression = 'NA'
error_message = (
'Low confidence in regression range. Test case crashes in '
'revision r%d but not later revision r%d' %
(revision, revision_list[min_index]))
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR, error_message)
return False
except errors.BadBuildError:
pass
return True
def execute_task(testcase_id, job_type):
"""Execute progression task."""
try:
find_fixed_range(testcase_id, job_type)
except errors.BuildSetupError as error:
# If we failed to setup a build, it is likely a bot error. We can retry
# the task in this case.
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = 'Build setup failed r%d' % error.revision
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
build_fail_wait = environment.get_value('FAIL_WAIT')
tasks.add_task('progression',
testcase_id,
job_type,
wait_time=build_fail_wait)
except errors.BadBuildError:
# Though bad builds when narrowing the range are recoverable, certain builds
# being marked as bad may be unrecoverable. Recoverable ones should not
# reach this point.
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = 'Unable to recover from bad build'
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
# If there is a fine grained bisection service available, request it. Both
# regression and fixed ranges are requested once. Regression is also requested
# here as the bisection service may require details that are not yet available
# (e.g. issue ID) at the time regress_task completes.
task_creation.request_bisection(testcase_id)
def execute_task(testcase_id, job_type):
"""Run regression task and handle potential errors."""
try:
find_regression_range(testcase_id, job_type)
except errors.BuildSetupError as error:
# If we failed to setup a build, it is likely a bot error. We can retry
# the task in this case.
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = 'Build setup failed r%d' % error.revision
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
build_fail_wait = environment.get_value('FAIL_WAIT')
tasks.add_task('regression',
testcase_id,
job_type,
wait_time=build_fail_wait)
except errors.BadBuildError:
# Though bad builds when narrowing the range are recoverable, certain builds
# being marked as bad may be unrecoverable. Recoverable ones should not
# reach this point.
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.regression = 'NA'
error_message = 'Unable to recover from bad build'
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
def _testcase_reproduces_in_revision(testcase,
testcase_file_path,
job_type,
revision,
should_log=True,
min_revision=None,
max_revision=None):
"""Test to see if a test case reproduces in the specified revision."""
if should_log:
log_message = 'Testing r%d' % revision
if min_revision is not None and max_revision is not None:
log_message += ' (current range %d:%d)' % (min_revision, max_revision)
testcase = data_handler.get_testcase_by_id(testcase.key.id())
data_handler.update_testcase_comment(testcase, data_types.TaskState.WIP,
log_message)
build_manager.setup_regular_build(revision)
app_path = environment.get_value('APP_PATH')
if not app_path:
raise errors.BuildSetupError(revision, job_type)
if tests.check_for_bad_build(job_type, revision):
log_message = 'Bad build at r%d. Skipping' % revision
testcase = data_handler.get_testcase_by_id(testcase.key.id())
data_handler.update_testcase_comment(testcase, data_types.TaskState.WIP,
log_message)
raise errors.BadBuildError(revision, job_type)
test_timeout = environment.get_value('TEST_TIMEOUT', 10)
result = tests.test_for_crash_with_retries(
testcase, testcase_file_path, test_timeout, http_flag=testcase.http_flag)
return result.is_crash()
def _check_fixed_for_custom_binary(testcase, job_type, testcase_file_path):
"""Simplified fixed check for test cases using custom binaries."""
revision = environment.get_value('APP_REVISION')
# Update comments to reflect bot information and clean up old comments.
testcase_id = testcase.key.id()
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
build_manager.setup_build()
if not build_manager.check_app_path():
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR,
'Build setup failed for custom binary')
build_fail_wait = environment.get_value('FAIL_WAIT')
tasks.add_task(
'progression', testcase_id, job_type, wait_time=build_fail_wait)
return
test_timeout = environment.get_value('TEST_TIMEOUT', 10)
result = testcase_manager.test_for_crash_with_retries(
testcase, testcase_file_path, test_timeout, http_flag=testcase.http_flag)
_log_output(revision, result)
# Re-fetch to finalize testcase updates in branches below.
testcase = data_handler.get_testcase_by_id(testcase.key.id())
# If this still crashes on the most recent build, it's not fixed. The task
# will be rescheduled by a cron job and re-attempted eventually.
if result.is_crash():
app_path = environment.get_value('APP_PATH')
command = testcase_manager.get_command_line_for_application(
testcase_file_path, app_path=app_path, needs_http=testcase.http_flag)
symbolized_crash_stacktrace = result.get_stacktrace(symbolized=True)
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
stacktrace = utils.get_crash_stacktrace_output(
command, symbolized_crash_stacktrace, unsymbolized_crash_stacktrace)
testcase.last_tested_crash_stacktrace = data_handler.filter_stacktrace(
stacktrace)
_update_completion_metadata(
testcase,
revision,
is_crash=True,
message='still crashes on latest custom build')
return
# Retry once on another bot to confirm our results and in case this bot is in
# a bad state which we didn't catch through our usual means.
if data_handler.is_first_retry_for_task(testcase, reset_after_retry=True):
tasks.add_task('progression', testcase_id, job_type)
_update_completion_metadata(testcase, revision)
return
# The bug is fixed.
testcase.fixed = 'Yes'
testcase.open = False
_update_completion_metadata(
testcase, revision, message='fixed on latest custom build')
def _skip_minimization(testcase, message, crash_result=None, command=None):
"""Skip minimization for a testcase."""
testcase.minimized_keys = testcase.fuzzed_keys
if crash_result:
_update_crash_result(testcase, crash_result, command)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
message)
create_additional_tasks(testcase)
def _update_completion_metadata(testcase,
revision,
is_crash=False,
message=None):
"""Update metadata the progression task completes."""
_clear_progression_pending(testcase)
testcase.set_metadata('last_tested_revision', revision, update_testcase=False)
if is_crash:
testcase.set_metadata(
'last_tested_crash_revision', revision, update_testcase=False)
testcase.set_metadata(
'last_tested_crash_time', utils.utcnow(), update_testcase=False)
data_handler.update_testcase_comment(testcase, data_types.TaskState.FINISHED,
message)
def test_update_comment_truncate(self):
"""Test truncating long comments."""
self.testcase.comments = '\n' * data_types.TESTCASE_COMMENTS_LENGTH_LIMIT
data_handler.update_testcase_comment(
self.testcase, data_types.TaskState.STARTED, 'message')
self.assertEqual(data_types.TESTCASE_COMMENTS_LENGTH_LIMIT,
len(self.testcase.comments))
expected_new = (
'[2019-01-01 00:00:00] bot: Progression task started: message.\n')
expected = (
'\n' * (data_types.TESTCASE_COMMENTS_LENGTH_LIMIT - len(expected_new)) +
expected_new)
self.assertEqual(expected, self.testcase.comments)
def mark_unreproducible_if_flaky(testcase, potentially_flaky):
"""Check to see if a test case appears to be flaky."""
task_name = environment.get_value('TASK_NAME')
# If this run does not suggest that we are flaky, clear the flag and assume
# that we are reproducible.
if not potentially_flaky:
testcase.set_metadata('potentially_flaky', False)
return
# If we have not been marked as potentially flaky in the past, don't mark
# mark the test case as unreproducible yet. It is now potentially flaky.
if not testcase.get_metadata('potentially_flaky'):
testcase.set_metadata('potentially_flaky', True)
# In this case, the current task will usually be in a state where it cannot
# be completed. Recreate it.
tasks.add_task(task_name, testcase.key.id(), testcase.job_type)
return
# At this point, this test case has been flagged as potentially flaky twice.
# It should be marked as unreproducible. Mark it as unreproducible, and set
# fields that cannot be populated accordingly.
if task_name == 'minimize' and not testcase.minimized_keys:
testcase.minimized_keys = 'NA'
if task_name in ['minimize', 'impact']:
testcase.set_impacts_as_na()
if task_name in ['minimize', 'regression']:
testcase.regression = 'NA'
if task_name in ['minimize', 'progression']:
testcase.fixed = 'NA'
testcase.one_time_crasher_flag = True
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
'Testcase appears to be flaky')
issue = data_handler.get_issue_for_testcase(testcase)
if issue and issue.has_label('Reproducible'):
issue.remove_label('Reproducible')
issue.add_label('Unreproducible')
issue.comment = ('ClusterFuzz testcase %d appears to be flaky, '
'updating reproducibility label.' % testcase.key.id())
issue.save()
# For unreproducible testcases, it is still beneficial to get second stack
# information from stack task and component information from blame task.
create_blame_task_if_needed(testcase)
create_stack_task_if_needed(testcase)
def save_regression_range(testcase_id, regression_range_start,
regression_range_end):
"""Saves the regression range and creates blame and impact task if needed."""
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.regression = '%d:%d' % (regression_range_start, regression_range_end)
data_handler.update_testcase_comment(testcase, data_types.TaskState.FINISHED)
write_to_big_query(testcase, regression_range_start, regression_range_end)
# Force impacts update after regression range is updated. In several cases,
# we might not have a production build to test with, so regression range is
# used to decide impacts.
task_creation.create_impact_task_if_needed(testcase)
# Get blame information using the regression range result.
task_creation.create_blame_task_if_needed(testcase)
def test_update_comment_clear(self):
"""Basic test on a testcase with existing comments, and clearing old
progression messages."""
self.testcase.comments = (
'[2018-01-01 00:00:00] bot: Foo.\n'
'[2018-01-01 00:00:00] bot: Progression task started: message.\n'
'[2018-01-01 00:00:00] bot: Bar.\n'
'[2018-01-01 00:00:00] bot: Progression task finished.\n'
'[2018-01-01 00:00:00] bot: Blah.\n')
data_handler.update_testcase_comment(
self.testcase, data_types.TaskState.STARTED, 'message')
self.assertEqual(
('[2018-01-01 00:00:00] bot: Foo.\n'
'[2018-01-01 00:00:00] bot: Bar.\n'
'[2018-01-01 00:00:00] bot: Blah.\n'
'[2019-01-01 00:00:00] bot: Progression task started: message.\n'),
self.testcase.comments)
def mark_unreproducible_if_flaky(testcase, potentially_flaky):
"""Check to see if a test case appears to be flaky."""
task_name = environment.get_value('TASK_NAME')
# If this run does not suggest that we are flaky, clear the flag and assume
# that we are reproducible.
if not potentially_flaky:
testcase.set_metadata('potentially_flaky', False)
return
# If we have not been marked as potentially flaky in the past, don't mark
# mark the test case as unreproducible yet. It is now potentially flaky.
if not testcase.get_metadata('potentially_flaky'):
testcase.set_metadata('potentially_flaky', True)
# In this case, the current task will usually be in a state where it cannot
# be completed. Recreate it.
tasks.add_task(task_name, testcase.key.id(), testcase.job_type)
return
# At this point, this test case has been flagged as potentially flaky twice.
# It should be marked as unreproducible. Mark it as unreproducible, and set
# fields that cannot be populated accordingly.
if task_name == 'minimize' and not testcase.minimized_keys:
testcase.minimized_keys = 'NA'
if task_name in ['minimize', 'impact']:
testcase.set_impacts_as_na()
if task_name in ['minimize', 'regression']:
testcase.regression = 'NA'
if task_name in ['minimize', 'progression']:
testcase.fixed = 'NA'
testcase.one_time_crasher_flag = True
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
'Testcase appears to be flaky')
# Issue update to flip reproducibility label is done in App Engine cleanup
# cron. This avoids calling the issue tracker apis from GCE.
# For unreproducible testcases, it is still beneficial to get component
# information from blame task.
create_blame_task_if_needed(testcase)
# Let bisection service know about flakiness.
request_bisection(testcase.key.id())
def execute_task(testcase_id, _):
"""Attempt to find the CL introducing the bug associated with testcase_id."""
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
# Make sure that predator topic is configured. If not, nothing to do here.
topic = db_config.get_value('predator_crash_topic')
if not topic:
logs.log('Predator is not configured, skipping blame task.')
return
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
# Prepare pubsub message to send to predator.
message = _prepare_predator_message(testcase)
if not message:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR,
'Failed to generate request for Predator.')
return
# Clear existing results and mark blame result as pending.
testcase = data_handler.get_testcase_by_id(testcase_id)
_clear_blame_result_and_set_pending_flag(testcase)
# Post request to pub sub.
client = pubsub.PubSubClient()
message_ids = client.publish(topic, [message])
logs.log('Successfully published testcase %s to Predator. Message IDs: %s.' %
(testcase_id, message_ids))
data_handler.update_testcase_comment(testcase, data_types.TaskState.FINISHED)
def finalize_testcase(testcase_id,
command,
last_crash_result,
flaky_stack=False):
"""Perform final updates on a test case and prepare it for other tasks."""
# Symbolize crash output if we have it.
testcase = data_handler.get_testcase_by_id(testcase_id)
if last_crash_result:
_update_crash_result(testcase, last_crash_result, command)
# Update remaining test case information.
testcase.flaky_stack = flaky_stack
if build_manager.is_custom_binary():
testcase.set_impacts_as_na()
testcase.regression = 'NA'
data_handler.update_testcase_comment(testcase, data_types.TaskState.FINISHED)
# We might have updated the crash state. See if we need to marked as duplicate
# based on other testcases.
data_handler.handle_duplicate_entry(testcase)
create_additional_tasks(testcase)
def setup_testcase(testcase, fuzzer_override=None):
"""Sets up the testcase and needed dependencies like fuzzer,
data bundle, etc."""
fuzzer_name = fuzzer_override or testcase.fuzzer_name
job_type = testcase.job_type
task_name = environment.get_value('TASK_NAME')
testcase_fail_wait = environment.get_value('FAIL_WAIT')
testcase_id = testcase.key.id()
# Clear testcase directories.
shell.clear_testcase_directories()
# Adjust the test timeout value if this is coming from an user uploaded
# testcase.
_set_timeout_value_from_user_upload(testcase_id)
# Update the fuzzer if necessary in order to get the updated data bundle.
if fuzzer_name:
try:
update_successful = update_fuzzer_and_data_bundles(fuzzer_name)
except errors.InvalidFuzzerError:
# Close testcase and don't recreate tasks if this fuzzer is invalid.
testcase.open = False
testcase.fixed = 'NA'
testcase.set_metadata('fuzzer_was_deleted', True)
logs.log_error('Closed testcase %d with invalid fuzzer %s.' %
(testcase_id, fuzzer_name))
error_message = 'Fuzzer %s no longer exists' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
return None, None, None
if not update_successful:
error_message = 'Unable to setup fuzzer %s' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# Extract the testcase and any of its resources to the input directory.
file_list, input_directory, testcase_file_path = unpack_testcase(testcase)
if not file_list:
error_message = 'Unable to setup testcase %s' % testcase_file_path
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# For Android/Fuchsia, we need to sync our local testcases directory with the
# one on the device.
if environment.platform() == 'ANDROID':
_copy_testcase_to_device_and_setup_environment(testcase,
testcase_file_path)
# Push testcases to worker.
if environment.is_trusted_host():
from bot.untrusted_runner import file_host
file_host.push_testcases_to_worker()
# Copy global blacklist into local blacklist.
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Get local blacklist without this testcase's entry.
leak_blacklist.copy_global_to_local_blacklist(
excluded_testcase=testcase)
prepare_environment_for_testcase(testcase)
return file_list, input_directory, testcase_file_path
def execute_task(testcase_id, job_type):
"""Run a test case with a second job type to generate a second stack trace."""
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.STARTED)
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
return
# Initialize timeout values.
test_timeout = environment.get_value('TEST_TIMEOUT', 10)
# Set up a custom or regular build. We explicitly omit the crash revision
# since we want to test against the latest build here.
build_manager.setup_build()
# Check if we have an application path. If not, our build failed to setup
# correctly.
app_path = environment.get_value('APP_PATH')
if not app_path:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
'Build setup failed')
return
# TSAN tool settings (if the tool is used).
if environment.tool_matches('TSAN', job_type):
environment.set_tsan_max_history_size()
command = tests.get_command_line_for_application(
testcase_file_path, app_path=app_path, needs_http=testcase.http_flag)
result = tests.test_for_crash_with_retries(testcase,
testcase_file_path,
test_timeout,
http_flag=testcase.http_flag,
compare_crash=False)
# Get revision information.
revision = environment.get_value('APP_REVISION')
# If a crash occurs, then we add the second stacktrace information.
if result.is_crash():
state = result.get_symbolized_data()
security_flag = result.is_security_issue()
one_time_crasher_flag = not tests.test_for_reproducibility(
testcase_file_path, state.crash_state, security_flag, test_timeout,
testcase.http_flag, testcase.gestures)
# Attach a header to indicate information on reproducibility flag.
if one_time_crasher_flag:
crash_stacktrace_header = 'Unreliable'
else:
crash_stacktrace_header = 'Fully reproducible'
crash_stacktrace_header += (' crash found using %s job.\n\n' %
job_type)
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
stacktrace = utils.get_crash_stacktrace_output(
command, state.crash_stacktrace, unsymbolized_crash_stacktrace)
crash_stacktrace = data_handler.filter_stacktrace(
'%s%s' % (crash_stacktrace_header, stacktrace))
else:
crash_stacktrace = 'No crash found using %s job.' % job_type
# Decide which stacktrace to update this stacktrace with.
testcase = data_handler.get_testcase_by_id(testcase_id)
if testcase.last_tested_crash_stacktrace == 'Pending':
# This case happens when someone clicks 'Update last tested stacktrace using
# trunk build' button.
testcase.last_tested_crash_stacktrace = crash_stacktrace
testcase.set_metadata('last_tested_crash_revision',
revision,
update_testcase=False)
else:
# Default case when someone defines |SECOND_STACK_JOB_TYPE| in the job
# type. This helps to test the unreproducible crash with a different memory
# debugging tool to get a second stacktrace (e.g. running TSAN on a flaky
# crash found in ASAN build).
testcase.second_crash_stacktrace = crash_stacktrace
testcase.set_metadata('second_crash_stacktrace_revision',
revision,
update_testcase=False)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED)
def find_regression_range(testcase_id, job_type):
"""Attempt to find when the testcase regressed."""
deadline = tasks.get_task_completion_deadline()
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
if testcase.regression:
logs.log_error('Regression range is already set as %s, skip.' %
testcase.regression)
return
# This task is not applicable for custom binaries.
if build_manager.is_custom_binary():
testcase.regression = 'NA'
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR,
'Not applicable for custom binaries')
return
data_handler.update_testcase_comment(testcase,
data_types.TaskState.STARTED)
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
'Failed to setup testcase')
tasks.add_task('regression', testcase_id, job_type)
return
release_build_bucket_path = environment.get_value(
'RELEASE_BUILD_BUCKET_PATH')
revision_list = build_manager.get_revisions_list(release_build_bucket_path,
testcase=testcase)
if not revision_list:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
'Failed to fetch revision list')
tasks.add_task('regression', testcase_id, job_type)
return
# Don't burden NFS server with caching these random builds.
environment.set_value('CACHE_STORE', False)
# Pick up where left off in a previous run if necessary.
min_revision = testcase.get_metadata('last_regression_min')
max_revision = testcase.get_metadata('last_regression_max')
first_run = not min_revision and not max_revision
if not min_revision:
min_revision = revisions.get_first_revision_in_list(revision_list)
if not max_revision:
max_revision = testcase.crash_revision
min_index = revisions.find_min_revision_index(revision_list, min_revision)
if min_index is None:
raise errors.BuildNotFoundError(min_revision, job_type)
max_index = revisions.find_max_revision_index(revision_list, max_revision)
if max_index is None:
raise errors.BuildNotFoundError(max_revision, job_type)
# Make sure that the revision where we noticed the crash, still crashes at
# that revision. Otherwise, our binary search algorithm won't work correctly.
max_revision = revision_list[max_index]
crashes_in_max_revision = _testcase_reproduces_in_revision(
testcase, testcase_file_path, job_type, max_revision, should_log=False)
if not crashes_in_max_revision:
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = ('Known crash revision %d did not crash' %
max_revision)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
task_creation.mark_unreproducible_if_flaky(testcase, True)
return
# If we've made it this far, the test case appears to be reproducible. Clear
# metadata from previous runs had it been marked as potentially flaky.
task_creation.mark_unreproducible_if_flaky(testcase, False)
# On the first run, check to see if we regressed near either the min or max
# revision.
if first_run and found_regression_near_extreme_revisions(
testcase, testcase_file_path, job_type, revision_list, min_index,
max_index):
return
while time.time() < deadline:
min_revision = revision_list[min_index]
max_revision = revision_list[max_index]
# If the min and max revisions are one apart (or the same, if we only have
# one build), this is as much as we can narrow the range.
if max_index - min_index <= 1:
# Verify that the regression range seems correct, and save it if so.
if not validate_regression_range(testcase, testcase_file_path,
job_type, revision_list,
min_index):
return
save_regression_range(testcase_id, min_revision, max_revision)
return
middle_index = (min_index + max_index) / 2
middle_revision = revision_list[middle_index]
try:
is_crash = _testcase_reproduces_in_revision(
testcase,
testcase_file_path,
job_type,
middle_revision,
min_revision=min_revision,
max_revision=max_revision)
except errors.BadBuildError:
# Skip this revision.
del revision_list[middle_index]
max_index -= 1
continue
if is_crash:
max_index = middle_index
else:
min_index = middle_index
_save_current_regression_range_indices(testcase_id,
revision_list[min_index],
revision_list[max_index])
# If we've broken out of the above loop, we timed out. We'll finish by
# running another regression task and picking up from this point.
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = 'Timed out, current range r%d:r%d' % (
revision_list[min_index], revision_list[max_index])
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
error_message)
tasks.add_task('regression', testcase_id, job_type)
def execute_task(testcase_id, job_type):
"""Attempt to minimize a given testcase."""
# Get deadline to finish this task.
deadline = tasks.get_task_completion_deadline()
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
# Update comments to reflect bot information.
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
# Setup testcase and its dependencies.
file_list, input_directory, testcase_file_path = setup.setup_testcase(
testcase)
if not file_list:
return
# Initialize variables.
max_timeout = environment.get_value('TEST_TIMEOUT', 10)
app_arguments = environment.get_value('APP_ARGS')
# Set up a custom or regular build based on revision.
last_tested_crash_revision = testcase.get_metadata(
'last_tested_crash_revision')
crash_revision = last_tested_crash_revision or testcase.crash_revision
build_manager.setup_build(crash_revision)
# Check if we have an application path. If not, our build failed
# to setup correctly.
app_path = environment.get_value('APP_PATH')
if not app_path:
logs.log_error('Unable to setup build for minimization.')
build_fail_wait = environment.get_value('FAIL_WAIT')
if environment.get_value('ORIGINAL_JOB_NAME'):
_skip_minimization(testcase, 'Failed to setup build for overridden job.')
else:
# Only recreate task if this isn't an overriden job. It's possible that a
# revision exists for the original job, but doesn't exist for the
# overriden job.
tasks.add_task(
'minimize', testcase_id, job_type, wait_time=build_fail_wait)
return
if environment.is_libfuzzer_job():
do_libfuzzer_minimization(testcase, testcase_file_path)
return
max_threads = utils.maximum_parallel_processes_allowed()
# Prepare the test case runner.
crash_retries = environment.get_value('CRASH_RETRIES')
warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
required_arguments = environment.get_value('REQUIRED_APP_ARGS', '')
# Add any testcase-specific required arguments if needed.
additional_required_arguments = testcase.get_metadata(
'additional_required_app_args')
if additional_required_arguments:
required_arguments = '%s %s' % (required_arguments,
additional_required_arguments)
test_runner = TestRunner(testcase, testcase_file_path, file_list,
input_directory, app_arguments, required_arguments,
max_threads, deadline)
# Verify the crash with a long timeout.
warmup_crash_occurred = False
result = test_runner.run(timeout=warmup_timeout, log_command=True)
if result.is_crash():
warmup_crash_occurred = True
logs.log('Warmup crash occurred in %d seconds.' % result.crash_time)
saved_unsymbolized_crash_state, flaky_stack, crash_times = (
check_for_initial_crash(test_runner, crash_retries, testcase))
# If the warmup crash occurred but we couldn't reproduce this in with
# multiple processes running in parallel, try to minimize single threaded.
if (len(crash_times) < tests.REPRODUCIBILITY_FACTOR * crash_retries and
warmup_crash_occurred and max_threads > 1):
logs.log('Attempting to continue single-threaded.')
max_threads = 1
test_runner = TestRunner(testcase, testcase_file_path, file_list,
input_directory, app_arguments, required_arguments,
max_threads, deadline)
saved_unsymbolized_crash_state, flaky_stack, crash_times = (
check_for_initial_crash(test_runner, crash_retries, testcase))
if flaky_stack:
testcase.flaky_stack = flaky_stack
testcase.put()
if len(crash_times) < tests.REPRODUCIBILITY_FACTOR * crash_retries:
if not crash_times:
# We didn't crash at all, so try again. This might be a legitimately
# unreproducible test case, so it will get marked as such after being
# retried on other bots.
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
'Unable to reproduce crash')
task_creation.mark_unreproducible_if_flaky(testcase, True)
else:
# We reproduced this crash at least once. It's too flaky to minimize, but
# maybe we'll have more luck in the other jobs.
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.minimized_keys = 'NA'
error_message = (
'Unable to reproduce crash reliably, skipping '
'minimization (crashed %d/%d)' % (len(crash_times), crash_retries))
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
error_message)
create_additional_tasks(testcase)
return
# If we've made it this far, the test case appears to be reproducible. Clear
# metadata from previous runs had it been marked as potentially flaky.
task_creation.mark_unreproducible_if_flaky(testcase, False)
test_runner.set_test_expectations(testcase.security_flag, flaky_stack,
saved_unsymbolized_crash_state)
# Use the max crash time unless this would be greater than the max timeout.
test_timeout = min(max(crash_times), max_timeout) + 1
logs.log('Using timeout %d (was %d)' % (test_timeout, max_timeout))
test_runner.timeout = test_timeout
logs.log('Starting minimization.')
if should_attempt_phase(testcase, MinimizationPhase.GESTURES):
gestures = minimize_gestures(test_runner, testcase)
# We can't call check_deadline_exceeded_and_store_partial_minimized_testcase
# at this point because we do not have a test case to store.
testcase = data_handler.get_testcase_by_id(testcase.key.id())
if testcase.security_flag and len(testcase.gestures) != len(gestures):
# Re-run security severity analysis since gestures affect the severity.
testcase.security_severity = severity_analyzer.get_security_severity(
testcase.crash_type, data_handler.get_stacktrace(testcase), job_type,
bool(gestures))
testcase.gestures = gestures
testcase.set_metadata('minimization_phase', MinimizationPhase.MAIN_FILE)
if time.time() > test_runner.deadline:
tasks.add_task('minimize', testcase.key.id(), job_type)
return
# Minimize the main file.
data = utils.get_file_contents_with_fatal_error_on_failure(testcase_file_path)
if should_attempt_phase(testcase, MinimizationPhase.MAIN_FILE):
data = minimize_main_file(test_runner, testcase_file_path, data)
if check_deadline_exceeded_and_store_partial_minimized_testcase(
deadline, testcase_id, job_type, input_directory, file_list, data,
testcase_file_path):
return
testcase.set_metadata('minimization_phase', MinimizationPhase.FILE_LIST)
# Minimize the file list.
if should_attempt_phase(testcase, MinimizationPhase.FILE_LIST):
if environment.get_value('MINIMIZE_FILE_LIST', True):
file_list = minimize_file_list(test_runner, file_list, input_directory,
testcase_file_path)
if check_deadline_exceeded_and_store_partial_minimized_testcase(
deadline, testcase_id, job_type, input_directory, file_list, data,
testcase_file_path):
return
else:
logs.log('Skipping minimization of file list.')
testcase.set_metadata('minimization_phase', MinimizationPhase.RESOURCES)
# Minimize any files remaining in the file list.
if should_attempt_phase(testcase, MinimizationPhase.RESOURCES):
if environment.get_value('MINIMIZE_RESOURCES', True):
for dependency in file_list:
minimize_resource(test_runner, dependency, input_directory,
testcase_file_path)
if check_deadline_exceeded_and_store_partial_minimized_testcase(
deadline, testcase_id, job_type, input_directory, file_list, data,
testcase_file_path):
return
else:
logs.log('Skipping minimization of resources.')
testcase.set_metadata('minimization_phase', MinimizationPhase.ARGUMENTS)
if should_attempt_phase(testcase, MinimizationPhase.ARGUMENTS):
app_arguments = minimize_arguments(test_runner, app_arguments)
# Arguments must be stored here in case we time out below.
testcase.minimized_arguments = app_arguments
testcase.put()
if check_deadline_exceeded_and_store_partial_minimized_testcase(
deadline, testcase_id, job_type, input_directory, file_list, data,
testcase_file_path):
return
command = tests.get_command_line_for_application(
testcase_file_path, app_args=app_arguments, needs_http=testcase.http_flag)
last_crash_result = test_runner.last_failing_result
store_minimized_testcase(testcase, input_directory, file_list, data,
testcase_file_path)
finalize_testcase(
testcase_id, command, last_crash_result, flaky_stack=flaky_stack)
def execute_task(testcase_id, job_type):
"""Run analyze task."""
# Reset redzones.
environment.reset_current_memory_tool_options(redzone_size=128)
# Unset window location size and position properties so as to use default.
environment.set_value('WINDOW_ARG', '')
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
metadata = data_types.TestcaseUploadMetadata.query(
data_types.TestcaseUploadMetadata.testcase_id == int(testcase_id)).get()
if not metadata:
logs.log_error(
'Testcase %s has no associated upload metadata.' % testcase_id)
testcase.key.delete()
return
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Creates empty local blacklist so all leaks will be visible to uploader.
leak_blacklist.create_empty_local_blacklist()
# Store the bot name and timestamp in upload metadata.
bot_name = environment.get_value('BOT_NAME')
metadata.bot_name = bot_name
metadata.timestamp = datetime.datetime.utcnow()
metadata.put()
# Adjust the test timeout, if user has provided one.
if metadata.timeout:
environment.set_value('TEST_TIMEOUT', metadata.timeout)
# Adjust the number of retries, if user has provided one.
if metadata.retries is not None:
environment.set_value('CRASH_RETRIES', metadata.retries)
# Setup testcase and get absolute testcase path.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
return
# Set up a custom or regular build based on revision.
build_manager.setup_build(testcase.crash_revision)
# Check if we have an application path. If not, our build failed
# to setup correctly.
app_path = environment.get_value('APP_PATH')
if not app_path:
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
'Build setup failed')
if data_handler.is_first_retry_for_task(testcase):
build_fail_wait = environment.get_value('FAIL_WAIT')
tasks.add_task(
'analyze', testcase_id, job_type, wait_time=build_fail_wait)
else:
close_invalid_testcase_and_update_status(testcase, metadata,
'Build setup failed')
return
# Update initial testcase information.
testcase.absolute_path = testcase_file_path
testcase.job_type = job_type
testcase.binary_flag = utils.is_binary_file(testcase_file_path)
testcase.queue = tasks.default_queue()
testcase.crash_state = ''
# Set initial testcase metadata fields (e.g. build url, etc).
data_handler.set_initial_testcase_metadata(testcase)
# Update minimized arguments and use ones provided during user upload.
if not testcase.minimized_arguments:
minimized_arguments = environment.get_value('APP_ARGS') or ''
additional_command_line_flags = testcase.get_metadata(
'uploaded_additional_args')
if additional_command_line_flags:
minimized_arguments += ' %s' % additional_command_line_flags
environment.set_value('APP_ARGS', minimized_arguments)
testcase.minimized_arguments = minimized_arguments
# Update other fields not set at upload time.
testcase.crash_revision = environment.get_value('APP_REVISION')
data_handler.set_initial_testcase_metadata(testcase)
testcase.put()
# Initialize some variables.
gestures = testcase.gestures
http_flag = testcase.http_flag
test_timeout = environment.get_value('TEST_TIMEOUT')
# Get the crash output.
result = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=http_flag,
compare_crash=False)
# If we don't get a crash, try enabling http to see if we can get a crash.
# Skip engine fuzzer jobs (e.g. libFuzzer, AFL) for which http testcase paths
# are not applicable.
if (not result.is_crash() and not http_flag and
not environment.is_engine_fuzzer_job()):
result_with_http = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=True,
compare_crash=False)
if result_with_http.is_crash():
logs.log('Testcase needs http flag for crash.')
http_flag = True
result = result_with_http
# Refresh our object.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
# Set application command line with the correct http flag.
application_command_line = (
testcase_manager.get_command_line_for_application(
testcase_file_path, needs_http=http_flag))
# Get the crash data.
crashed = result.is_crash()
crash_time = result.get_crash_time()
state = result.get_symbolized_data()
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
# Get crash info object with minidump info. Also, re-generate unsymbolized
# stacktrace if needed.
crash_info, _ = (
crash_uploader.get_crash_info_and_stacktrace(
application_command_line, state.crash_stacktrace, gestures))
if crash_info:
testcase.minidump_keys = crash_info.store_minidump()
if not crashed:
# Could not reproduce the crash.
log_message = (
'Testcase didn\'t crash in %d seconds (with retries)' % test_timeout)
data_handler.update_testcase_comment(
testcase, data_types.TaskState.FINISHED, log_message)
# For an unreproducible testcase, retry once on another bot to confirm
# our results and in case this bot is in a bad state which we didn't catch
# through our usual means.
if data_handler.is_first_retry_for_task(testcase):
testcase.status = 'Unreproducible, retrying'
testcase.put()
tasks.add_task('analyze', testcase_id, job_type)
return
# In the general case, we will not attempt to symbolize if we do not detect
# a crash. For user uploads, we should symbolize anyway to provide more
# information about what might be happening.
crash_stacktrace_output = utils.get_crash_stacktrace_output(
application_command_line, state.crash_stacktrace,
unsymbolized_crash_stacktrace)
testcase.crash_stacktrace = data_handler.filter_stacktrace(
crash_stacktrace_output)
close_invalid_testcase_and_update_status(testcase, metadata,
'Unreproducible')
# A non-reproducing testcase might still impact production branches.
# Add the impact task to get that information.
task_creation.create_impact_task_if_needed(testcase)
return
# Update http flag and re-run testcase to store dependencies (for bundled
# archives only).
testcase.http_flag = http_flag
if not store_testcase_dependencies_from_bundled_testcase_archive(
metadata, testcase, testcase_file_path):
return
# Update testcase crash parameters.
testcase.crash_type = state.crash_type
testcase.crash_address = state.crash_address
testcase.crash_state = state.crash_state
# Try to guess if the bug is security or not.
security_flag = crash_analyzer.is_security_issue(
state.crash_stacktrace, state.crash_type, state.crash_address)
testcase.security_flag = security_flag
# If it is, guess the severity.
if security_flag:
testcase.security_severity = severity_analyzer.get_security_severity(
state.crash_type, state.crash_stacktrace, job_type, bool(gestures))
log_message = ('Testcase crashed in %d seconds (r%d)' %
(crash_time, testcase.crash_revision))
data_handler.update_testcase_comment(testcase, data_types.TaskState.FINISHED,
log_message)
# See if we have to ignore this crash.
if crash_analyzer.ignore_stacktrace(state.crash_stacktrace):
close_invalid_testcase_and_update_status(testcase, metadata, 'Irrelavant')
return
# Test for reproducibility.
one_time_crasher_flag = not testcase_manager.test_for_reproducibility(
testcase_file_path, state.crash_state, security_flag, test_timeout,
http_flag, gestures)
testcase.one_time_crasher_flag = one_time_crasher_flag
# Check to see if this is a duplicate.
project_name = data_handler.get_project_name(job_type)
existing_testcase = data_handler.find_testcase(
project_name, state.crash_type, state.crash_state, security_flag)
if existing_testcase:
# If the existing test case is unreproducible and we are, replace the
# existing test case with this one.
if existing_testcase.one_time_crasher_flag and not one_time_crasher_flag:
duplicate_testcase = existing_testcase
original_testcase = testcase
else:
duplicate_testcase = testcase
original_testcase = existing_testcase
metadata.status = 'Duplicate'
metadata.duplicate_of = existing_testcase.key.id()
duplicate_testcase.status = 'Duplicate'
duplicate_testcase.duplicate_of = original_testcase.key.id()
duplicate_testcase.put()
# Set testcase and metadata status if not set already.
if testcase.status != 'Duplicate':
testcase.status = 'Processed'
metadata.status = 'Confirmed'
# Add new leaks to global blacklist to avoid detecting duplicates.
# Only add if testcase has a direct leak crash and if it's reproducible.
if is_lsan_enabled:
leak_blacklist.add_crash_to_global_blacklist_if_needed(testcase)
# Add application specific information in the trace.
crash_stacktrace_output = utils.get_crash_stacktrace_output(
application_command_line, state.crash_stacktrace,
unsymbolized_crash_stacktrace)
testcase.crash_stacktrace = data_handler.filter_stacktrace(
crash_stacktrace_output)
# Update the testcase values.
testcase.put()
# Update the upload metadata.
metadata.security_flag = security_flag
metadata.put()
# Create tasks to
# 1. Minimize testcase (minimize).
# 2. Find regression range (regression).
# 3. Find testcase impact on production branches (impact).
# 4. Check whether testcase is fixed (progression).
# 5. Get second stacktrace from another job in case of
# one-time crashers (stack).
task_creation.create_tasks(testcase)
def execute_task(testcase_id, job_type):
"""Attempt to find if the testcase affects release branches on Chromium."""
# This shouldn't ever get scheduled, but check just in case.
if not utils.is_chromium():
return
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
# If this testcase is fixed, we should no longer be doing impact testing.
if testcase.fixed and testcase.is_impact_set_flag:
return
# For testcases with status unreproducible, we just do impact analysis just
# once.
if testcase.is_status_unreproducible() and testcase.is_impact_set_flag:
return
# Update comments only after checking the above bailout conditions.
data_handler.update_testcase_comment(testcase,
data_types.TaskState.STARTED)
# This task is not applicable to unreproducible testcases.
if testcase.one_time_crasher_flag:
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR,
'Not applicable for unreproducible testcases')
return
# This task is not applicable for custom binaries. We cannot remove the
# creation of such tasks specifically for custom binary testcase in cron,
# so exit gracefully.
if build_manager.is_custom_binary():
data_handler.update_testcase_comment(
testcase, data_types.TaskState.FINISHED,
'Not applicable for custom binaries')
return
# If we don't have a stable or beta build url pattern, we try to use build
# information url to make a guess.
if not build_manager.has_production_builds():
if not testcase.regression:
data_handler.update_testcase_comment(
testcase, data_types.TaskState.FINISHED,
'Cannot run without regression range, will re-run once regression '
'task finishes')
return
impacts = get_impacts_from_url(testcase.regression, testcase.job_type)
testcase = data_handler.get_testcase_by_id(testcase_id)
set_testcase_with_impacts(testcase, impacts)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED)
return
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
return
# Setup stable, beta builds and get impact and crash stacktrace.
try:
impacts = get_impacts_on_prod_builds(testcase, testcase_file_path)
except BuildFailedException as error:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error.message)
tasks.add_task('impact',
testcase_id,
job_type,
wait_time=environment.get_value('FAIL_WAIT'))
return
testcase = data_handler.get_testcase_by_id(testcase_id)
set_testcase_with_impacts(testcase, impacts)
# Set stacktrace in case we have a unreproducible crash on trunk,
# but it crashes on one of the production builds.
if testcase.is_status_unreproducible() and impacts.get_extra_trace():
testcase.crash_stacktrace = data_handler.filter_stacktrace(
'%s\n\n%s' %
(data_handler.get_stacktrace(testcase), impacts.get_extra_trace()))
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED)
def find_fixed_range(testcase_id, job_type):
"""Attempt to find the revision range where a testcase was fixed."""
deadline = tasks.get_task_completion_deadline()
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
if testcase.fixed:
logs.log_error('Fixed range is already set as %s, skip.' %
testcase.fixed)
return
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
return
# Set a flag to indicate we are running progression task. This shows pending
# status on testcase report page and avoid conflicting testcase updates by
# triage cron.
testcase.set_metadata('progression_pending', True)
# Custom binaries are handled as special cases.
if build_manager.is_custom_binary():
_check_fixed_for_custom_binary(testcase, job_type, testcase_file_path)
return
release_build_bucket_path = environment.get_value(
'RELEASE_BUILD_BUCKET_PATH')
revision_list = build_manager.get_revisions_list(release_build_bucket_path,
testcase=testcase)
if not revision_list:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
'Failed to fetch revision list')
tasks.add_task('progression', testcase_id, job_type)
return
# Use min, max_index to mark the start and end of revision list that is used
# for bisecting the progression range. Set start to the revision where noticed
# the crash. Set end to the trunk revision. Also, use min, max from past run
# if it timed out.
min_revision = testcase.get_metadata('last_progression_min')
max_revision = testcase.get_metadata('last_progression_max')
last_tested_revision = testcase.get_metadata('last_tested_crash_revision')
known_crash_revision = last_tested_revision or testcase.crash_revision
if not min_revision:
min_revision = known_crash_revision
if not max_revision:
max_revision = revisions.get_last_revision_in_list(revision_list)
min_index = revisions.find_min_revision_index(revision_list, min_revision)
if min_index is None:
raise errors.BuildNotFoundError(min_revision, job_type)
max_index = revisions.find_max_revision_index(revision_list, max_revision)
if max_index is None:
raise errors.BuildNotFoundError(max_revision, job_type)
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.STARTED,
'r%d' % max_revision)
# Check to see if this testcase is still crashing now. If it is, then just
# bail out.
result = _testcase_reproduces_in_revision(testcase, testcase_file_path,
job_type, max_revision)
if result.is_crash():
logs.log('Found crash with same signature on latest revision r%d.' %
max_revision)
app_path = environment.get_value('APP_PATH')
command = testcase_manager.get_command_line_for_application(
testcase_file_path,
app_path=app_path,
needs_http=testcase.http_flag)
symbolized_crash_stacktrace = result.get_stacktrace(symbolized=True)
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
stacktrace = utils.get_crash_stacktrace_output(
command, symbolized_crash_stacktrace,
unsymbolized_crash_stacktrace)
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.last_tested_crash_stacktrace = data_handler.filter_stacktrace(
stacktrace)
_update_completion_metadata(
testcase,
max_revision,
is_crash=True,
message='still crashes on latest revision r%s' % max_revision)
# Since we've verified that the test case is still crashing, clear out any
# metadata indicating potential flake from previous runs.
task_creation.mark_unreproducible_if_flaky(testcase, False)
# For chromium project, save latest crash information for later upload
# to chromecrash/.
state = result.get_symbolized_data()
crash_uploader.save_crash_info_if_needed(testcase_id, max_revision,
job_type, state.crash_type,
state.crash_address,
state.frames)
return
# Don't burden NFS server with caching these random builds.
environment.set_value('CACHE_STORE', False)
# Verify that we do crash in the min revision. This is assumed to be true
# while we are doing the bisect.
result = _testcase_reproduces_in_revision(testcase, testcase_file_path,
job_type, min_revision)
if result and not result.is_crash():
testcase = data_handler.get_testcase_by_id(testcase_id)
# Retry once on another bot to confirm our result.
if data_handler.is_first_retry_for_task(testcase,
reset_after_retry=True):
tasks.add_task('progression', testcase_id, job_type)
error_message = (
'Known crash revision %d did not crash, will retry on another bot to '
'confirm result' % known_crash_revision)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
_update_completion_metadata(testcase, max_revision)
return
_clear_progression_pending(testcase)
error_message = ('Known crash revision %d did not crash' %
known_crash_revision)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
task_creation.mark_unreproducible_if_flaky(testcase, True)
return
# Start a binary search to find last non-crashing revision. At this point, we
# know that we do crash in the min_revision, and do not crash in max_revision.
while time.time() < deadline:
min_revision = revision_list[min_index]
max_revision = revision_list[max_index]
# If the min and max revisions are one apart this is as much as we can
# narrow the range.
if max_index - min_index == 1:
_save_fixed_range(testcase_id, min_revision, max_revision)
return
# Test the middle revision of our range.
middle_index = (min_index + max_index) // 2
middle_revision = revision_list[middle_index]
testcase = data_handler.get_testcase_by_id(testcase_id)
log_message = 'Testing r%d (current range %d:%d)' % (
middle_revision, min_revision, max_revision)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.WIP,
log_message)
try:
result = _testcase_reproduces_in_revision(testcase,
testcase_file_path,
job_type,
middle_revision)
except errors.BadBuildError:
# Skip this revision.
del revision_list[middle_index]
max_index -= 1
continue
if result.is_crash():
min_index = middle_index
else:
max_index = middle_index
_save_current_fixed_range_indices(testcase_id,
revision_list[min_index],
revision_list[max_index])
# If we've broken out of the loop, we've exceeded the deadline. Recreate the
# task to pick up where we left off.
testcase = data_handler.get_testcase_by_id(testcase_id)
error_message = ('Timed out, current range r%d:r%d' %
(revision_list[min_index], revision_list[max_index]))
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
error_message)
tasks.add_task('progression', testcase_id, job_type)
def execute_task(testcase_id, job_type):
"""Run a test case with a different job type to see if they reproduce."""
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
if environment.is_engine_fuzzer_job(
testcase.job_type) != environment.is_engine_fuzzer_job(job_type):
# We should never reach here. But in case we do, we should bail out as
# otherwise we will run into exceptions.
return
# Use a cloned testcase entity with different fuzz target paramaters for
# a different fuzzing engine.
original_job_type = testcase.job_type
testcase = _get_variant_testcase_for_job(testcase, job_type)
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase, job_type)
if not file_list:
return
# Set up a custom or regular build. We explicitly omit the crash revision
# since we want to test against the latest build here.
build_manager.setup_build()
# Check if we have an application path. If not, our build failed to setup
# correctly.
if not build_manager.check_app_path():
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(
testcase,
data_types.TaskState.ERROR,
"Build setup failed with job: " + job_type,
)
return
# Disable gestures if we're running on a different platform from that of
# the original test case.
use_gestures = testcase.platform == environment.platform().lower()
# Reproduce the crash.
app_path = environment.get_value("APP_PATH")
command = testcase_manager.get_command_line_for_application(
testcase_file_path, app_path=app_path, needs_http=testcase.http_flag)
test_timeout = environment.get_value("TEST_TIMEOUT", 10)
revision = environment.get_value("APP_REVISION")
result = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=testcase.http_flag,
use_gestures=use_gestures,
compare_crash=False,
)
if result.is_crash() and not result.should_ignore():
crash_state = result.get_state()
crash_type = result.get_type()
security_flag = result.is_security_issue()
gestures = testcase.gestures if use_gestures else None
one_time_crasher_flag = not testcase_manager.test_for_reproducibility(
testcase.fuzzer_name,
testcase.actual_fuzzer_name(),
testcase_file_path,
crash_state,
security_flag,
test_timeout,
testcase.http_flag,
gestures,
)
if one_time_crasher_flag:
status = data_types.TestcaseVariantStatus.FLAKY
else:
status = data_types.TestcaseVariantStatus.REPRODUCIBLE
crash_comparer = CrashComparer(crash_state, testcase.crash_state)
is_similar = (crash_comparer.is_similar()
and security_flag == testcase.security_flag)
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
symbolized_crash_stacktrace = result.get_stacktrace(symbolized=True)
crash_stacktrace_output = utils.get_crash_stacktrace_output(
command, symbolized_crash_stacktrace,
unsymbolized_crash_stacktrace)
else:
status = data_types.TestcaseVariantStatus.UNREPRODUCIBLE
is_similar = False
crash_type = None
crash_state = None
security_flag = False
crash_stacktrace_output = "No crash occurred."
if original_job_type == job_type:
# This case happens when someone clicks 'Update last tested stacktrace using
# trunk build' button.
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.last_tested_crash_stacktrace = data_handler.filter_stacktrace(
crash_stacktrace_output)
testcase.set_metadata("last_tested_crash_revision",
revision,
update_testcase=True)
else:
# Regular case of variant analysis.
variant = data_handler.get_testcase_variant(testcase_id, job_type)
variant.status = status
variant.revision = revision
variant.crash_type = crash_type
variant.crash_state = crash_state
variant.security_flag = security_flag
variant.is_similar = is_similar
# Explicitly skipping crash stacktrace for now as it make entities larger
# and we plan to use only crash paramaters in UI.
variant.put()
def execute_task(testcase_id, job_type):
"""Execute a symbolize command."""
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
# We should atleast have a symbolized debug or release build.
if not build_manager.has_symbolized_builds():
return
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase, job_type)
if not file_list:
return
# Initialize variables.
build_fail_wait = environment.get_value("FAIL_WAIT")
old_crash_stacktrace = data_handler.get_stacktrace(testcase)
sym_crash_type = testcase.crash_type
sym_crash_address = testcase.crash_address
sym_crash_state = testcase.crash_state
sym_redzone = DEFAULT_REDZONE
warmup_timeout = environment.get_value("WARMUP_TIMEOUT")
# Decide which build revision to use.
if testcase.crash_stacktrace == "Pending":
# This usually happen when someone clicked the 'Update stacktrace from
# trunk' button on the testcase details page. In this case, we are forced
# to use trunk. No revision -> trunk build.
build_revision = None
else:
build_revision = testcase.crash_revision
# Set up a custom or regular build based on revision.
build_manager.setup_build(build_revision)
# Get crash revision used in setting up build.
crash_revision = environment.get_value("APP_REVISION")
if not build_manager.check_app_path():
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
"Build setup failed")
tasks.add_task(
"symbolize", testcase_id, job_type, wait_time=build_fail_wait)
return
# ASAN tool settings (if the tool is used).
# See if we can get better stacks with higher redzone sizes.
# A UAF might actually turn out to be OOB read/write with a bigger redzone.
if environment.tool_matches("ASAN", job_type) and testcase.security_flag:
redzone = MAX_REDZONE
while redzone >= MIN_REDZONE:
environment.reset_current_memory_tool_options(
redzone_size=testcase.redzone, disable_ubsan=testcase.disable_ubsan)
process_handler.terminate_stale_application_instances()
command = testcase_manager.get_command_line_for_application(
testcase_file_path, needs_http=testcase.http_flag)
return_code, crash_time, output = process_handler.run_process(
command, timeout=warmup_timeout, gestures=testcase.gestures)
crash_result = CrashResult(return_code, crash_time, output)
if crash_result.is_crash() and "AddressSanitizer" in output:
state = crash_result.get_symbolized_data()
security_flag = crash_result.is_security_issue()
if (not crash_analyzer.ignore_stacktrace(state.crash_stacktrace) and
security_flag == testcase.security_flag and
state.crash_type == testcase.crash_type and
(state.crash_type != sym_crash_type or
state.crash_state != sym_crash_state)):
logs.log("Changing crash parameters.\nOld : %s, %s, %s" %
(sym_crash_type, sym_crash_address, sym_crash_state))
sym_crash_type = state.crash_type
sym_crash_address = state.crash_address
sym_crash_state = state.crash_state
sym_redzone = redzone
old_crash_stacktrace = state.crash_stacktrace
logs.log("\nNew : %s, %s, %s" %
(sym_crash_type, sym_crash_address, sym_crash_state))
break
redzone /= 2
# We should have atleast a symbolized debug or a release build.
symbolized_builds = build_manager.setup_symbolized_builds(crash_revision)
if not symbolized_builds or (
not build_manager.check_app_path() and
not build_manager.check_app_path("APP_PATH_DEBUG")):
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
"Build setup failed")
tasks.add_task(
"symbolize", testcase_id, job_type, wait_time=build_fail_wait)
return
# Increase malloc_context_size to get all stack frames. Default is 30.
environment.reset_current_memory_tool_options(
redzone_size=sym_redzone,
malloc_context_size=STACK_FRAME_COUNT,
symbolize_inline_frames=True,
disable_ubsan=testcase.disable_ubsan,
)
# TSAN tool settings (if the tool is used).
if environment.tool_matches("TSAN", job_type):
environment.set_tsan_max_history_size()
# Do the symbolization if supported by this application.
result, sym_crash_stacktrace = get_symbolized_stacktraces(
testcase_file_path, testcase, old_crash_stacktrace, sym_crash_state)
# Update crash parameters.
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.crash_type = sym_crash_type
testcase.crash_address = sym_crash_address
testcase.crash_state = sym_crash_state
testcase.crash_stacktrace = data_handler.filter_stacktrace(
sym_crash_stacktrace)
if not result:
data_handler.update_testcase_comment(
testcase,
data_types.TaskState.ERROR,
"Unable to reproduce crash, skipping "
"stacktrace update",
)
else:
# Switch build url to use the less-optimized symbolized build with better
# stacktrace.
build_url = environment.get_value("BUILD_URL")
if build_url:
testcase.set_metadata("build_url", build_url, update_testcase=False)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED)
testcase.symbolized = True
testcase.crash_revision = crash_revision
testcase.put()
# We might have updated the crash state. See if we need to marked as duplicate
# based on other testcases.
data_handler.handle_duplicate_entry(testcase)
task_creation.create_blame_task_if_needed(testcase)
# Switch current directory before builds cleanup.
root_directory = environment.get_value("ROOT_DIR")
os.chdir(root_directory)
# Cleanup symbolized builds which are space-heavy.
symbolized_builds.delete()
def execute_task(testcase_id, job_type):
"""Run a test case with a different job type to see if they reproduce."""
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
if (environment.is_engine_fuzzer_job(testcase.job_type) !=
environment.is_engine_fuzzer_job(job_type)):
# We should never reach here. But in case we do, we should bail out as
# otherwise we will run into exceptions.
return
# Setup testcase and its dependencies.
fuzzer_override = builtin_fuzzers.get_fuzzer_for_job(job_type)
file_list, _, testcase_file_path = setup.setup_testcase(
testcase, fuzzer_override=fuzzer_override)
if not file_list:
return
# Set up a custom or regular build. We explicitly omit the crash revision
# since we want to test against the latest build here.
build_manager.setup_build()
# Check if we have an application path. If not, our build failed to setup
# correctly.
app_path = environment.get_value('APP_PATH')
if not app_path:
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(
testcase, data_types.TaskState.ERROR,
'Build setup failed with job: ' + job_type)
return
# Reproduce the crash.
command = testcase_manager.get_command_line_for_application(
testcase_file_path, app_path=app_path, needs_http=testcase.http_flag)
test_timeout = environment.get_value('TEST_TIMEOUT', 10)
revision = environment.get_value('APP_REVISION')
result = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=testcase.http_flag,
compare_crash=False)
if result.is_crash() and not result.should_ignore():
crash_state = result.get_state()
crash_type = result.get_type()
security_flag = result.is_security_issue()
one_time_crasher_flag = not testcase_manager.test_for_reproducibility(
testcase_file_path, crash_state, security_flag, test_timeout,
testcase.http_flag, testcase.gestures)
if one_time_crasher_flag:
status = data_types.TestcaseVariantStatus.FLAKY
else:
status = data_types.TestcaseVariantStatus.REPRODUCIBLE
crash_comparer = CrashComparer(crash_state, testcase.crash_state)
is_similar = (crash_comparer.is_similar()
and security_flag == testcase.security_flag)
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
symbolized_crash_stacktrace = result.get_stacktrace(symbolized=True)
crash_stacktrace_output = utils.get_crash_stacktrace_output(
command, symbolized_crash_stacktrace,
unsymbolized_crash_stacktrace)
else:
status = data_types.TestcaseVariantStatus.UNREPRODUCIBLE
is_similar = False
crash_type = None
crash_state = None
security_flag = False
crash_stacktrace_output = 'No crash occurred.'
testcase = data_handler.get_testcase_by_id(testcase_id)
if testcase.job_type == job_type:
# This case happens when someone clicks 'Update last tested stacktrace using
# trunk build' button.
testcase.last_tested_crash_stacktrace = (
data_handler.filter_stacktrace(crash_stacktrace_output))
testcase.set_metadata('last_tested_crash_revision',
revision,
update_testcase=False)
else:
# Regular case of variant analysis.
variant = data_handler.get_testcase_variant(testcase_id, job_type)
variant.status = status
variant.revision = revision
variant.crash_type = crash_type
variant.crash_state = crash_state
variant.security_flag = security_flag
variant.is_similar = is_similar
# Explicitly skipping crash stacktrace for now as it make entities larger
# and we plan to use only crash paramaters in UI.
variant.put()
def setup_testcase(testcase):
"""Sets up the testcase and needed dependencies like fuzzer,
data bundle, etc."""
fuzzer_name = testcase.fuzzer_name
job_type = testcase.job_type
task_name = environment.get_value('TASK_NAME')
testcase_fail_wait = environment.get_value('FAIL_WAIT')
testcase_id = testcase.key.id()
# Clear testcase directories.
shell.clear_testcase_directories()
# Setup memory debugging tool environment.
environment.reset_current_memory_tool_options(
redzone_size=testcase.redzone)
# Adjust the test timeout value if this is coming from an user uploaded
# testcase.
_set_timeout_value_from_user_upload(testcase_id)
if task_name == 'minimize':
# Allow minimizing with a different fuzzer set up.
minimize_fuzzer_override = environment.get_value(
'MINIMIZE_FUZZER_OVERRIDE')
fuzzer_name = minimize_fuzzer_override or fuzzer_name
# Update the fuzzer if necessary in order to get the updated data bundle.
if fuzzer_name:
try:
update_successful = update_fuzzer_and_data_bundles(fuzzer_name)
except errors.InvalidFuzzerError:
# Close testcase and don't recreate tasks if this fuzzer is invalid.
testcase.open = False
testcase.fixed = 'NA'
testcase.set_metadata('fuzzer_was_deleted', True)
logs.log_error('Closed testcase %d with invalid fuzzer %s.' %
(testcase_id, fuzzer_name))
error_message = 'Fuzzer %s no longer exists.' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
return None, None, None
if not update_successful:
error_message = 'Unable to setup fuzzer %s.' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# Extract the testcase and any of its resources to the input directory.
file_list, input_directory, testcase_file_path = unpack_testcase(testcase)
if not file_list:
error_message = 'Unable to setup testcase %s.' % testcase_file_path
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# For Android/Fuchsia, we need to sync our local testcases directory with the
# one on the device.
if environment.platform() == 'ANDROID':
_copy_testcase_to_device_and_setup_environment(testcase,
testcase_file_path)
if environment.platform() == 'FUCHSIA':
fuchsia.device.copy_testcase_to_device(testcase_file_path)
# Push testcases to worker.
if environment.is_trusted_host():
from bot.untrusted_runner import file_host
file_host.push_testcases_to_worker()
# Copy global blacklist into local blacklist.
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Get local blacklist without this testcase's entry.
leak_blacklist.copy_global_to_local_blacklist(
excluded_testcase=testcase)
# Setup environment variable for windows size and location properties.
# Explicitly use empty string to indicate use of default window properties.
if hasattr(testcase, 'window_argument'):
environment.set_value('WINDOW_ARG', testcase.window_argument)
# Adjust timeout based on the stored multiplier (if available).
if hasattr(testcase, 'timeout_multiplier') and testcase.timeout_multiplier:
test_timeout = environment.get_value('TEST_TIMEOUT')
environment.set_value('TEST_TIMEOUT',
int(test_timeout * testcase.timeout_multiplier))
# Override APP_ARGS with minimized arguments (if available).
if (hasattr(testcase, 'minimized_arguments')
and testcase.minimized_arguments):
environment.set_value('APP_ARGS', testcase.minimized_arguments)
# Add FUZZ_TARGET to environment if this is a fuzz target testcase.
fuzz_target = testcase.get_metadata('fuzzer_binary_name')
if fuzz_target:
environment.set_value('FUZZ_TARGET', fuzz_target)
return file_list, input_directory, testcase_file_path
