def test_group_translation(responses):
branch = "integration/autoland",
rev = "abcdef"
query = "/push/{}/{}/schedules".format(branch, rev)
url = BUGBUG_BASE_URL + query
responses.add(
responses.GET,
url,
json={
"groups": {
"dom/indexedDB": 1,
"testing/web-platform/tests/IndexedDB": 1,
"testing/web-platform/mozilla/tests/IndexedDB": 1,
}
},
status=200,
)
assert len(push_schedules) == 0
data = push_schedules(branch, rev)
print(data)
assert sorted(data["groups"]) == [
"/IndexedDB",
"/_mozilla/IndexedDB",
"dom/indexedDB",
]
assert len(push_schedules) == 1
# Value is memoized.
responses.reset()
push_schedules(branch, rev)
assert len(push_schedules) == 1
def get_manifests(self, suite, mozinfo):
manifests = super(BugbugLoader, self).get_manifests(suite, mozinfo)
# Don't prune any manifests if we're on a backstop push or there was a timeout.
if self.params["backstop"] or self.timedout:
return manifests
try:
data = push_schedules(self.params["project"],
self.params["head_rev"])
except BugbugTimeoutException:
logger.warning(
"Timed out waiting for bugbug, loading all test manifests.")
self.timedout = True
return self.get_manifests(suite, mozinfo)
bugbug_manifests = {
m
for m, c in data.get("groups", {}).items()
if c >= self.CONFIDENCE_THRESHOLD
}
manifests["active"] = list(set(manifests["active"]) & bugbug_manifests)
manifests["skipped"] = list(
set(manifests["skipped"]) & bugbug_manifests)
return manifests
def should_remove_task(self, task, params, importance):
if params['project'] not in ("autoland", "try"):
return False
branch = urlsplit(params['head_repository']).path.strip('/')
rev = params['head_rev']
if self.timedout:
return seta.is_low_value_task(task.label, params['project'])
try:
data = push_schedules(branch, rev)
except BugbugTimeoutException:
if not self.fallback:
raise
self.timedout = True
return self.should_remove_task(task, params, importance)
key = "reduced_tasks" if self.use_reduced_tasks else "tasks"
tasks = set(
task
for task, confidence in data.get(key, {}).items()
if confidence >= self.confidence_threshold
)
test_manifests = task.attributes.get('test_manifests')
if test_manifests is None or self.use_reduced_tasks:
if data.get("known_tasks") and task.label not in data["known_tasks"]:
return False
if task.label not in tasks:
return True
return False
# If a task contains more than one group, use the max confidence.
groups = data.get("groups", {})
confidences = [c for g, c in groups.items() if g in test_manifests]
if not confidences or max(confidences) < self.confidence_threshold:
return True
# Store group importance so future optimizers can access it.
for manifest in test_manifests:
if manifest not in groups:
continue
confidence = groups[manifest]
if confidence >= CT_HIGH:
importance[manifest] = 'high'
elif confidence >= CT_MEDIUM:
importance[manifest] = 'medium'
elif confidence >= CT_LOW:
importance[manifest] = 'low'
else:
importance[manifest] = 'lowest'
return False
def get_manifests(self, suite, mozinfo):
manifests = super(BugbugLoader, self).get_manifests(suite, mozinfo)
data = push_schedules(self.params['project'], self.params['head_rev'])
bugbug_manifests = {
m
for m, c in data.get('groups', {}).items()
if c >= self.CONFIDENCE_THRESHOLD
}
manifests['active'] = list(set(manifests['active']) & bugbug_manifests)
return manifests
def test_only_important_manifests(params, full_task_graph, filter_tasks):
data = push_schedules(params["project"], params["head_rev"])
important_manifests = {
m
for m, c in data.get("groups", {}).items()
if c >= BugbugLoader.CONFIDENCE_THRESHOLD
}
# Ensure we don't schedule unimportant manifests.
for task in filter_tasks(full_task_graph, lambda t: t.kind == "test"):
attr = task.attributes.get
if attr("unittest_suite") in CHUNK_SUITES_BLACKLIST:
assert not attr("test_manifests")
else:
unimportant = [
t for t in attr("test_manifests")
if t not in important_manifests
]
assert unimportant == []
def test_only_important_manifests(params, full_task_graph, filter_tasks):
data = push_schedules(params["project"], params["head_rev"])
important_manifests = {
m
for m, c in data.get("groups", {}).items()
if c >= BugbugLoader.CONFIDENCE_THRESHOLD
}
# Ensure we don't schedule unimportant manifests.
for task in filter_tasks(full_task_graph, lambda t: t.kind == "test"):
attr = task.attributes.get
if "test_manifests" in task.attributes:
unimportant = [
t for t in attr("test_manifests") if t not in important_manifests
]
# Manifest scheduling is disabled for mochitest-ally.
if attr("unittest_suite") == "mochitest-a11y":
assert len(unimportant) > 0
else:
assert unimportant == []
def should_remove_task(self, task, params, importance):
project = params["project"]
if project not in ("autoland", "try"):
return False
current_push_id = int(params["pushlog_id"])
branch = urlsplit(params["head_repository"]).path.strip("/")
rev = params["head_rev"]
if self.timedout:
return registry[self.fallback].should_remove_task(
task, params, importance)
data = {}
start_push_id = current_push_id - self.num_pushes + 1
if self.num_pushes != 1:
push_data = get_push_data(params["head_repository"], project,
start_push_id, current_push_id - 1)
for push_id in range(start_push_id, current_push_id + 1):
if push_id == current_push_id:
rev = params["head_rev"]
else:
rev = push_data[push_id]["changesets"][-1]
try:
new_data = push_schedules(branch, rev)
merge_bugbug_replies(data, new_data)
except BugbugTimeoutException:
if not self.fallback:
raise
self.timedout = True
return self.should_remove_task(task, params, importance)
key = "reduced_tasks" if self.use_reduced_tasks else "tasks"
tasks = set(task for task, confidence in data.get(key, {}).items()
if confidence >= self.confidence_threshold)
test_manifests = task.attributes.get("test_manifests")
if test_manifests is None or self.tasks_only:
if data.get(
"known_tasks") and task.label not in data["known_tasks"]:
return False
if task.label not in tasks:
return True
return False
# If a task contains more than one group, use the max confidence.
groups = data.get("groups", {})
confidences = [c for g, c in groups.items() if g in test_manifests]
if not confidences or max(confidences) < self.confidence_threshold:
return True
# If the task configuration doesn't match the ones selected by bugbug for
# the manifests, optimize out.
if self.select_configs:
selected_groups = [
g for g, c in groups.items()
if g in test_manifests and c > self.confidence_threshold
]
config_groups = data.get("config_groups", defaultdict(list))
# Configurations returned by bugbug are in a format such as
# `test-windows10-64/opt-*-e10s`, while task labels are like
# test-windows10-64-qr/opt-mochitest-browser-chrome-e10s-6.
# In order to match the strings, we need to ignore the chunk number
# from the task label.
parts = task.label.split("-")
label_without_chunk_number = "-".join(
parts[:-1] if parts[-1].isdigit() else parts)
if not any(
fnmatch(label_without_chunk_number, config)
for group in selected_groups
for config in config_groups[group]):
return True
# Store group importance so future optimizers can access it.
for manifest in test_manifests:
if manifest not in groups:
continue
confidence = groups[manifest]
if confidence >= CT_HIGH:
importance[manifest] = "high"
elif confidence >= CT_MEDIUM:
importance[manifest] = "medium"
elif confidence >= CT_LOW:
importance[manifest] = "low"
else:
importance[manifest] = "lowest"
return False
def should_remove_task(self, task, params, importance):
project = params['project']
if project not in ("autoland", "try"):
return False
current_push_id = int(params['pushlog_id'])
branch = urlsplit(params['head_repository']).path.strip('/')
rev = params['head_rev']
if self.timedout:
return registry[self.fallback].should_remove_task(
task, params, importance)
data = {}
start_push_id = current_push_id - self.num_pushes + 1
if self.num_pushes != 1:
push_data = get_push_data(params["head_repository"], project,
start_push_id, current_push_id - 1)
for push_id in range(start_push_id, current_push_id + 1):
if push_id == current_push_id:
rev = params["head_rev"]
else:
rev = push_data[push_id]["changesets"][-1]
try:
new_data = push_schedules(branch, rev)
for key, value in new_data.items():
if isinstance(value, dict):
if key not in data:
data[key] = {}
for name, confidence in value.items():
if name not in data[
key] or data[key][name] < confidence:
data[key][name] = confidence
elif isinstance(value, list):
if key not in data:
data[key] = set()
data[key].update(value)
except BugbugTimeoutException:
if not self.fallback:
raise
self.timedout = True
return self.should_remove_task(task, params, importance)
key = "reduced_tasks" if self.use_reduced_tasks else "tasks"
tasks = set(task for task, confidence in data.get(key, {}).items()
if confidence >= self.confidence_threshold)
test_manifests = task.attributes.get('test_manifests')
if test_manifests is None or self.tasks_only:
if data.get(
"known_tasks") and task.label not in data["known_tasks"]:
return False
if task.label not in tasks:
return True
return False
# If a task contains more than one group, use the max confidence.
groups = data.get("groups", {})
confidences = [c for g, c in groups.items() if g in test_manifests]
if not confidences or max(confidences) < self.confidence_threshold:
return True
# Store group importance so future optimizers can access it.
for manifest in test_manifests:
if manifest not in groups:
continue
confidence = groups[manifest]
if confidence >= CT_HIGH:
importance[manifest] = 'high'
elif confidence >= CT_MEDIUM:
importance[manifest] = 'medium'
elif confidence >= CT_LOW:
importance[manifest] = 'low'
else:
importance[manifest] = 'lowest'
return False