1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
|
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
import json
import taskgraph
from taskgraph.transforms.base import TransformSequence
from taskgraph.util.attributes import keymatch
from taskgraph.util.treeherder import join_symbol, split_symbol
from gecko_taskgraph.util.attributes import is_try
from gecko_taskgraph.util.chunking import (
DefaultLoader,
chunk_manifests,
get_manifest_loader,
get_runtimes,
guess_mozinfo_from_task,
)
from gecko_taskgraph.util.copy_task import copy_task
from gecko_taskgraph.util.perfile import perfile_number_of_chunks
DYNAMIC_CHUNK_DURATION = 20 * 60 # seconds
"""The approximate time each test chunk should take to run."""
DYNAMIC_CHUNK_MULTIPLIER = {
# Desktop xpcshell tests run in parallel. Reduce the total runtime to
# compensate.
"^(?!android).*-xpcshell.*": 0.2,
}
"""A multiplication factor to tweak the total duration per platform / suite."""
transforms = TransformSequence()
@transforms.add
def set_test_verify_chunks(config, tasks):
"""Set the number of chunks we use for test-verify."""
for task in tasks:
if any(task["suite"].startswith(s) for s in ("test-verify", "test-coverage")):
env = config.params.get("try_task_config", {}) or {}
env = env.get("templates", {}).get("env", {})
task["chunks"] = perfile_number_of_chunks(
is_try(config.params),
env.get("MOZHARNESS_TEST_PATHS", ""),
frozenset(config.params["files_changed"]),
task["test-name"],
)
# limit the number of chunks we run for test-verify mode because
# test-verify is comprehensive and takes a lot of time, if we have
# >30 tests changed, this is probably an import of external tests,
# or a patch renaming/moving files in bulk
maximum_number_verify_chunks = 3
if task["chunks"] > maximum_number_verify_chunks:
task["chunks"] = maximum_number_verify_chunks
yield task
@transforms.add
def set_test_manifests(config, tasks):
"""Determine the set of test manifests that should run in this task."""
for task in tasks:
# When a task explicitly requests no 'test_manifest_loader', test
# resolving will happen at test runtime rather than in the taskgraph.
if "test-manifest-loader" in task and task["test-manifest-loader"] is None:
yield task
continue
# Set 'tests_grouped' to "1", so we can differentiate between suites that are
# chunked at the test runtime and those that are chunked in the taskgraph.
task.setdefault("tags", {})["tests_grouped"] = "1"
if taskgraph.fast:
# We want to avoid evaluating manifests when taskgraph.fast is set. But
# manifests are required for dynamic chunking. Just set the number of
# chunks to one in this case.
if task["chunks"] == "dynamic":
task["chunks"] = 1
yield task
continue
manifests = task.get("test-manifests")
if manifests:
if isinstance(manifests, list):
task["test-manifests"] = {"active": manifests, "skipped": []}
yield task
continue
mozinfo = guess_mozinfo_from_task(
task, config.params.get("head_repository", "")
)
loader_name = task.pop(
"test-manifest-loader", config.params["test_manifest_loader"]
)
loader = get_manifest_loader(loader_name, config.params)
task["test-manifests"] = loader.get_manifests(
task["suite"],
frozenset(mozinfo.items()),
)
# When scheduling with test paths, we often find manifests scheduled but all tests
# are skipped on a given config. This will remove the task from the task set if
# no manifests have active tests for the given task/config
mh_test_paths = {}
if "MOZHARNESS_TEST_PATHS" in config.params.get("try_task_config", {}).get(
"env", {}
):
mh_test_paths = json.loads(
config.params["try_task_config"]["env"]["MOZHARNESS_TEST_PATHS"]
)
if task["attributes"]["unittest_suite"] in mh_test_paths.keys():
input_paths = mh_test_paths[task["attributes"]["unittest_suite"]]
remaining_manifests = []
# if we have web-platform tests incoming, just yield task
for m in input_paths:
if m.startswith("testing/web-platform/tests/"):
if not isinstance(loader, DefaultLoader):
task["chunks"] = "dynamic"
yield task
break
# input paths can exist in other directories (i.e. [../../dir/test.js])
# we need to look for all [active] manifests that include tests in the path
for m in input_paths:
if [tm for tm in task["test-manifests"]["active"] if tm.startswith(m)]:
remaining_manifests.append(m)
# look in the 'other' manifests
for m in input_paths:
man = m
for tm in task["test-manifests"]["other_dirs"]:
matched_dirs = [
dp
for dp in task["test-manifests"]["other_dirs"].get(tm)
if dp.startswith(man)
]
if matched_dirs:
if tm not in task["test-manifests"]["active"]:
continue
if m not in remaining_manifests:
remaining_manifests.append(m)
if remaining_manifests == []:
continue
# The default loader loads all manifests. If we use a non-default
# loader, we'll only run some subset of manifests and the hardcoded
# chunk numbers will no longer be valid. Dynamic chunking should yield
# better results.
if not isinstance(loader, DefaultLoader):
task["chunks"] = "dynamic"
yield task
@transforms.add
def resolve_dynamic_chunks(config, tasks):
"""Determine how many chunks are needed to handle the given set of manifests."""
for task in tasks:
if task["chunks"] != "dynamic":
yield task
continue
if not task.get("test-manifests"):
raise Exception(
"{} must define 'test-manifests' to use dynamic chunking!".format(
task["test-name"]
)
)
runtimes = {
m: r
for m, r in get_runtimes(task["test-platform"], task["suite"]).items()
if m in task["test-manifests"]["active"]
}
# Truncate runtimes that are above the desired chunk duration. They
# will be assigned to a chunk on their own and the excess duration
# shouldn't cause additional chunks to be needed.
times = [min(DYNAMIC_CHUNK_DURATION, r) for r in runtimes.values()]
avg = round(sum(times) / len(times), 2) if times else 0
total = sum(times)
# If there are manifests missing from the runtimes data, fill them in
# with the average of all present manifests.
missing = [m for m in task["test-manifests"]["active"] if m not in runtimes]
total += avg * len(missing)
# Apply any chunk multipliers if found.
key = "{}-{}".format(task["test-platform"], task["test-name"])
matches = keymatch(DYNAMIC_CHUNK_MULTIPLIER, key)
if len(matches) > 1:
raise Exception(
"Multiple matching values for {} found while "
"determining dynamic chunk multiplier!".format(key)
)
elif matches:
total = total * matches[0]
chunks = int(round(total / DYNAMIC_CHUNK_DURATION))
# Make sure we never exceed the number of manifests, nor have a chunk
# length of 0.
task["chunks"] = min(chunks, len(task["test-manifests"]["active"])) or 1
yield task
@transforms.add
def split_chunks(config, tasks):
"""Based on the 'chunks' key, split tests up into chunks by duplicating
them and assigning 'this-chunk' appropriately and updating the treeherder
symbol.
"""
for task in tasks:
# If test-manifests are set, chunk them ahead of time to avoid running
# the algorithm more than once.
chunked_manifests = None
if "test-manifests" in task:
# TODO: hardcoded to "2", ideally this should be centralized somewhere
if (
config.params["try_task_config"].get("new-test-config", False)
and task["chunks"] > 1
):
task["chunks"] *= 2
task["max-run-time"] = int(task["max-run-time"] * 2)
manifests = task["test-manifests"]
chunked_manifests = chunk_manifests(
task["suite"],
task["test-platform"],
task["chunks"],
manifests["active"],
)
# Add all skipped manifests to the first chunk of backstop pushes
# so they still show up in the logs. They won't impact runtime much
# and this way tools like ActiveData are still aware that they
# exist.
if config.params["backstop"] and manifests["active"]:
chunked_manifests[0].extend(manifests["skipped"])
for i in range(task["chunks"]):
this_chunk = i + 1
# copy the test and update with the chunk number
chunked = copy_task(task)
chunked["this-chunk"] = this_chunk
if chunked_manifests is not None:
chunked["test-manifests"] = sorted(chunked_manifests[i])
group, symbol = split_symbol(chunked["treeherder-symbol"])
if task["chunks"] > 1 or not symbol:
# add the chunk number to the TH symbol
symbol += str(this_chunk)
chunked["treeherder-symbol"] = join_symbol(group, symbol)
yield chunked
|