Debian Codegraph @

testing/performance/mach-try-perf/perfdocs/index.rst

raw
1 #############
2 Mach Try Perf
3 #############
4
5 .. contents::
6 :depth: 2
7 :local:
8
9 To make it easier for developers to find the tests they need to run we built a perf-specific try selector called ``./mach try perf``. With this tool, you no longer need to remember the obfuscated platform and test names that you need to target for your tests. Instead, the new interface shows test categories along with a simplified name of the platform that they will run on.
10
11 When you trigger a try run from the perf selector, two try runs will be created. One with your changes, and one without. The push without your changes will be done on the revision that your patches are based on (which we call the base revision). In your console, after you trigger the try runs, you'll find a PerfCompare link that will bring you directly to a comparison of the two pushes when they have completed.
12
13 The tool is built to be conservative about the number of tests to run, so if you are looking for something that is not listed, it's likely hidden behind a flag found in the ``--help``. Here's a list of what you'll find there::
14
15 $ ./mach try perf --help
16
17 optional arguments:
18 -h, --help show this help message and exit
19 perf arguments:
20 --show-all Show all available tasks. Alternatively, --full may be used.
21 --chrome Show tests available for Chrome-based browsers (disabled by
22 default).
23 --custom-car Show tests available for Custom Chromium-as-Release (disabled by
24 default)
25 --safari Show tests available for Safari (disabled by default).
26 --safari-tp Show tests available for Safari Technology Preview(disabled by
27 default).
28 --live-sites Run tasks with live sites (if possible). You can also use the
29 `live-sites` variant.
30 --profile Run tasks with profiling (if possible). You can also use the
31 `profiling` variant.
32 --single-run Run tasks without a comparison
33 -q QUERY, --query QUERY
34 Query to run in either the perf-category selector, or the fuzzy
35 selector if --show-all/--full is provided.
36 --browsertime-upload-apk BROWSERTIME_UPLOAD_APK
37 Path to an APK to upload. Note that this will replace the APK
38 installed in all Android Performance tests. If the Activity,
39 Binary Path, or Intents required change at all relative to the
40 existing GeckoView, and Fenix tasks, then you will need to make
41 fixes in the associated taskcluster files (e.g.
42 taskcluster/kinds/test/browsertime-mobile.yml). Alternatively, set
43 MOZ_FIREFOX_ANDROID_APK_OUTPUT to a path to an APK, and then run
44 the command with --browsertime-upload-apk firefox-android. This
45 option will only copy the APK for browsertime, see --mozperftest-
46 upload-apk to upload APKs for startup tests.
47 --mozperftest-upload-apk MOZPERFTEST_UPLOAD_APK
48 See --browsertime-upload-apk. This option does the same thing
49 except it's for mozperftest tests such as the startup ones. Note
50 that those tests only exist through --show-all/--full as they aren't
51 contained in any existing categories.
52 --detect-changes Adds a task that detects performance changes using MWU.
53 --comparator COMPARATOR
54 Either a path to a file to setup a custom comparison, or a builtin
55 name. See the Firefox source docs for mach try perf for examples
56 of how to build your own, along with the interface.
57 --comparator-args [ARG=VALUE [ARG=VALUE ...]]
58 Arguments provided to the base, and new revision setup stages of
59 the comparator.
60 --variants [ [ ...]] Select variants to display in the selector from: fission,
61 bytecode-cached, live-sites, profiling, swr
62 --platforms [ [ ...]]
63 Select specific platforms to target.
64 Available platforms: android-a55, android, windows,
65 linux, macosx, desktop
66 --apps [ [ ...]] Select specific applications to target from: firefox, chrome,
67 geckoview, fenix, chrome-m, safari, safari-tp, custom-car, cstm-
68 car-m
69 --clear-cache Deletes the try_perf_revision_cache file
70 --alert ALERT Run all tests that produced this alert summary ID based on the
71 alert summary table in either the alerts view or the regression
72 bug. The comparison that is produced will be based on the base
73 revision in your local repository (i.e. the base revision your
74 patches, if any, are based on). If only specific tests need to
75 run, use --tests to specify them (e.g. --tests webaudio).
76 --extra-args [ [ ...]]
77 Set the extra args (e.x, --extra-args verbose post-startup-
78 delay=1)
79 --non-pgo Use opt/non-pgo builds instead of shippable/pgo builds. Setting
80 this flag will result in faster try runs.
81 --tests [TESTS [TESTS ...]], -t [TESTS [TESTS ...]]
82 Select from all tasks that run these specific tests (e.g. amazon, or
83 speedometer3).
84
85 task configuration arguments:
86 --artifact Force artifact builds where possible.
87 --no-artifact Disable artifact builds even if being used locally.
88 --browsertime Use browsertime during Raptor tasks.
89 --disable-pgo Don't run PGO builds
90 --env ENV Set an environment variable, of the form FOO=BAR. Can
91 be passed in multiple times.
92 --gecko-profile Create and upload a gecko profile during talos/raptor
93 tasks. Copy paste the parameters used in this profiling
94 run directly from about:profiling in Nightly.
95 --gecko-profile-interval GECKO_PROFILE_INTERVAL
96 How frequently to take samples (ms)
97 --gecko-profile-entries GECKO_PROFILE_ENTRIES
98 How many samples to take with the profiler
99 --gecko-profile-features GECKO_PROFILE_FEATURES
100 Set the features enabled for the profiler.
101 --gecko-profile-threads GECKO_PROFILE_THREADS
102 Comma-separated list of threads to sample.
103 paths Run tasks containing tests under the specified
104 path(s).
105 --rebuild [2-20] Rebuild all selected tasks the specified number of
106 times.
107
108
109
110 Workflow
111 --------
112
113 Below, you'll find an overview of the features available in ``./mach try perf``. If you'd like to learn more about how to use this tool to enhance your development process, see the :ref:`Standard Workflow with Mach Try Perf` page.
114
115 Standard Usage
116 --------------
117
118 To use mach try perf simply call ``./mach try perf``. This will open an interface for test selection like so:
119
120
121 .. image:: ./standard-try-perf.png
122 :alt: Mach try perf with default options
123 :scale: 75%
124 :align: center
125
126
127 Select the categories you'd like to run, hit enter, and wait for the tool to finish the pushes. **Note that it can take some time to do both pushes, and you might not see logging for some time.**
128
129
130 Retrigger
131 ---------
132 After the push is done, you will receive a Treeherder link that you can open to view your push. Access the Treeherder link to see all your tests.
133
134 To launch a retrigger, first select the task that you want to retrigger:
135
136 .. image:: ./th_select_task.png
137 :width: 300
138
139
140 Then, click the rotating arrow icon in the task action bar, or press 'r' on your keyboard:
141
142 .. image:: ./th_retrigger.png
143 :width: 300
144
145
146 Additionally, you can add the flag ``--rebuild=2-20`` to the try perf command to specify how many times you want to run the tests. If you want to learn more about retriggering please `visit this page <../treeherder-try/index.html#retrigger-r>`__.
147
148
149 Add new jobs (mass retriggers)
150 ------------------------------
151
152 The add new job function can be used to retrigger many tasks multiple times. To add a new job, follow these steps:
153 * Navigate to the push you want to add jobs on Treeherder.
154 * Click on the arrow drop-down on the top right of the push.
155 * Select the ``Custom push action`` from the menu.
156
157 .. image:: ./th_custom_push_action.png
158 :width: 500
159
160 You can copy the values from the ``target-tasks.json`` file from your ``Decision`` task and paste them into the ``task`` option. This method is useful for mass retriggers if needed.
161 After you have pasted the json values, press the ``Trigger`` button.
162
163 .. image:: ./th_custom_job_action.png
164 :width: 500
165
166 Ideally, you should be able to use compare view to be more specific in the retriggers you do for tasks/tests that show a difference that they want to double-check.
167
168
169 Add extra-arguments/options to try run
170 --------------------------------------
171
172 To add additional arguments to a try run, there are several approaches you can consider:
173
174
175 Use Treeherder
176 ^^^^^^^^^^^^^^
177
178 This method assumes that you already have the job that has been run and you want to run it again, but this time to add extra options as well. First select the task that you want to add extra options:
179
180 .. image:: ./th_select_task.png
181 :width: 300
182
183 Then, click the three dots icon in the task action bar and select ``Custom Action``:
184
185 .. image:: ./th_custom_action.png
186 :width: 300
187
188 A window will open where you need to select ``raptor-extra-options``. There you can add all the options you need (e.g. extra_options: 'verbose browser-cycles=3'). After finishing, press the ``Trigger`` button.
189
190 .. image:: ./th_raptor_extra_option.png
191 :width: 500
192
193 Modify the yml file
194 ^^^^^^^^^^^^^^^^^^^
195
196 This method involves identifying the YML file that contains the test you are interested in and modifying or adding the extra-options key. Under this key you can add all the parameters you desire.
197
198 .. image:: ./extra-options.png
199 :width: 500
200
201 Use extra-args option
202 ^^^^^^^^^^^^^^^^^^^^^
203
204 An alternative method is to utilize the ``--extra-args`` argument to try perf command (e.g. --extra-args verbose post-startup-delay=1).
205
206
207 .. _Running Alert Tests:
208
209 Running Alert Tests
210 -------------------
211
212 To run all the tests that triggered a given alert, use ``./mach try perf --alert <ALERT-NUMBER>``. Using this command will run all the tests that generated the alert summary ID provided in the regression bug. **It's recommended to use this when working with performance alerts.** The alert number can be found in comment 0 on any alert bug, `such as this one <https://bugzilla.mozilla.org/show_bug.cgi?id=1844510>`_. As seen in the image below, the alert number can be found just above the summary table. The comparison that is produced will be based on the base revision in your local repository (i.e. the base revision your patches, if any, are based on).
213
214 .. image:: ./comment-zero-alert-number.png
215 :alt: Comment 0 containing an alert number just above the table.
216 :scale: 50%
217 :align: center
218
219
220 Running Tasks of a Specific Test
221 --------------------------------
222
223 Using the ``--tests`` option, you can run all tasks that run a specific test. This is based on the test name that is used in the command that runs in the task. For raptor, this is the test specified by ``--test``. For talos, it can either be a specific test in a suite like ``tp5n`` from ``xperf``, or the suite ``xperf`` can be specified. For AWSY though, there are no specific tests that can be selected so the only option to select AWSY tests is to specify ``AWSY`` as the test.
224
225 If it's used with ``--alert <NUM>``, only the tasks that run the specific test will be run on try. If it's used with ``--show-all`` or ``--full``, you will only see the tasks that run the specific test in the fuzzy interface. Finally, if it's used without either of those, then categories of the tests that were specified will be displayed in the fuzzy interface. For example, if ``--tests amazon`` is used, then categories like ``amazon linux firefox`` or ``amazon desktop`` will be displayed.
226
227 Chrome
228 ------
229
230 Chrome tests are disabled by default as they are often unneeded and waste our limited resources. If you need chrome tests you can add ``--chrome`` to the command like so ``./mach try perf --chrome``:
231
232
233 .. image:: ./android-chrome-try-perf.png
234 :alt: Mach try perf with android, and chrome options
235 :scale: 75%
236 :align: center
237
238
239 Variants
240 --------
241
242 If you are looking for any variants (e.g. no-fission, bytecode-cached, live-sites), use the ``--variants`` options like so ``./mach try perf --variants live-sites``. This will select all possible categories that could have live-sites tests.
243
244
245 .. image:: ./variants-try-perf.png
246 :alt: Mach try perf with variants
247 :scale: 75%
248 :align: center
249
250
251 Note that it is expected that the offered categories have extra variants (such as bytecode-cached) as we are showing all possible combinations that can include live-sites.
252
253 Platforms
254 ---------
255
256 To target a particular platform you can use ``--platforms`` to only show categories with the given platforms.
257
258 Categories
259 ----------
260
261 In the future, this section will be populated dynamically. If you are wondering what the categories you selected will run, you can use ``--no-push`` to print out a list of tasks that will run like so::
262
263 $ ./mach try perf --no-push
264
265 Artifact builds enabled, pass --no-artifact to disable
266 Gathering tasks for Benchmarks desktop category
267 Executing queries: 'browsertime 'benchmark, !android 'shippable !-32 !clang, !live, !profil, !chrom
268 estimates: Runs 66 tasks (54 selected, 12 dependencies)
269 estimates: Total task duration 8:45:58
270 estimates: In the shortest 38% of durations (thanks!)
271 estimates: Should take about 1:04:58 (Finished around 2022-11-22 15:08)
272 Commit message:
273 Perf selections=Benchmarks desktop (queries='browsertime 'benchmark&!android 'shippable !-32 !clang&!live&!profil&!chrom)
274 Pushed via `mach try perf`
275 Calculated try_task_config.json:
276 {
277 "env": {
278 "TRY_SELECTOR": "fuzzy"
279 },
280 "tasks": [
281 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-ares6",
282 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-assorted-dom",
283 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-jetstream2",
284 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-matrix-react-bench",
285 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-animometer",
286 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-htmlsuite",
287 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-speedometer",
288 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-stylebench",
289 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-sunspider",
290 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-twitch-animation",
291 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-unity-webgl",
292 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-firefox-webaudio",
293 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot",
294 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-baseline",
295 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-optimizing",
296 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc",
297 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-baseline",
298 "test-linux1804-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-optimizing",
299 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-ares6",
300 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-assorted-dom",
301 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-jetstream2",
302 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-matrix-react-bench",
303 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-animometer",
304 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-htmlsuite",
305 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-speedometer",
306 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-stylebench",
307 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-sunspider",
308 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-twitch-animation",
309 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-unity-webgl",
310 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-firefox-webaudio",
311 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot",
312 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-baseline",
313 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-optimizing",
314 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc",
315 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-baseline",
316 "test-macosx1015-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-optimizing",
317 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-ares6",
318 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-assorted-dom",
319 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-jetstream2",
320 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-matrix-react-bench",
321 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-animometer",
322 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-motionmark-htmlsuite",
323 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-speedometer",
324 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-stylebench",
325 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-sunspider",
326 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-twitch-animation",
327 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-unity-webgl",
328 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-firefox-webaudio",
329 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot",
330 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-baseline",
331 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-godot-optimizing",
332 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc",
333 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-baseline",
334 "test-windows10-64-shippable-qr/opt-browsertime-benchmark-wasm-firefox-wasm-misc-optimizing"
335 ],
336 "use-artifact-builds": true,
337 "version": 1
338 }
339
340
341 Adding a New Category
342 ---------------------
343
344 It's very easy to add a new category if needed, and you can do so by modifying the `PerfParser categories attribute here <https://searchfox.org/mozilla-central/source/tools/tryselect/selectors/perf.py#179>`_. The following is an example of a complex category that gives a good idea of what you have available::
345
346 "Resource Usage": {
347 "query": {
348 "talos": ["'talos 'xperf | 'tp5"],
349 "raptor": ["'power 'osx"],
350 "awsy": ["'awsy"],
351 },
352 "suites": ["talos", "raptor", "awsy"],
353 "platform-restrictions": ["desktop"],
354 "variant-restrictions": {
355 "raptor": [],
356 "talos": [],
357 },
358 "app-restrictions": {
359 "raptor": ["firefox"],
360 "talos": ["firefox"],
361 },
362 "tasks": [],
363 },
364
365 The following fields are available:
366 * **query**: Set the queries to use for each suite you need.
367 * **suites**: The suites that are needed for this category.
368 * **tasks**: A hard-coded list of tasks to select.
369 * **platform-restrictions**: The platforms that it can run on.
370 * **app-restrictions**: A list of apps that the category can run.
371 * **variant-restrictions**: A list of variants available for each suite.
372
373 Note that setting the App/Variant-Restriction fields should be used to restrict the available apps and variants, not expand them as the suites, apps, and platforms combined already provide the largest coverage. The restrictions should be used when you know certain things definitely won't work, or will never be implemented for this category of tests. For instance, our ``Resource Usage`` tests only work on Firefox even though they may exist in Raptor which can run tests with Chrome.
374
375 Comparators
376 -----------
377
378 If the standard/default push-to-try comparison is not enough, you can build your own "comparator" that can setup the base, and new revisions. The default comparator ``BasePerfComparator`` runs the standard mach-try-perf comparison, and there also exists a custom comparator called ``BenchmarkComparator`` for running custom benchmark comparisons on try (using Github PR links).
379
380 If you'd like to add a custom comparator, you can either create it in a separate file and pass it in the ``--comparator``, or add it to the ``tools/tryselect/selectors/perfselector/perfcomparators.py`` and use the name of the class as the ``--comparator`` argument (e.g. ``--comparator BenchmarkComparator``). You can pass additional arguments to it using the ``--comparator-args`` option that accepts arguments in the format ``NAME=VALUE``.
381
382 The custom comparator needs to be a subclass of ``BasePerfComparator``, and optionally overrides its methods. See the comparators file for more information about the interface available. Here's the general interface for it (subject to change), note that the ``@comparator`` decorator is required when making a builtin comparator::
383
384 @comparator
385 class BasePerfComparator:
386 def __init__(self, vcs, compare_commit, current_revision_ref, comparator_args):
387 """Initialize the standard/default settings for Comparators.
388
389 :param vcs object: Used for updating the local repo.
390 :param compare_commit str: The base revision found for the local repo.
391 :param current_revision_ref str: The current revision of the local repo.
392 :param comparator_args list: List of comparator args in the format NAME=VALUE.
393 """
394
395 def setup_base_revision(self, extra_args):
396 """Setup the base try run/revision.
397
398 The extra_args can be used to set additional
399 arguments for Raptor (not available for other harnesses).
400
401 :param extra_args list: A list of extra arguments to pass to the try tasks.
402 """
403
404 def teardown_base_revision(self):
405 """Teardown the setup for the base revision."""
406
407 def setup_new_revision(self, extra_args):
408 """Setup the new try run/revision.
409
410 Note that the extra_args are reset between the base, and new revision runs.
411
412 :param extra_args list: A list of extra arguments to pass to the try tasks.
413 """
414
415 def teardown_new_revision(self):
416 """Teardown the new run/revision setup."""
417
418 def teardown(self):
419 """Teardown for failures.
420
421 This method can be used for ensuring that the repo is cleaned up
422 when a failure is hit at any point in the process of doing the
423 new/base revision setups, or the pushes to try.
424 """
425
426 Frequently Asked Questions (FAQ)
427 --------------------------------
428
429 If you have any questions which aren't already answered below please reach out to us in the `perftest matrix channel <https://matrix.to/#/#perftest:mozilla.org>`_.
430
431 * **How can I tell what a category or a set of selections will run?**
432
433 At the moment, you need to run your command with an additional option to see what will be run: ``./mach try perf --no-push``. See the `Categories`_ section for more information about this. In the future, we plan on having an dynamically updated list for the tasks in the `Categories`_ section of this document.
434
435 * **What's the difference between ``Pageload desktop``, and ``Pageload desktop firefox``?**
436
437 If you simply ran ``./mach try perf`` with no additional options, then there is no difference. If you start adding additional browsers to the try run with commands like ``./mach try perf --chrome``, then ``Pageload desktop`` will select all tests available for ALL browsers available, and ``Pageload desktop firefox`` will only select Firefox tests. When ``--chrome`` is provided, you'll also see a ``Pageload desktop chrome`` option.
438
439 * **Help! I can't find a test in any of the categories. What should I do?**
440
441 Use the option ``--show-all`` or ``--full``. This will let you select tests from the ``./mach try fuzzy --full`` interface directly instead of the categories. You will always be able to find your tests this way. Please be careful with your task selections though as it's easy to run far too many tests in this way!
442
443 Future Work
444 -----------
445
446 The future work for this tool can be `found in this bug <https://bugzilla.mozilla.org/show_bug.cgi?id=1799178>`_. Feel free to file improvements, and bugs against it.