This set of pages documents the setup and operation of the GPU bots and try servers, which verify the correctness of Chrome's graphically accelerated rendering pipeline.
The GPU bots run a different set of tests than the majority of the Chromium test machines. The GPU bots specifically focus on tests which exercise the graphics processor, and whose results are likely to vary between graphics card vendors.
Most of the tests on the GPU bots are run via the Telemetry framework. Telemetry was originally conceived as a performance testing framework, but has proven valuable for correctness testing as well. Telemetry directs the browser to perform various operations, like page navigation and test execution, from external scripts written in Python. The GPU bots launch the full Chromium browser via Telemetry for the majority of the tests. Using the full browser to execute tests, rather than smaller test harnesses, has yielded several advantages: testing what is shipped, improved reliability, and improved performance.
A subset of the tests, called “pixel tests”, grab screen snapshots of the web page in order to validate Chromium's rendering architecture end-to-end. Where necessary, GPU-specific results are maintained for these tests. Some of these tests verify just a few pixels, using handwritten code, in order to use the same validation for all brands of GPUs.
The GPU bots use the Chrome infrastructure team‘s recipe framework, and specifically the chromium and chromium_trybot recipes, to describe what tests to execute. Compared to the legacy master-side buildbot scripts, recipes make it easy to add new steps to the bots, change the bots’ configuration, and run the tests locally in the same way that they are run on the bots. Additionally, the chromium and chromium_trybot recipes make it possible to send try jobs which add new steps to the bots. This single capability is a huge step forward from the previous configuration where new steps were added blindly, and could cause failures on the tryservers. For more details about the configuration of the bots, see the GPU bot details.
The physical hardware for the GPU bots lives in the Swarming pool*. The Swarming infrastructure (new docs, older but currently more complete docs) provides many benefits:
(* All but a few one-off GPU bots are in the swarming pool. The exceptions to the rule are described in the GPU bot details.)
The bots on the chromium.gpu.fyi waterfall are configured to always test top-of-tree ANGLE. This setup is done with a few lines of code in the tools/build workspace; search the code for “angle”.
These aspects of the bots are described in more detail below, and in linked pages. There is a presentation which gives a brief overview of this documentation and links back to various portions.
Please see the GPU Pixel Wrangling instructions for links to dashboards showing the status of various bots in the GPU fleet.
Most Chromium developers interact with the GPU bots in two ways:
The GPU bots are grouped on the chromium.gpu and chromium.gpu.fyi waterfalls. Their current status can be easily observed there.
To send try jobs, you must first upload your CL to the codereview server. Then, either clicking the “CQ dry run” link or running from the command line:
git cl try
Sends your job to the default set of try servers.
The GPU tests are part of the default set for Chromium CLs, and are run as part of the following tryservers' jobs:
tryserver.chromium.linux waterfalltryserver.chromium.mac waterfalltryserver.chromium.win waterfallScan down through the steps looking for the text “GPU”; that identifies those tests run on the GPU bots. For each test the “trigger” step can be ignored; the step further down for the test of the same name contains the results.
It's usually not necessary to explicitly send try jobs just for verifying GPU tests. If you want to, you must invoke “git cl try” separately for each tryserver master you want to reference, for example:
git cl try -b linux-rel git cl try -b mac-rel git cl try -b win7-rel
Alternatively, the Gerrit UI can be used to send a patch set to these try servers.
Three optional tryservers are also available which run additional tests. As of this writing, they ran longer-running tests that can't run against all Chromium CLs due to lack of hardware capacity. They are added as part of the included tryservers for code changes to certain sub-directories.
Tryservers for the ANGLE project are also present on the tryserver.chromium.angle waterfall. These are invoked from the Gerrit user interface. They are configured similarly to the tryservers for regular Chromium patches, and run the same tests that are run on the chromium.gpu.fyi waterfall, in the same way (e.g., against ToT ANGLE).
If you find it necessary to try patches against other sub-repositories than Chromium (src/) and ANGLE (src/third_party/angle/), please file a bug with component Internals>GPU>Testing.
All of the GPU tests running on the bots can be run locally from a Chromium build. Many of the tests are simple executables:
angle_unittestsgl_testsgl_unitteststab_capture_end2end_testsSome run only on the chromium.gpu.fyi waterfall, either because there isn‘t enough machine capacity at the moment, or because they’re closed-source tests which aren't allowed to run on the regular Chromium waterfalls:
angle_deqp_gles2_testsangle_deqp_gles3_testsangle_end2end_testsaudio_unittestsThe remaining GPU tests are run via Telemetry. In order to run them, just build the chrome target and then invoke src/content/test/gpu/run_gpu_integration_test.py with the appropriate argument. The tests this script can invoke are in src/content/test/gpu/gpu_tests/. For example:
run_gpu_integration_test.py context_lost --browser=releaserun_gpu_integration_test.py webgl_conformance --browser=release --webgl-conformance-version=1.0.2run_gpu_integration_test.py maps --browser=releaserun_gpu_integration_test.py screenshot_sync --browser=releaserun_gpu_integration_test.py trace_test --browser=releaseThe pixel tests are a bit special. See the section on running them locally for details.
If you're testing on Android and have built and deployed ChromePublic.apk to the device, use --browser=android-chromium to invoke it.
Note: The tests require some third-party Python packages. Obtaining these packages is handled automatically by vpython, and the script‘s shebang should use vpython if running the script directly. If you’re used to invoking python to run a script, simply use vpython instead, e.g. vpython run_gpu_integration_test.py ....
You can run a subset of tests with this harness:
run_gpu_integration_test.py webgl_conformance --browser=release --test-filter=conformance_attribsThe exact command used to invoke the test on the bots can be found in one of two ways:
requests[task_slices][command]. The arguments after the last -- are used to actually run the test.In both cases, the following can be omitted when running locally since they're only necessary on swarming:
testing/test_env.pytesting/scripts/run_gpu_integration_test_as_googletest.py--isolated-script-test-output--isolated-script-test-perf-outputThe Maps test requires you to authenticate to cloud storage in order to access the Web Page Reply archive containing the test. See Cloud Storage Credentials for documentation on setting this up.
The Telemetry-based tests are all technically the same target, telemetry_gpu_integration_test, just run with different runtime arguments. The first positional argument passed determines which suite will run, and additional runtime arguments may cause the step name to change on the bots. Here is a list of all suites and resulting step names as of April 15th 2021:
context_lostcontext_lost_passthrough_testscontext_lost_testscontext_lost_validating_testsgl_renderer_context_lost_testsdepth_capturedepth_capture_testsgl_renderer_depth_capture_testshardware_accelerated_featuregl_renderer_hardware_accelerated_feature_testshardware_accelerated_feature_testsgpu_processgl_renderer_gpu_process_launch_testsgpu_process_launch_testsinfo_collectioninfo_collection_testsmapsgl_renderer_maps_pixel_testsmaps_pixel_passthrough_testmaps_pixel_testmaps_pixel_validating_testmaps_testspixelandroid_webview_pixel_skia_gold_testdawn_pixel_skia_gold_testegl_pixel_skia_gold_testgl_renderer_pixel_skia_gold_testspixel_skia_gold_passthrough_testpixel_skia_gold_validating_testpixel_testsskia_renderer_pixel_skia_gold_testvulkan_pixel_skia_gold_testpowerpower_measurement_testscreenshot_syncgl_renderer_screenshot_sync_testsscreenshot_sync_passthrough_testsscreenshot_sync_testsscreenshot_sync_validating_teststrace_testtrace_testwebgl_conformancewebgl2_conformance_d3d11_passthrough_testswebgl2_conformance_gl_passthrough_testswebgl2_conformance_gles_passthrough_testswebgl2_conformance_testswebgl2_conformance_validating_testswebgl_conformance_d3d11_passthrough_testswebgl_conformance_d3d9_passthrough_testswebgl_conformance_fast_call_testswebgl_conformance_gl_passthrough_testswebgl_conformance_gles_passthrough_testswebgl_conformance_metal_passthrough_testswebgl_conformance_swangle_passthrough_testswebgl_conformance_swiftshader_validating_testswebgl_conformance_testswebgl_conformance_validating_testswebgl_conformance_vulkan_passthrough_testsThe pixel tests are a special case because they use an external Skia service called Gold to handle image approval and storage. See GPU Pixel Testing With Gold for specifics.
TL;DR is that the pixel tests use a binary called goldctl to download and upload data when running pixel tests.
Normally, goldctl uploads images and image metadata to the Gold server when used. This is not desirable when running locally for a couple reasons:
Additionally, the tests normally rely on the Gold server for viewing images produced by a test run. This does not work if the data is not actually uploaded.
The pixel tests contain logic to automatically determine whether they are running on a workstation or not, as well as to determine what git revision is being tested. This should mean that the pixel tests will automatically work when run locally. However, if the local run detection code fails for some reason, you can manually pass some flags to force the same behavior:
In order to get around the local run issues, simply pass the --local-pixel-tests flag to the tests. This will disable uploading, but otherwise go through the same steps as a test normally would. Each test will also print out file:// URLs to the produced image, the closest image for the test known to Gold, and the diff between the two.
Because the image produced by the test locally is likely slightly different from any of the approved images in Gold, local test runs are likely to fail during the comparison step. In order to cut down on the amount of noise, you can also pass the --no-skia-gold-failure flag to not fail the test on a failed image comparison. When using --no-skia-gold-failure, you'll also need to pass the --passthrough flag in order to actually see the link output.
Example usage: run_gpu_integration_test.py pixel --no-skia-gold-failure --local-pixel-tests --passthrough
If, for some reason, the local run code is unable to determine what the git revision is, simply pass --git-revision aabbccdd. Note that aabbccdd must be replaced with an actual Chromium src revision (typically whatever revision origin/main is currently synced to) in order for the tests to work. This can be done automatically using: run_gpu_integration_test.py pixel --no-skia-gold-failure --local-pixel-tests --passthrough --git-revision `git rev-parse origin/main`
Any binary run remotely on a bot can also be run locally, assuming the local machine loosely matches the architecture and OS of the bot.
The easiest way to do this is to find the ID of the swarming task and use “swarming.py reproduce” to re-run it:
./src/tools/luci-go/swarming reproduce -S https://chromium-swarm.appspot.com [task ID]The task ID can be found in the stdio for the “trigger” step for the test. For example, look at a recent build from the Mac Release (Intel) bot, and look at the gl_unittests step. You will see something like:
Triggered task: gl_unittests on Intel GPU on Mac/Mac-10.12.6/[TRUNCATED_ISOLATE_HASH]/Mac Release (Intel)/83664 To collect results, use: swarming.py collect -S https://chromium-swarm.appspot.com --json /var/folders/[PATH_TO_TEMP_FILE].json Or visit: https://chromium-swarm.appspot.com/user/task/[TASK_ID]
There is a difference between the isolate‘s hash and Swarming’s task ID. Make sure you use the task ID and not the isolate's hash.
As of this writing, there seems to be a bug when attempting to re-run the Telemetry based GPU tests in this way. For the time being, this can be worked around by instead downloading the contents of the isolate. To do so, look into the “Reproducing the task locally” section on a swarming task, which contains something like:
Download inputs files into directory foo:
# (if needed, use "\${platform}" as-is) cipd install "infra/tools/luci/cas/\${platform}" -root bar
# (if needed) ./bar/cas login
./bar/cas download -cas-instance projects/chromium-swarm/instances/default_instance -digest 68ae1d6b22673b0ab7b4427ca1fc2a4761c9ee53474105b9076a23a67e97a18a/647 -dir foo
Before attempting to download an isolate, you must ensure you have permission to access the isolate server. Full instructions can be found here. For most cases, you can simply run:
./src/tools/luci-go/isolate loginThe above link requires that you log in with your @google.com credentials. It‘s not known at the present time whether this works with @chromium.org accounts. Email kbr@ if you try this and find it doesn’t work.
See the Swarming documentation for instructions on how to upload your binaries to the isolate server and trigger execution on Swarming.
Be sure to use the correct swarming dimensions for your desired GPU e.g. “1002:6613” instead of “AMD Radeon R7 240 (1002:6613)” which is how it appears on swarming task page. You can query bots in the chromium.tests.gpu pool to find the correct dimensions:
tools\luci-go\swarming bots -S chromium-swarm.appspot.com -d pool=chromium.tests.gpuTo create a zip archive of your personal Chromium build plus all of the Telemetry-based GPU tests' dependencies, which you can then move to another machine for testing:
out/Release in this example).vpython tools/mb/mb.py zip out/Release/ telemetry_gpu_integration_test out/telemetry_gpu_integration_test.zipThen copy telemetry_gpu_integration_test.zip to another machine. Unzip it, and cd into the resulting directory. Invoke content/test/gpu/run_gpu_integration_test.py as above.
This workflow has been tested successfully on Windows with a statically-linked Release build of Chrome.
Note: on one macOS machine, this command failed because of a broken strip-json-comments symlink in src/third_party/catapult/common/node_runner/node_runner/node_modules/.bin. Deleting that symlink allowed it to proceed.
Note also: on the same macOS machine, with a component build, this command failed to zip up a working Chromium binary. The browser failed to start with the following error:
[0626/180440.571670:FATAL:chrome_main_delegate.cc(1057)] Check failed: service_manifest_data_pack_.
In a pinch, this command could be used to bundle up everything, but the “out” directory could be deleted from the resulting zip archive, and the Chromium binaries moved over to the target machine. Then the command line arguments --browser=exact --browser-executable=[path] can be used to launch that specific browser.
See the user guide for mb, the meta-build system, for more details.
The goal of the GPU bots is to avoid regressions in Chrome‘s rendering stack. To that end, let’s add as many tests as possible that will help catch regressions in the product. If you see a crazy bug in Chrome's rendering which would be easy to catch with a pixel test running in Chrome and hard to catch in any of the other test harnesses, please, invest the time to add a test!
There are a couple of different ways to add new tests to the bots:
Adding new tests to the GTest-based harnesses is straightforward and essentially requires no explanation.
As of this writing it isn‘t as easy as desired to add a new test to one of the Telemetry based harnesses. See Issue 352807. Let’s collectively work to address that issue. It would be great to reduce the number of steps on the GPU bots, or at least to avoid significantly increasing the number of steps on the bots. The WebGL conformance tests should probably remain a separate step, but some of the smaller Telemetry based tests (context_lost_tests, memory_test, etc.) should probably be combined into a single step.
If you are adding a new test to one of the existing tests (e.g., pixel_test), all you need to do is make sure that your new test runs correctly via isolates. See the documentation from the GPU bot details on adding new isolated tests for the gn args and authentication needed to upload isolates to the isolate server. Most likely the new test will be Telemetry based, and included in the telemetry_gpu_test_run isolate.
The tests that are run by the GPU bots are described by a couple of JSON files in the Chromium workspace:
These files are autogenerated by the following script:
This script is documented in testing/buildbot/README.md. The JSON files are parsed by the chromium and chromium_trybot recipes, and describe two basic types of tests:
base/test/launcher/ frameworks.The majority of the GPU tests are however:
A prerequisite of adding a new test to the bots is that that test run via isolates. Once that is done, modify test_suites.pyl to add the test to the appropriate set of bots. Be careful when adding large new test steps to all of the bots, because the GPU bots are a limited resource and do not currently have the capacity to absorb large new test suites. It is safer to get new tests running on the chromium.gpu.fyi waterfall first, and expand from there to the chromium.gpu waterfall (which will also make them run against every Chromium CL by virtue of the linux-rel, mac-rel, win7-rel and android-marshmallow-arm64-rel tryservers' mirroring of the bots on this waterfall – so be careful!).
Tryjobs which add new test steps to the chromium.gpu.json file will run those new steps during the tryjob, which helps ensure that the new test won't break once it starts running on the waterfall.
Tryjobs which modify chromium.gpu.fyi.json can be sent to the win_optional_gpu_tests_rel, mac_optional_gpu_tests_rel and linux_optional_gpu_tests_rel tryservers to help ensure that they won't break the FYI bots.
If pixel tests fail on the bots, the build step will contain either one or more links titled gold_triage_link for <test name> or a single link titled Too many artifacts produced to link individually, click for links, which itself will contain links. In either case, these links will direct to Gold pages showing the image produced by the image and the approved image that most closely matches it.
Note that for the tests which programatically check colors in certain regions of the image (tests with expected_colors fields in pixel_test_pages), there likely won't be a closest approved image since those tests only upload data to Gold in the event of a failure.
If your CL adds a new pixel test or modifies existing ones, it's likely that you will have to approve new images. Simply run your CL through the CQ and follow the steps outline here under the “Check if any pixel test failures are actual failures or need to be rebaselined.” step.
If you are adding a new pixel test, it is beneficial to set the grace_period_end argument in the test‘s definition. This will allow the test to run for a period without actually failing on the waterfall bots, giving you some time to triage any additional images that show up on them. This helps prevent new tests from making the bots red because they’re producing slightly different but valid images from the ones triaged while the CL was in review. Example:
from datetime import date ... PixelTestPage( 'foo_pixel_test.html', ... grace_period_end=date(2020, 1, 1) )
You should typically set the grace period to end 1-2 days after the the CL will land.
Once your CL passes the CQ, you should be mostly good to go, although you should keep an eye on the waterfall bots for a short period after your CL lands in case any configurations not covered by the CQ need to have images approved, as well. All untriaged images for your test can be found by substituting your test name into:
https://chrome-gpu-gold.skia.org/search?query=name%3D<test name>
NOTE If you have a grace period active for your test, then Gold will be told to ignore results for the test. This is so that it does not comment on unrelated CLs about untriaged images if your test is noisy. Images will still be uploaded to Gold and can be triaged, but will not show up on the main page's untriaged image list, and you will need to enable the “Ignored” toggle at the top of the page when looking at the triage page specific to your test.
It‘s critically important to aggressively investigate and eliminate the root cause of any flakiness seen on the GPU bots. The bots have been known to run reliably for days at a time, and any flaky failures that are tolerated on the bots translate directly into instability of the browser experienced by customers. Critical bugs in subsystems like WebGL, affecting high-profile products like Google Maps, have escaped notice in the past because the bots were unreliable. After much re-work, the GPU bots are now among the most reliable automated test machines in the Chromium project. Let’s keep them that way.
Flakiness affecting the GPU tests can come in from highly unexpected sources. Here are some examples:
sem_post/sem_wait primitives breaking V8’s parallel garbage collection (Issue 609249).If you notice flaky test failures either on the GPU waterfalls or try servers, please file bugs right away with the component Internals>GPU>Testing and include links to the failing builds and copies of the logs, since the logs expire after a few days. GPU pixel wranglers should give the highest priority to eliminating flakiness on the tree.