Repository navigation
android_hardware_smoke_test: Improve reliability - #191374
Conversation
…king, engine cache eviction, and infrastructure failure exit codes
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request refactors the Android hardware smoke tests to improve EGL failure detection and engine cache management. It reduces retry attempts, checks both logcat and driver output for EGL errors, and introduces an engine cache eviction mechanism. Feedback recommends making the engine cache eviction synchronous when called from non-main threads to prevent race conditions during retries, and setting the window pixel format before calling super.onCreate to avoid initialization issues.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request refactors the Android hardware smoke tests to improve reliability and EGL failure handling. Key changes include locking the window pixel format to 8-bit sRGB to prevent negotiation errors, updating the retry loop to check both logcat and driver output for EGL failures, and introducing a centralized evictEngineCache method to clear the cached Flutter engine. The review feedback suggests simplifying the concurrency logic in evictEngineCache by running it directly on the main thread and wrapping its invocations in the test class with runOnMainSync to prevent potential CI hangs.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request refactors the Android hardware smoke tests to improve EGL/graphics failure detection and recovery. Key changes include locking the window pixel format to 8-bit sRGB in MainActivity to prevent format negotiation errors, introducing an engine cache eviction mechanism on failures, and updating the test runner to inspect both logcat and driver output for EGL pipeline collapses. Feedback suggests expanding the driver output check in the test runner to include EglInitializationException so that explicit initialization failures are correctly caught and handled.
There was a problem hiding this comment.
Code Review
This pull request updates the Android hardware smoke tests by locking the window pixel format to 8-bit sRGB (RGBA_8888) in MainActivity to prevent SwiftShader negotiation errors, introducing a centralized engine cache eviction helper, and refactoring the retry loop in the test runner to check both logcat and driver output for EGL failures. Feedback was provided regarding the direct call to io.exit(2), which can bypass critical cleanup steps (such as restoring AndroidManifest.xml in finally blocks), suggesting throwing a custom exception instead.
b-luk
left a comment
There was a problem hiding this comment.
LGTM other than a small nit
Roll Flutter from c2437523d308 to 65c9a8dc60bc (195 revisions) flutter/flutter@c243752...65c9a8d 2026-08-22 [email protected] Roll ICU from d578f2e8b7bd to 8cc91d9b6ab9 (1 revision) (flutter/flutter#191542) 2026-08-22 [email protected] Roll Skia from 666a9b3d5cf0 to ad2106c0bb64 (3 revisions) (flutter/flutter#191527) 2026-08-22 [email protected] Roll Fuchsia Test Scripts from KaOq3EE4qJ9fnaaaK... to 0iCv10IlKfiilEBOU... (flutter/flutter#191524) 2026-08-22 [email protected] [flutter_tools] Add tests for negative lookahead regex in test runner and batch entrypoints (flutter/flutter#191438) 2026-08-22 [email protected] Roll Skia from 0c37868737fa to 666a9b3d5cf0 (2 revisions) (flutter/flutter#191518) 2026-08-21 [email protected] Revert "[input] Migrate fuchsia.ui.pointerinjector to TouchSource (#190855) (flutter/flutter#191509) 2026-08-21 [email protected] Fixes windows gallery benchmarks by forcing mobile layout (flutter/flutter#191507) 2026-08-21 [email protected] [flutter_tools] Restrict WebAssetServer source resolution to source map extensions (flutter/flutter#191501) 2026-08-21 [email protected] Roll Skia from f6900c5b8439 to 0c37868737fa (2 revisions) (flutter/flutter#191504) 2026-08-21 [email protected] Refactor `FlutterDevice.connect` and VM service discovery (flutter/flutter#191221) 2026-08-21 [email protected] [flutter_tools] Fix crash when migrating flow-style exclude lists in analysis_options.yaml (flutter/flutter#191269) 2026-08-21 [email protected] [rules] Add packages/flutter_tools/gradle/AGENTS.md rules (flutter/flutter#191486) 2026-08-21 [email protected] [flutter_tools] refactor CLI argument architecture with typed option descriptors and bundles (PoC) (flutter/flutter#191018) 2026-08-21 [email protected] tools: Extract Dart SDK to temp directory before moving to final location (flutter/flutter#191263) 2026-08-21 [email protected] [web] Move CanvasKit fragment shader classes to canvaskit/fragment_shader.dart (flutter/flutter#191451) 2026-08-21 [email protected] Roll Fuchsia Linux SDK from GCQlmt6h-esJsNubS... to ic6GjOSn-KN508XyK... (flutter/flutter#191485) 2026-08-21 [email protected] [Widget Preview] Isolate PageStorage scope in widget preview group expansion tile (flutter/flutter#191378) 2026-08-21 [email protected] [flutter_tools] Deprecate --build and --no-build flags on flutter run (flutter/flutter#191358) 2026-08-21 [email protected] Roll Skia from 70988bed1b3b to f6900c5b8439 (2 revisions) (flutter/flutter#191481) 2026-08-21 [email protected] Roll Packages from 1785501 to 252bb33 (6 revisions) (flutter/flutter#191480) 2026-08-21 [email protected] Roll Fuchsia Linux SDK from 20IJas24bZiNmCZTK... to GCQlmt6h-esJsNubS... (flutter/flutter#191415) 2026-08-21 [email protected] Roll Skia from 2ba6971bd0d1 to 70988bed1b3b (2 revisions) (flutter/flutter#191476) 2026-08-21 [email protected] Roll Skia from 1d5f72537ba6 to 2ba6971bd0d1 (2 revisions) (flutter/flutter#191473) 2026-08-21 [email protected] Roll Skia from 2c25efd2e369 to 1d5f72537ba6 (1 revision) (flutter/flutter#191472) 2026-08-21 [email protected] Roll Skia from 09b1b810850a to 2c25efd2e369 (9 revisions) (flutter/flutter#191470) 2026-08-21 [email protected] [Impeller] fix position of cached single glyph text shadows (flutter/flutter#191325) 2026-08-21 [email protected] Roll Skia from abdf8821f313 to 09b1b810850a (15 revisions) (flutter/flutter#191458) 2026-08-21 [email protected] Document that programmatic TextEditingController changes do not run input formatters (flutter/flutter#190166) 2026-08-21 [email protected] Adds new gallery benchmarks to windows (skia and impeller) (flutter/flutter#191454) 2026-08-21 [email protected] iOS: Deprecate FlutterEngine.isGpuDisabled (flutter/flutter#191393) 2026-08-21 [email protected] Revert: [web] Unskip decoration image lerp tests (flutter/flutter#191462) 2026-08-20 [email protected] android_hardware_smoke_test: Improve reliability (flutter/flutter#191374) 2026-08-20 [email protected] Don't size or offset the Android platform view before it is laid out (flutter/flutter#190895) 2026-08-20 [email protected] [iOS][add2app] Skip building SwiftPM plugins when generating CocoaPods artifacts (flutter/flutter#190736) 2026-08-20 [email protected] Optimize JSONMessageCodec UTF-8 conversion (flutter/flutter#190529) 2026-08-20 [email protected] [FML] Replace deprecated wstring_convert with Win32 APIs (flutter/flutter#191394) 2026-08-20 [email protected] SliverFillRemaining extends beyond viewport size when fillOverscroll is true (flutter/flutter#191236) 2026-08-20 [email protected] [flutter_tools] Update argParser usageLineLength when --wrap-column is passed (flutter/flutter#191264) 2026-08-20 [email protected] [flutter_tools] Fix UNC path resolution in depfile parsing on Windows (flutter/flutter#191265) 2026-08-20 [email protected] [web] Unskip decoration image lerp tests (flutter/flutter#191426) 2026-08-20 [email protected] Do not inject 'type' and 'method' into service extension responses (flutter/flutter#190946) 2026-08-20 [email protected] [flutter_tools] Fix crash in symbolize command on stream error (flutter/flutter#191273) 2026-08-20 [email protected] [flutter_tools] Prevent deletion of shared native asset hooks outputs as stale (flutter/flutter#191272) 2026-08-20 [email protected] [flutter_tools] Support package wildcard assets in app pubspec (flutter/flutter#191266) 2026-08-20 [email protected] Add SliverClipRect and SliverClipRRect (flutter/flutter#179003) 2026-08-20 [email protected] [flutter_tools] Implement Diagnostics extension slice and doctor integration (flutter/flutter#191162) ...
flutter#192197) Promotes the `Linux_android_emu_vulkan_stable android_hardware_smoke_vulkan_instrumented_tests` and `Linux_android_emu android_hardware_smoke_opengles_instrumented_tests` shards out of bringup. [PR flutter#191374 ](flutter#191374) seems to have successfully resolved the common failure modes for these tests. I'm setting these as post-submit only because they're more expensive than the driver tests, which already run in presubmit and should catch most relevant problems. ## Pre-launch Checklist - [x] I read the [Contributor Guide] and followed the process outlined there for submitting PRs. - [x] I read the [AI contribution guidelines] and understand my responsibilities, or I am not using AI tools. - [x] I read the [Tree Hygiene] wiki page, which explains my responsibilities. - [x] I read and followed the [Flutter Style Guide], including [Features we expect every widget to implement]. - [x] I signed the [CLA]. - [x] I listed at least one issue that this PR fixes in the description above. - [x] I updated/added relevant in-code documentation (doc comments with `///`). - [x] If this PR introduces a new feature or capability, I created and linked a website documentation issue or PR in [flutter/website] (or verified none is needed). - [x] I added new tests to check the change I am making, or this PR is [test-exempt]. - [x] I followed the [breaking change policy] and added [Data Driven Fixes] where supported. - [x] All existing and new tests are passing. <!-- Links --> [Contributor Guide]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#overview [AI contribution guidelines]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#ai-contribution-guidelines [Tree Hygiene]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md [test-exempt]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#tests [Flutter Style Guide]: https://github.com/flutter/flutter/blob/main/docs/contributing/Style-guide-for-Flutter-repo.md [Features we expect every widget to implement]: https://github.com/flutter/flutter/blob/main/docs/contributing/Style-guide-for-Flutter-repo.md#features-we-expect-every-widget-to-implement [CLA]: https://cla.developers.google.com/ [flutter/tests]: https://github.com/flutter/tests [flutter/website]: https://github.com/flutter/website [breaking change policy]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#handling-breaking-changes [Discord]: https://github.com/flutter/flutter/blob/main/docs/contributing/Chat.md [Data Driven Fixes]: https://github.com/flutter/flutter/blob/main/docs/contributing/Data-driven-Fixes.md
Issue #191348 shows that the detection and retry behavior introduced in PR #190110 did not solve the issue.
Previously, engine cache cleanup was placed solely in
@AfterClass tearDownClass(). When a test run failed mid-suite on attempt 1, JUnit aborted before@AfterClasscould run. When attempt 2 launched,MainActivityretrieved the corrupted engine from cache, guaranteeing that subsequent attempts failed with blank screenshots, which matches what we see in the run history.To fix this, we move cache cleanup to
MainActivity.evictEngineCache().FlutterActivityTest.ktinvokes this cleanup during class teardown and whenever anEglInitializationExceptionorBlankScreenshotExceptionoccurs, ensuring any retry starts with a clean engine instance.We also reduce the retry cap in
run_android_hardware_smoke_tests.dartfrom 3 attempts to 2 to avoid 30-minute LUCI timeouts on broken emulators, which we saw some of in tryjobs.I think this should be an infrastructure failure. So if an unrecoverable EGL collapse occurs across both attempts, the runner logs an infrastructure failure and exits with code 2 (
INFRA_FAILURE)We also saw some logs which suggest HWUI repeatedly attempted 10-bit color format negotiation (
101010-2), which fails on SwiftShader CPU drivers. This could contribute to an increased rate of EGL errors, soMainActivity.onCreate()now explicitly locks the window pixel format toPixelFormat.RGBA_8888.Pre-launch Checklist
///).