JMR-dev/LibreMediaConverter
1
0
Fork 0
Code Issues 16 Pull Requests Actions Packages Projects Releases Wiki Activity

Drop API 37 from the E2E matrix and write down why

Browse Source 
The android-37.0 emulator image crash-loops surfaceflinger inside its own
gralloc mapper: RegionSamplingThread calls GraphicBuffer::lock, which reaches
GoldfishMapper::readFromHost, which asserts that the host has not negotiated
ReadColorBufferDma. It has, so surfaceflinger aborts, restarts, and aborts
again. Nothing this app does can survive that, and it reproduces on a GitHub
runner under swiftshader_indirect and on a workstation under -gpu host alike.

There is no ATD image at android-37.0 to fall back to, and -feature -GLDMA is
accepted by the emulator but does not prevent the assertion.

Correcting the previous commit, which is already pushed so its message stands:
ram-size was not the cause of that failure. Setting it did move the job from
failing at install to failing during the test run, which is how the real
crash became visible, but at 2560M the guest had 1.5 GB free when it died.
The setting is kept because the emulator's own floor varies by API level --
2048M at 33, 2560M at 34 to 36 -- and pinning it makes the matrix uniform.

Also corrected: a comment claiming this could not be reproduced locally. It
can, and the local crash was the same one all along.

Dropped the dmesg probe. adb shell is not root, so klogctl is denied and it
only ever printed a permission error -- which a later reader would reasonably
misread as "no OOM kills".

docs/api-37-emulator-crash.md carries the evidence, the ruled-out fixes, the
reproduction, and how to file it upstream, so re-adding the row later starts
from what is already known rather than from scratch.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
JMR-dev
2026-08-21 21:35:04 -05:00
co-authored by Claude Opus 5
parent 2fe1aa9f9c
commit 4e6fe6b75a
2 changed files with 212 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/status_check.yml
+27 -27
View File
@@ -103,9 +103,12 @@ jobs:
# ---------------------------------------------------------------------------
# One runner per API level, across the whole supported range.
#
# The range is the point: minSdk is 33 and targetSdk is 37, and the foreground
# service type differs across it -- none below 34, dataSync at 34, mediaProcessing
# from 35. Testing a single level would leave two thirds of that branch unexercised.
# The range is the point: minSdk is 33, and the foreground service type differs
# across it -- none below 34, dataSync at 34, mediaProcessing from 35. Testing a
# single level would leave two thirds of that branch unexercised.
#
# It stops at 36 rather than targetSdk 37 because the android-37.0 emulator image
# is broken, not because 37 does not matter. See docs/api-37-emulator-crash.md.
#
# FFmpeg is not built here. The AAR is committed under bin/, so a red run means the
# code is broken rather than that a cross-compile hiccuped.
@@ -130,12 +133,13 @@ jobs:
api-level: "35"
- label: "36"
api-level: "36"
# API 37 ships as android-37.0. The emulator action installs
# "platforms;android-${api-level}" and there is no platforms;android-37,
# so a bare 37 fails during SDK setup before an emulator ever starts.
# system-image-api-level alone does not fix it: that only names the image.
- label: "37"
api-level: "37.0"
# No API 37 row. targetSdk is 37, but the android-37.0 emulator image
# crash-loops surfaceflinger inside its own gralloc mapper, so every test
# fails there no matter what this app does. Ruling that in took four CI
# rounds, so the evidence and the ruled-out fixes are written down rather
# than left to be rediscovered: docs/api-37-emulator-crash.md. That file
# also records what to try first when re-adding it -- note that the row
# needs api-level "37.0", since a bare 37 fails during SDK setup.
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
@@ -172,29 +176,25 @@ jobs:
emulator-options: -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none
disable-animations: true
# The default userdata partition is not big enough for this APK once the
# FFmpeg libraries are in it. API 37's system image leaves the least room and
# was the level that actually failed, with "Requested internal only, but not
# enough space" -- but the margin was thin everywhere, so give them all room.
# FFmpeg libraries are in it. One level failed outright with "Requested
# internal only, but not enough space", and the margin was thin everywhere
# else, so give them all room.
disk-size: 8G
# The emulator raises an undersized guest to 2560M by itself, but only for
# API levels it recognises, and it does not recognise "37.0". Every green job
# in this matrix was quietly running at 2560M while API 37 ran at the pixel_6
# default of 1536M, lost system_server partway through installing the 82 MB
# APK, and reported it as "Can't find service: package". 2560M is not a guess
# at a sufficient value: it is the value the other four levels already pass
# at. Setting it explicitly makes the matrix uniform instead of leaving one
# level at the mercy of that heuristic.
# Pinned because the emulator's own default is not uniform: it raises an
# undersized guest to a floor that varies by API level -- 2048M at 33, 2560M
# at 34 through 36 -- and skips levels it does not recognise entirely. 2560M
# is the highest of those floors, so no level gets less memory than it
# already had, and none of them depend on that heuristic any more.
ram-size: 2560M
# Build only the ABI the emulator can execute. FFmpeg's native libraries
# dominate the APK, so shipping arm64 to an x86_64 emulator doubles the
# install for code that can never run: 114 MB against 80 MB.
#
# The memory probes exist because this failure cannot be reproduced locally:
# API 37 will not boot on a workstation under either GPU mode -- host aborts
# surfaceflinger in the goldfish mapper, swiftshader_indirect segfaults the
# emulator. CI is the only instrument, so it has to report enough to be
# conclusive. The first line proves what the guest actually got regardless of
# the result; the rest runs only on failure, so a green run is unchanged.
# The probe lines survive from diagnosing the API 37 crash and are kept
# because a red instrumented run is otherwise near-impossible to read from a
# log alone. The first reports what the guest actually got, so a wrong
# emulator configuration is visible on a green run too; the crash dump runs
# only on failure, so a green run is unchanged.
#
# Each line here is a separate `sh -c` -- the action splits the script on
# newlines -- so the failure handler has to stay on one line. The action does
@@ -203,7 +203,7 @@ jobs:
# diagnostic must never be the thing that turns a run red.
script: |
adb shell cat /proc/meminfo | grep -E 'MemTotal|MemAvailable|SwapTotal' || true
./gradlew :app:connectedDebugAndroidTest -PabiFilters=x86_64 || { echo "=== guest memory at failure ==="; adb shell cat /proc/meminfo | grep -E 'MemTotal|MemAvailable|SwapTotal'; echo "=== kernel OOM kills ==="; adb shell dmesg | grep -iE 'oom|lowmemory|killed process' | tail -20; echo "=== native crashes ==="; adb logcat -d -b crash | tail -40; exit 1; }
./gradlew :app:connectedDebugAndroidTest -PabiFilters=x86_64 || { echo "=== guest memory at failure ==="; adb shell cat /proc/meminfo | grep -E 'MemTotal|MemAvailable|SwapTotal'; echo "=== native crashes ==="; adb logcat -d -b crash | tail -60; exit 1; }
- uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
if: always()