Optimize reboots with Android emulator initialization by jardondiego · Pull Request #5280 · google/clusterfuzz

jardondiego · 2026-05-16T16:02:31Z

Goal

Significantly improve the startup efficiency and fuzzing throughput of Android bots by eliminating redundant, time-consuming device reboots during the initialization and sanitizer setup phases.

Background

Historically, Android bots suffered from excessive hard reboots (~60s penalty each) during task execution:

Unconditional Final Reboots: initialize_device() unconditionally triggered a hard reboot at the end of its sequence to ensure a "good state," even if setup_asan_if_needed() had just performed a clean shell restart seconds prior.
Sanitizer Option Reboots: Setting ASan options (sanitizer.set_options) writes to the /system partition. The helper function adb.write_data_to_file() contained a hardcoded rule forcing a reboot anytime a file in /system was modified.

IvanBM18

LGTM, minus some NITs

fernandofloresg

lgtm, would be nice to add those docstrings

letitz

I'm a bit confused about the main logic change in initialize_device(), see comment below.

letitz · 2026-05-19T08:35:10Z

    return

-  adb.write_data_to_file(sanitizer_options, sanitizer_options_file_path)
+  # Skip reboot as the app will pick up the options file on restart.


Just to make sure I understand, this means that the next time the app is started, it will read from the options file?

Yes. We write the ASan options to the system file. When the target fuzzer application is launched in a new process shortly after this, the new options file are read from disk during its initialization. A full device reboot isn't required for this step, so we can safely skip it to save time. Forcing a ~60s kernel reboot to update a text string that a process is going to read 2 seconds later is redundant.

I see, thanks. IIUC the logic in write_data_to_file(), that means we'll leave /system in read-write mode. Is that intentional?

Yes. I don't think it would be a problem for emulated devices since they would be launched and destroyed in every task bit it might be an issue for non-virtual devices since a fuzzing job could break a system file and we would have to address that issue in a physical device, and that could mean having to physically check such device. We could either skip this optimization for virtual devices or manually trigger a read-only re-mounting. What do you suggest?

Option A: ea78515

Option B: ccffa14

I applied Option A for now, but I'm all ears for any suggestions from you. Thanks!

See other comment, I don't think option A was applied in the end.

In any case, I don't think it's a big deal to lose 1m per fuzzing session - they last 3 hours anyway. I would rather go for option B and avoid the risk of weird issues where /system files get corrupted. I agree with you that for emulated devices that should be safe, but I'm not 100% confident :)

I think something weird happened in my workspace, I was moving around several branches and got confused. Agreed! I will adjust as suggested!

letitz

One substantive comment around skipping reboots after writing ASan options, otherwise just two small nits.

letitz · 2026-05-21T09:35:41Z

    return

-  adb.write_data_to_file(sanitizer_options, sanitizer_options_file_path)
+  # Skip reboot as the app will pick up the options file on restart.


I see, thanks. IIUC the logic in write_data_to_file(), that means we'll leave /system in read-write mode. Is that intentional?

This change reduces the number of reboots during Android device setup by: - Adding a wait_for_reboot parameter to adb.write_data_to_file. - Tracking reboot status in initialize_device to skip the final reboot if one already occurred. - Disabling reboots when setting sanitizer options since the app restart is sufficient. These optimizations improve bot startup efficiency and overall fuzzing throughput.

… tracking and short-circuit issues

Adds `-> bool` return type hints and explanatory docstrings to: - `configure_system_build_properties` - `setup_asan_if_needed` These updates clarify that the returned boolean indicates whether a device reboot occurred during the setup step.

- Revert configure_system_build_properties to not return a bool. - Simplify initialize_device to only reboot if ASan setup did not reboot the device. This fixes the bug where it would reboot even if no properties changed. - Refactor device_test.py to use helpers.patch instead of manual mock setups, matching the codebase style.

This restricts the should_reboot=False optimization to emulators only. Physical devices will always fall back to a hard reboot after modifying a system file to guarantee the /system partition is safely locked back to read-only.

Removed redundant 'helpers' import in AddTestAccountsIfNeededTest.setUp and fixed spacing.

letitz

LGTM % option b and actually fixing the docstring for write_data_to_file().

letitz · 2026-05-22T08:24:18Z

    return

-  adb.write_data_to_file(sanitizer_options, sanitizer_options_file_path)
+  # Skip reboot as the app will pick up the options file on restart.


See other comment, I don't think option A was applied in the end.

In any case, I don't think it's a big deal to lose 1m per fuzzing session - they last 3 hours anyway. I would rather go for option B and avoid the risk of weird issues where /system files get corrupted. I agree with you that for emulated devices that should be safe, but I'm not 100% confident :)

Previously, the initialization sequence forced physical Android devices to perform a full kernel reboot after modifying the /system partition (e.g., when updating asan.options) to guarantee the partition was safely locked back to read-only. This commit transitions from that conservative approach (Option A) to a shell-based fallback (Option B). Now, when the 'should_reboot=False' optimization is requested, both emulators AND physical devices skip the 60+ second reboot penalty. To maintain security and prevent accidental filesystem corruption from rogue fuzzers, the script explicitly locks the partition back down using the 'mount -o ro,remount /system' shell command.

jardondiego requested review from IvanBM18, Xeicker and fernandofloresg May 16, 2026 16:05

jardondiego marked this pull request as ready for review May 17, 2026 00:44

jardondiego requested a review from a team as a code owner May 17, 2026 00:44

IvanBM18 reviewed May 18, 2026

View reviewed changes

Comment thread src/clusterfuzz/_internal/platforms/android/device.py

IvanBM18 reviewed May 18, 2026

View reviewed changes

Comment thread src/clusterfuzz/_internal/platforms/android/device.py

IvanBM18 reviewed May 18, 2026

View reviewed changes

Comment thread src/clusterfuzz/_internal/platforms/android/sanitizer.py Outdated

IvanBM18 approved these changes May 18, 2026

View reviewed changes

jardondiego requested a review from letitz May 18, 2026 19:27

fernandofloresg approved these changes May 18, 2026

View reviewed changes

Comment thread src/clusterfuzz/_internal/platforms/android/device.py

letitz requested changes May 19, 2026

View reviewed changes

jardondiego changed the title ~~Optimize Android emulator initialization by batching reboots~~ Optimize reboots with Android emulator initialization May 20, 2026

jardondiego requested review from ViniciustCosta and decoNR May 20, 2026 17:16

jardondiego force-pushed the optimize-android-reboots branch from bff881e to 2888eab Compare May 21, 2026 02:09

jardondiego requested a review from letitz May 21, 2026 02:28

letitz reviewed May 21, 2026

View reviewed changes

Diego Jardon added 12 commits May 21, 2026 18:33

Fix reboot_done tracking to ensure device settings are applied

8c579ed

Revert previous fix and use accumulated state for reboot_done

af62e75

Invert logic to needs_reboot and use bitwise OR operator to fix state…

a3c840d

… tracking and short-circuit issues

Fix needs_reboot logic so ASan setup clears pending reboots

c2fd3e2

Fix initialization reboot logic to correctly preserve base requirements

09273e2

Add unit tests for android device reboot logic

d6ab58f

Fix formatting issues from linter in device_test.py

4b340a5

Rename wait_for_reboot to should_reboot

9d8cd5c

Update docstrings for test cases

baf5766

Diego Jardon added 2 commits May 21, 2026 18:33

Add descriptive docstring for InitializeEnvironmentTest

48f3468

Add docstring to write_data_to_file

2c08a43

jardondiego force-pushed the optimize-android-reboots branch from 2888eab to 2c08a43 Compare May 21, 2026 18:34

Diego Jardon and others added 5 commits May 21, 2026 18:36

Move helpers import to module level in device_test.py

336400b

Ensure physical devices always reboot to lock /system

ccffa14

This restricts the should_reboot=False optimization to emulators only. Physical devices will always fall back to a hard reboot after modifying a system file to guarantee the /system partition is safely locked back to read-only.

Merge branch 'master' into optimize-android-reboots

829ac70

Fix linter errors in device_test.py

c347ad6

Removed redundant 'helpers' import in AddTestAccountsIfNeededTest.setUp and fixed spacing.

Merge branch 'master' into optimize-android-reboots

baec846

letitz approved these changes May 22, 2026

View reviewed changes

Diego Jardon added 2 commits May 22, 2026 14:59

Update docstring for write_data_to_file based on PR feedback

511b555

Conversation

jardondiego commented May 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Goal

Background

Uh oh!

Uh oh!

Uh oh!

Uh oh!

IvanBM18 left a comment

Choose a reason for hiding this comment

Uh oh!

fernandofloresg left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

letitz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jardondiego May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

letitz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

letitz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

jardondiego commented May 16, 2026 •

edited

Loading

jardondiego May 21, 2026 •

edited

Loading