Possible memory leak when shared content transition is involved [349646655]

Fixed

Bug

Status Update

No update yet.

Description

eb...@gmail.com

created issue #1

Jun 27, 2024 08:38AM

Jetpack Compose version: 1.7.0-beta04

Jetpack Compose component used: Both happened with navigation-compose 2.8.0-beta03 and also after the update, 2.8.0-beta04

Android Studio Build: 2024.1.1

Kotlin version: 2.0.0

Steps to Reproduce or Code Sample to Reproduce: This happens consistently on the CI but I couldn't reproduce it locally unfortunately, guess due to memory cap differences

Clone this project https://github.com/persian-calendar/persian-calendar (the app is a fully offline, not even internet permission, so should be safe to run locally)
Revert this change https://github.com/persian-calendar/persian-calendar/commit/aab44f3978d285aeacb9c7ea46910baf80c8c290
Run the UI tests, at least in the CI UI tests halt until timeout with this message https://github.com/persian-calendar/persian-calendar/actions/runs/9692947971/job/26747357826#step:6:297

Stack trace (if applicable):

com.byagowi.persiancalendar.ui.MapScreenTest > mapScreenNavigateUp[emulator-5554 - 10] FAILED 
	java.lang.OutOfMemoryError: Failed to allocate a 189720 byte allocation with 64296 free bytes and 62KB until OOM, target footprint 16777216, growth limit 16777216
	at java.util.Arrays.copyOf(Arrays.java:3136)

emulator-5554 - 10 Tests 30/43 completed. (0 skipped) (1 failed)

And this only happens when I reenable shared content transition for the map so I think something is wrong.

Comments

eb...@gmail.com <eb...@gmail.com> #2Jun 27, 2024 09:02AM

Replace the second step with git revert aab44f3978d285aeacb9c7ea46910baf80c8c290

Also happens with the revert of https://github.com/persian-calendar/persian-calendar/commit/a142fc343c7c6d7e1331d0245e6458cfbc7cb32d but with not memory related message https://github.com/persian-calendar/persian-calendar/actions/runs/9693459009/job/26748947134#step:6:292 only hangs forever. There are other shared content uses on the CI, weird that some are defect and some not.

eb...@gmail.com <eb...@gmail.com> #3Jun 27, 2024 05:19PM

Now more I think this has some similar root as https://issuetracker.google.com/issues/346608857 given this just times out on the second change and no mention of OOM.

(and I'll understand if you see this issue as invalid as it's my duty to replicate that locally first even though it's GitHub Actions and don't have access to the machine, still)

Message last modified on Jun 27, 2024 05:24PM

jo...@google.com <jo...@google.com> Jun 28, 2024 03:37PM

Assigned to jb...@google.com.

jb...@google.com <jb...@google.com> #4Jul 3, 2024 05:10PM

Assigned to ti...@google.com.

Happening with shared element transitions. Doris, any ideas here?

iv...@playbackbone.com <iv...@playbackbone.com> #5Jul 5, 2024 05:18PM

Sharing here in case it's the same issue, LeakCanary detected a memory leak as soon as I tried migrating a screen to use navigation-compose:2.8.0-beta04 . I've attached an (obfuscated) memory leak trace in case it's helpful.

Navigation Compose beta memory leak.txt

4.3 KB

View

Download

ti...@google.com <ti...@google.com> #6Jul 9, 2024 01:34PM

Could you confirm which version of animation library you are using? It'd be helpful to output a dependency tree gradle dependencies

eb...@gmail.com <eb...@gmail.com> #7Jul 9, 2024 05:41PM

This has all the dependencies informations, https://github.com/persian-calendar/persian-calendar/blob/main/PersianCalendar/runtime-dependencies-report.txt

eb...@gmail.com <eb...@gmail.com> #8Jul 9, 2024 05:53PM

androidx.compose.animation:animation-graphics-android was on 1.6.8 but upgraded it to androidx.compose.animation:animation:1.6.8 -> 1.7.0-beta04 (c) which still had the issue.

ti...@google.com <ti...@google.com> #9Jul 10, 2024 04:16PM

We fixed a memory leak in SeekableTransitionObserver in 1.7.0-beta04 here: https://issuetracker.google.com/336167334

Note: The lib that the fix landed in is androidx.compose.animation:animation-core.

Could you invalidate studio cache and double check that the animation-core:1.7.0-beta04 is the version being used when you see the leak?

ti...@google.com <ti...@google.com> #10Jul 10, 2024 04:30PM

More specifically, please make sure your test dependency (i.e. androidTestImplementation) is set up to use animation-core:1.7.0-beta04, since you mentioned in comment #1 that the leak showed up in CI.

eb...@gmail.com <eb...@gmail.com> #11Jul 10, 2024 08:15PM

First of all, thank you so much for the help, it's even possible that this is actually a combination of multiple bugs but since there is possibility of outdated library, is there a beta BoM so I can test and to make sure we are in sync?

In any case it was my duty to make this reproduce locally first. I should've provided a minimal case so I'll work on it just right now. Thanks

eb...@gmail.com <eb...@gmail.com> #12Jul 11, 2024 08:57AM

This is all dependencies of a run that is stuck https://github.com/persian-calendar/persian-calendar/actions/runs/9888317836/job/27311949949

So far I've failed to create a reduction :/

Message last modified on Jul 11, 2024 09:03AM

ti...@google.com <ti...@google.com> #13Jul 11, 2024 12:28PM

Thanks for the update.

Some investigation has revealed the aforementioned fix for this leak in comment #9 did not make it into 1.7 beta. We are requesting for it to be included in the next beta release. Sorry about the confusion.

ti...@google.com <ti...@google.com> #14Aug 9, 2024 10:09PM

The fix has been cherry-picked into 1.7.0-beta06. Could you try the beta06 or beta07 and see if the leak is still reproducible?

Message last modified on Aug 9, 2024 10:09PM

eb...@gmail.com <eb...@gmail.com> #15Aug 9, 2024 10:22PM

This is my last attempt https://github.com/persian-calendar/persian-calendar/actions/runs/10300145378/job/28508939416 and no sign of that "java.lang.OutOfMemoryError" on log but my UI tests hang forever only when a certain shared element transition is enabled https://github.com/persian-calendar/persian-calendar/commit/693938b42c78e2c60d5658b51275a5ac9ba1bca0 and here is my dependencies https://github.com/persian-calendar/persian-calendar/blob/main/PersianCalendar/runtime-dependencies-report.txt

Tbh since I can't reproduce this locally asking you for a fix is just too much so feel free to close just right away specially since a part of this is resolved already apparently. I just think something else is also going on that sooner or later someone will provide a better test case than me and of course 'leak' part is gone the far I understand, so thanks 😊

Message last modified on Aug 9, 2024 10:41PM

ti...@google.com <ti...@google.com> #16Aug 9, 2024 10:48PM

Marked as fixed.

Thanks for verifying.

If you are able to repro the ANR(?) locally, please feel free to file a separate issue. I'll close this issue as fixed.

eb...@gmail.com <eb...@gmail.com> #17Sep 5, 2024 11:41AM

That weird CI bug also gone with upgrading BoM, fantastic, thanks.