[lucas-zimerman]

📜 Description

Scheduler.stop() only awaited the delay timer between scheduled runs, not the in-flight capture itself. ScheduledScreenshotRecorder.pause()/stop() could therefore return while a screenshot capture was still rendering, disposing its image, and releasing its GPU texture on the engine's message loop.

Scheduler.stop() now also awaits the currently in-flight capture (_runningCallback) before returning.

💡 Motivation and Context

Fixes #3923. A production user hit a native SIGABRT ~3s after MainActivity was destroyed, in ~TextureVK during a message-loop task flush on Android/Impeller (Vulkan backend). The crash was attributed to the session replay recorder: a screenshot capture that was still in flight when the app was backgrounded/destroyed could complete and post its GPU texture release after the Activity's Vulkan surface had already been torn down, aborting inside the driver.

pause()/stop() are invoked from native lifecycle callbacks and are awaited there, but the underlying Scheduler.stop() wasn't actually waiting for the capture to finish — only for a delay timer that's a loose proxy for it. This closes that gap so GPU resources are released while the surface is still alive.

💚 How did you test it?

• Added a regression test (test/replay/scheduler_test.dart) that reproduces the race: it starts a capture gated on an uncompleted Completer, calls stop(), and asserts stop() does not resolve until the capture completes. This test fails against the old implementation and passes with the fix.
• Ran the full test/replay/ suite plus flutter analyze — all green.
• Sanity-tested the example app (replay enabled at 100% sample rate) on an Android emulator: cycled backgrounding, a tight background→foreground race, and a hard force-stop while a replay segment was actively encoding. No crashes, ANRs, or hangs; pause() returned promptly in all cases.
• Note: the emulator's Impeller backend resolves to OpenGLES rather than Vulkan (engine-level device detection avoids Vulkan on virtual GPUs), so the exact native TextureVK path from the report can't be reproduced on an emulator — this fix was verified functionally rather than via a native crash repro. Atm I don't have an Android device with vulkan support for testing this.

📝 Checklist

• I reviewed submitted code
• I added tests to verify changes
• No new PII added or SDK only sends newly added PII if sendDefaultPii is enabled
• I updated the docs if needed
• All tests passing
• Public API changes reviewed by another Mobile SDK team member or implemented according to the develop docs spec
• No breaking changes

🔮 Next steps

None planned; this is a scoped fix to Scheduler.stop().

[github-actions]

Semver Impact of This PR

🟢 Patch (bug fixes)

📋 Changelog Preview

This is how your changes will appear in the changelog.
Entries from this PR are highlighted with a left border (blockquote style).

---

Fixes

• (flutter) Await in-flight replay capture on scheduler stop by lucas-zimerman in #3935

Dependencies

• chore(deps): update Native SDK to v0.16.6 by github-actions in #4013

Internal Changes

• (symbols) Collect Apple framework symbols by sentry-junior in #4003

---

_{🤖 This preview updates automatically when you update the PR.}

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 92.69%. Comparing base (f0bcefc) to head (7d0e7ed).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...ges/flutter/lib/src/replay/scheduled_recorder.dart	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3935      +/-   ##
==========================================
+ Coverage   88.06%   92.69%   +4.63%     
==========================================
  Files         347      114     -233     
  Lines       12933     4424    -8509     
==========================================
- Hits        11389     4101    -7288     
+ Misses       1544      323    -1221     

Flag	Coverage Δ
sentry	?
sentry_dio	?
sentry_drift	?
sentry_file	?
sentry_firebase_remote_config	100.00% <ø> (ø)
sentry_flutter	92.42% <93.33%> (+0.28%)	⬆️
sentry_grpc	?
sentry_hive	?
sentry_isar	?
sentry_link	?
sentry_logging	?
sentry_sqflite	?
sentry_supabase	97.27% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[github-actions]

Android Performance metrics 🚀

	Plain	With Sentry	Diff
Startup time	380.11 ms	401.91 ms	21.80 ms
Size	14.77 MiB	16.02 MiB	1.25 MiB

Baseline results on branch: main

Startup times

Revision	Plain	With Sentry	Diff
dc53d48	358.91 ms	366.17 ms	7.26 ms
d18c337	355.25 ms	342.15 ms	-13.10 ms
6e7d494	397.35 ms	378.91 ms	-18.43 ms
2d34233	470.54 ms	558.90 ms	88.36 ms
57eccdf	437.96 ms	437.94 ms	-0.02 ms
e45c0e1	447.29 ms	558.33 ms	111.04 ms
d8f332d	365.31 ms	349.98 ms	-15.33 ms
b4b7781	399.82 ms	394.26 ms	-5.56 ms
79f6b41	469.66 ms	525.90 ms	56.24 ms
7701e0e	361.66 ms	366.66 ms	5.00 ms

App size

Revision	Plain	With Sentry	Diff
dc53d48	14.09 MiB	15.28 MiB	1.19 MiB
d18c337	14.77 MiB	16.02 MiB	1.25 MiB
6e7d494	13.93 MiB	15.06 MiB	1.13 MiB
2d34233	6.54 MiB	7.55 MiB	1.01 MiB
57eccdf	14.30 MiB	15.49 MiB	1.19 MiB
e45c0e1	6.54 MiB	7.70 MiB	1.16 MiB
d8f332d	14.55 MiB	15.87 MiB	1.31 MiB
b4b7781	14.31 MiB	15.49 MiB	1.19 MiB
79f6b41	6.54 MiB	7.69 MiB	1.15 MiB
7701e0e	14.31 MiB	15.49 MiB	1.19 MiB

Previous results on branch: lz/sigart-vk

Startup times

Revision	Plain	With Sentry	Diff
1f5bd67	415.40 ms	404.56 ms	-10.84 ms
7ea65be	350.87 ms	350.58 ms	-0.29 ms
3989d55	371.38 ms	399.09 ms	27.71 ms
ec5779f	353.59 ms	358.57 ms	4.98 ms
d297b6e	440.62 ms	436.60 ms	-4.01 ms
e2b43f3	359.67 ms	357.70 ms	-1.97 ms
005190f	390.72 ms	417.40 ms	26.68 ms

App size

Revision	Plain	With Sentry	Diff
1f5bd67	14.77 MiB	16.08 MiB	1.31 MiB
7ea65be	14.77 MiB	16.08 MiB	1.31 MiB
3989d55	14.77 MiB	16.02 MiB	1.25 MiB
ec5779f	14.77 MiB	16.08 MiB	1.31 MiB
d297b6e	14.55 MiB	15.87 MiB	1.31 MiB
e2b43f3	14.77 MiB	16.08 MiB	1.31 MiB
005190f	14.77 MiB	16.02 MiB	1.25 MiB

[github-actions]

iOS Performance metrics 🚀

	Plain	With Sentry	Diff
Startup time	1249.14 ms	1245.84 ms	-3.31 ms
Size	5.91 MiB	6.35 MiB	450.94 KiB

Baseline results on branch: main

Startup times

Revision	Plain	With Sentry	Diff
2d42aac	1245.65 ms	1255.93 ms	10.29 ms
e200a70	1245.37 ms	1252.35 ms	6.98 ms
a2c8611	1253.83 ms	1246.92 ms	-6.91 ms
17bac94	1249.71 ms	1260.17 ms	10.46 ms
5b9a0da	1249.69 ms	1250.71 ms	1.03 ms
9b99523	1256.06 ms	1270.33 ms	14.27 ms
3135a81	1258.11 ms	1247.48 ms	-10.63 ms
e2d675d	1238.48 ms	1242.76 ms	4.28 ms
1837f6c	1255.00 ms	1255.08 ms	0.08 ms
6bcdc99	1257.16 ms	1264.96 ms	7.79 ms

App size

Revision	Plain	With Sentry	Diff
2d42aac	5.73 MiB	6.18 MiB	466.16 KiB
e200a70	5.53 MiB	5.96 MiB	444.86 KiB
a2c8611	5.66 MiB	6.10 MiB	452.11 KiB
17bac94	5.91 MiB	6.35 MiB	450.93 KiB
5b9a0da	5.53 MiB	5.96 MiB	444.32 KiB
9b99523	7.86 MiB	9.44 MiB	1.58 MiB
3135a81	5.73 MiB	6.17 MiB	453.76 KiB
e2d675d	7.86 MiB	9.44 MiB	1.58 MiB
1837f6c	5.91 MiB	6.35 MiB	451.30 KiB
6bcdc99	5.53 MiB	6.00 MiB	479.95 KiB

Previous results on branch: lz/sigart-vk

Startup times

Revision	Plain	With Sentry	Diff
1f5bd67	1255.26 ms	1261.20 ms	5.94 ms
ec5779f	1242.96 ms	1241.33 ms	-1.63 ms
3989d55	1260.12 ms	1261.05 ms	0.92 ms
d297b6e	1244.51 ms	1247.76 ms	3.25 ms
005190f	1236.06 ms	1249.98 ms	13.92 ms
7ea65be	1252.07 ms	1256.60 ms	4.53 ms
e2b43f3	1250.00 ms	1256.35 ms	6.35 ms

App size

Revision	Plain	With Sentry	Diff
1f5bd67	5.91 MiB	6.35 MiB	451.35 KiB
ec5779f	5.91 MiB	6.35 MiB	451.30 KiB
3989d55	5.91 MiB	6.35 MiB	450.94 KiB
d297b6e	5.83 MiB	6.28 MiB	457.79 KiB
005190f	5.91 MiB	6.35 MiB	450.93 KiB
7ea65be	5.91 MiB	6.35 MiB	451.25 KiB
e2b43f3	5.91 MiB	6.35 MiB	451.24 KiB

[buenaflor]

LGTM, but before I approve I wanna test it on my physical Android device

[buenaflor]

Sry for the delay, I'll properly test it tomorrow

[lucas-zimerman]

Sry for the delay, I'll properly test it tomorrow

No rush, I ordered an android device to help with tests and it should arrive on day 20

[cursor]

Stale Bugbot comment from a previous run.

[buenaflor]

Pls check if the AI reviews are legitimate

[lucas-zimerman]

Pls check if the AI reviews are legitimate

They were valid, fixed them on 7ea65be

[buenaflor]

I am not 100% sure to be honest if this is actually mitigating / fixing the core issue. Check my comment in the issue, I think this is a Flutter bug

[lucas-zimerman]

@buenaflor It may be unrelated to that flutter bug, but still, would you rather me to block this PR/issue until a fix on Flutter is released?

[buenaflor]

@lucas-zimerman okay I think this PR should land but I'm unsure if 2s timeout is too aggressive

[buenaflor]

Do you think it also makes sense to check if we can capture at all? e.g

from this:

_scheduler = Scheduler(
  frameDuration,
  (_) => capture(_onImageCaptured),
  _addPostFrameCallback,
);

to this:

// in scheduled_recorder.dart
_scheduler = Scheduler(
  frameDuration,
  (_) async {
    if (!_canStartCapture) {
      internalLogger.debug(() =>
          '$logName: skipping capture, app is not resumed.'));
      return;
    }
    await capture(_onImageCaptured);
  },
  _addPostFrameCallback,
);

/// The engine's surface can be torn down as soon as the app leaves
/// [AppLifecycleState.resumed]. A capture started after that point can
/// outlive the surface backing its GPU texture, and releasing the texture
/// against a dead Vulkan context aborts in the driver (#3923).
///
/// Captures already in flight when the transition happens are handled by
/// [Scheduler.stop]; this only prevents new ones.
bool get _canStartCapture {
  final state = options.bindingUtils.instance?.lifecycleState;
  // null until the first lifecycle message arrives, i.e. during startup,
  // where the app is foreground by definition.
  return state == null || state == AppLifecycleState.resumed;
}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 3989d55. Configure here.}