[Bug]: $Proxy4.uniffi_ldk_node_fn_free_node — TimeoutException on Node destroy (Play 2.2.0)

[piotr-iohk]

What happened?

Google Play Console reports a production crash during LDK Node native teardown via UniFFI/JNA.

Play cluster label: Unknown Source - $Proxy4.uniffi_ldk_node_fn_free_node

Play report: https://play.google.com/console/u/1/developers/5690237613698939049/app/4973032799696771282/vitals/crashes/5d6157871f2b5f277ad8f0eeefefc2cf/details?days=60&isUserPerceived=true

Exception: java.util.concurrent.TimeoutException while calling uniffi_ldk_node_fn_free_node through JNA (Native.invokeVoid).

Call path: UniFFI pointer destroy / cleaner — Node$UniffiPointerDestroyer.destroy → UniffiCleanerAction.run on FinalizerDaemon (not app main thread). Suggests Node is being freed by GC/finalizer while native free_node blocks too long or deadlocks.

Play Vitals (60-day window):

• 4 events, 4 affected users.
• Versions: mostly 181 (2.2.0) (~3), 180 (2.1.2) (~1).
• Android: Android 16 Beta SDK 36 (~3), Android 15 SDK 35 (~1).
• Devices: Pixel 9 Pro, Pixel 9, Motorola Edge 50 Pro, Samsung Galaxy S21 FE.
• Visibility: 100% foreground.

Priority: Medium — lower than #980–#982; separate from those clusters.

Expected behavior

• Stopping or releasing the LDK Node must not crash the app.
• Native node teardown (free_node) must complete within JNA timeout or be coordinated with explicit LightningService.stop() / LightningRepo.stop() so finalizer does not race active node use.

Steps to Reproduce

TBD — no deterministic repro. Suggested matrix:

1. Android 15 or 16 device, mainnet wallet, notifications on (LightningNodeService / node running).
2. Scenarios that stop or replace the node: background/foreground, notification Stop, FGS onTimeout ([Bug]: ForegroundServiceDidNotStopInTimeException on LightningNodeService (Android 15+, 2.2.0) #980), rapid wallet ops, memory pressure.
3. Logcat: TimeoutException on uniffi_ldk_node_fn_free_node or FinalizerDaemon.
4. Check whether Node references can be GC'd while node still running or during async lightningRepo.stop().

Logs / Screenshots / Recordings

Play Console: https://play.google.com/console/u/1/developers/5690237613698939049/app/4973032799696771282/vitals/crashes/5d6157871f2b5f277ad8f0eeefefc2cf/details?days=60&isUserPerceived=true

Full stack trace (most recent Play report):

Exception java.util.concurrent.TimeoutException:
 at com.sun.jna.Native.invokeVoid
 at com.sun.jna.Function.invoke (Function.java:418)
 at com.sun.jna.Function.invoke (Function.java:364)
 at com.sun.jna.Library$Handler.invoke (Library.java:270)
 at java.lang.reflect.Proxy.invoke (Proxy.java:1006)
 at $Proxy4.uniffi_ldk_node_fn_free_node (Unknown Source)
 at org.lightningdevkit.ldknode.Node$UniffiPointerDestroyer$destroy$lambda$1$$inlined$uniffiRustCall$1.invoke (ldk_node.android.kt:14367)
 at org.lightningdevkit.ldknode.Node$UniffiPointerDestroyer$destroy$lambda$1$$inlined$uniffiRustCall$1.invoke (ldk_node.android.kt:313)
 at org.lightningdevkit.ldknode.UniffiRustCallStatusHelper.withReference (ldk_node.android.kt:413)
 at org.lightningdevkit.ldknode.Node$UniffiPointerDestroyer.destroy (ldk_node.android.kt:14368)
 at org.lightningdevkit.ldknode.UniffiCleanerAction.run (ldk_node.android.kt:4595)
 at jdk.internal.ref.CleanerImpl$PhantomCleanableRef.performCleanup (CleanerImpl.java:212)
 at jdk.internal.ref.PhantomCleanable.clean (PhantomCleanable.java:133)
 at java.lang.Daemons$FinalizerDaemon.doClean (Daemons.java:407)
 at java.lang.Daemons$FinalizerDaemon.processReference (Daemons.java:377)
 at java.lang.Daemons$FinalizerDaemon.runInternal (Daemons.java:341)
 at java.lang.Daemons$Daemon.run (Daemons.java:135)
 at java.lang.Thread.run (Thread.java:1564)

Bitkit Version

2.2.0 (181) mostly, some 2.1.2 (180), mainnet to.bitkit. Native dep at v2.2.0 tag: ldk-node-android 0.7.0-rc.36.

Device / OS

Android 15 (SDK 35), Android 16 Beta (SDK 36). Pixel 9 / 9 Pro, Motorola lamu, Samsung r0s.

Reproducibility

Rarely (once or twice)

Additional context

• FinalizerDaemon → likely GC/finalizer-driven Node destruction.
• May interact with async lightningRepo.stop() in LightningNodeService ([Bug]: ForegroundServiceDidNotStopInTimeException on LightningNodeService (Android 15+, 2.2.0) #980, [Bug]: ForegroundServiceStartNotAllowedException in LightningNodeService.onCreate (dataSync quota, Android 15+) #981).
• Related ldk-node native: crash: Production SIGABRT in arm64 native code #982 (SIGABRT) — cross-link, different symptom.
• Suggested: explicit Node lifecycle; await LightningService.stop(); investigate free_node in ldk-node bindings.
• Optimise Node Foreground Service #443

[piotr-iohk]

Closing for now based on Play Vitals monitoring: this cluster has not appeared in the last two weekly production checks for 2.3.0 / build 182.

This was low-frequency and tied mostly to older 2.2.0 / 2.1.2 reports. We can reopen if the free_node timeout / node teardown path appears again in current production releases.

[ovitrif]

The FinalizerDaemon timeout remains consistent with the ldk-node teardown family now tracked in synonymdev/ldk-node#94, but this report does not contain the native program counters needed to prove the same Tokio runtime self-drop mechanism.

The affected rc.36 library also lacks a usable GNU build ID and matching published debug-symbol artifact, so retrospective native symbolication is unavailable. Keeping this issue closed after no recurrence is appropriate. A recurrence should reopen it and capture a build-ID-matched native tombstone plus all-thread stacks so the teardown path can be attributed directly.