GraphQL subscriptions can get stuck in connecting after repeated iOS background/foreground transitions #4220

Describe the bug

Summary

We have a repro repo for an iOS GraphQL subscription reconnection issue after app lifecycle transitions.

With the current feature/multi-subscription-repro branch, we are now able to reproduce a stronger failure mode that is closer to our production app:

• one or more subscriptions enter connecting
• we have not observed them recover without fully terminating the app
• the app does not recover by foregrounding again or by normal in-app actions

This stuck state is not reproducing on the simpler main branch of the repro repo. It is currently reproducing only on the multi-subscription variant.

We also observe a difference between build configurations:

• the issue is harder to reproduce in Debug
• the issue reproduces more easily in Release

Repro repository

• Repo: https://github.com/akiramur/amplify-swift-ios-repros
• Branch: feature/multi-subscription-repro
• Commit: 6daca6d404771ab3cdac435976eacf0577958b40

Repro shape

This repro app uses:

• Cognito auth with Hosted UI sign-in
• AppSync GraphQL API
• one GraphQL model
• three concurrent onCreate subscription workers
• background/foreground restart logic across all workers
• in-app logging for worker connection state and lifecycle transitions

Environment

• Amplify Swift: 2.58.1
• Xcode: 26.3
• iOS: 26.4.2
• Device: iPhone 12
• Repro backend: Amplify Gen 2 sandbox

Actual behavior

After repeating lock/unlock and app switching several times, one or more subscription workers eventually enter connecting and remain stuck there.

Once this happens:

• one or more stuck subscription workers do not return to connected
• normal app usage does not recover them
• repeated foreground/background transitions do not recover them
• the app remains in a degraded state until it is fully terminated and relaunched

In other words, after the stuck state is reached, a full app restart is the only reliable recovery we have observed.

Important observation

This behavior is currently reproducing only on the feature/multi-subscription-repro branch, which runs multiple concurrent subscription workers.

We are not reproducing the same stuck connecting state on the simpler main branch of the repro repo.

This multi-subscription branch appears closer to our production shape, where multiple subscriptions and lifecycle-driven restarts are active concurrently.

This suggests the issue may be related to one or more of:

• multiple concurrent subscriptions
• lifecycle-driven restart timing across multiple workers
• reconnection races that do not appear with a single subscription

Steps To Reproduce

Steps to reproduce the behavior:

1. Start the backend sandbox and generate `amplify_outputs.json`.
2. Generate the Xcode project and run the app from the `feature/multi-subscription-repro` branch.
3. Sign in with Hosted UI.
4. Confirm all subscription workers reach `connected`.
5. Lock the device screen and leave the app idle for around 2 to 3 minutes.
6. Unlock the device.
7. Repeatedly switch between this app and other apps so the repro app cycles through background and foreground several times after unlocking.
8. Observe the aggregate connection badge, worker cards, and event log.

Expected behavior

Expected behavior

After app lock/unlock and repeated background/foreground transitions, all subscription workers should eventually reconnect and return to connected without requiring a full app restart.

Amplify Framework Version

2.58.1

Amplify Categories

API

Dependency manager

Swift PM

Swift version

5.9

CLI version

13.0.0

Xcode version

26.3 (17C529)

Relevant log output

<details>
<summary>Log Messages</summary>


INSERT LOG MESSAGES HERE

</details>

Is this a regression?

No

Regression additional context

No response

Platforms

No response

OS Version

26.4.2

Device

iPhone 12

Specific to simulators

No response

Additional context

No response