Skip to main content

Lifecycle-aware concurrency, testing, and interviews

Android ownership​

  • viewModelScope: screen/business work that should survive configuration change but stop when ViewModel clears.
  • lifecycleScope: work owned by a Lifecycle.
  • repeatOnLifecycle: start collection when active and cancel it when below the target state.
  • WorkManager: durable deferrable work that must survive process death/reboot constraints.
viewLifecycleOwner.lifecycleScope.launch {
    viewLifecycleOwner.repeatOnLifecycle(Lifecycle.State.STARTED) {
        viewModel.uiState.collect(::render)
    }
}

Android-only: requires AndroidX lifecycle and coroutines.

Testing​

Use runTest, test dispatchers, virtual time, fakes, and libraries such as Turbine where appropriate. Assert emissions, cancellation, errors, and absence of unwanted values—not merely the happy path.

Corner cases​

  • Rapid repeated clicks may start duplicate jobs; disable, debounce, serialize, or make operations idempotent.
  • finally runs during cancellation, but suspending cleanup needs careful NonCancellable use.
  • StateFlow may suppress equal states; immutable equality must reflect visible changes.
  • Buffer overflow strategy can drop events.
  • A caught exception may accidentally convert cancellation into an error state.
  • Process death is not coroutine cancellation; in-memory state disappears entirely.

Interview questions​

Does suspend mean background thread?

No. It means the function may suspend. Execution context comes from the coroutine context/dispatcher.

StateFlow versus SharedFlow?

StateFlow always represents a current value and replays it. SharedFlow broadcasts values with configurable replay and buffering and need not have a current value.

Why structured concurrency?

It makes lifetime, cancellation, failure, and completion follow a parent-child structure instead of leaving detached work.

Exercises​

  1. Convert sequential independent calls to async inside coroutineScope.
  2. Demonstrate cancellation of a CPU loop with isActive.
  3. Design search using debounce and flatMapLatest.
  4. Decide delivery guarantees for a snackbar versus payment confirmation.
  5. Test a StateFlow ViewModel with virtual time.
Architect takeaway

Concurrency correctness comes from ownership, cancellation, error policy, dispatcher choice, and delivery guarantees. Coroutine syntax alone does not provide those decisions.