Skip to content

Stateflow sharedflow and channels

← Back to Concurrency

StateFlow, SharedFlow, and Channels Deep Dive

Overview

These primitives represent different messaging semantics: state, broadcast events, and point-to-point queues.

Core Concepts

  • StateFlow: state holder with current value
  • SharedFlow: multicast stream with configurable replay
  • Channel: queue/handoff semantics
  • delivery guarantees and subscriber topology

Internal Implementation

StateFlow is a specialized SharedFlow optimized for current-state access. SharedFlow maintains replay cache and subscriber coordination. Channel coordinates send/receive with capacity/backpressure semantics.

Threading Model

All three are coroutine-friendly and can be used across dispatchers, but ownership and collection lifecycle should remain explicit in architecture.

Coroutine / Flow Behavior

UI state should generally be StateFlow. One-off UI events usually fit SharedFlow with low replay. Work-queue or request-response pipelines often map better to Channel.

Code Examples

private val _state = MutableStateFlow(UiState())
val state: StateFlow<UiState> = _state
private val _events = MutableSharedFlow<UiEvent>(replay = 0)
val events: SharedFlow<UiEvent> = _events
private val workQueue = Channel<WorkItem>(capacity = Channel.BUFFERED)

Common Interview Questions

  • Q: Why are one-off events awkward with plain StateFlow? A: Start from delivery semantics: use StateFlow for durable state, SharedFlow or Channel for transient events, and lifecycle-aware collection to prevent duplicate work.
  • Q: When is a Channel safer than SharedFlow? A: Start from delivery semantics: use StateFlow for durable state, SharedFlow or Channel for transient events, and lifecycle-aware collection to prevent duplicate work.
  • Q: How does replay affect duplicate UI actions? A: Answer with correctness first and throughput second: cancellation model, dispatcher choice, bounded parallelism, and contention or latency measurements.
  • Q: How do you model state and events separately in ViewModel? A: Answer with correctness first and throughput second: cancellation model, dispatcher choice, bounded parallelism, and contention or latency measurements.

Production Considerations

  • separate state stream from event stream
  • keep replay minimal for transient events
  • document consumer assumptions for channels
  • test collector restart behavior

Performance Insights

Misconfigured replay/capacity can create memory pressure or dropped signals. Tune by event criticality and collector speed.

Senior-Level Insights

Great answers center on semantics first, API second: choose primitive by delivery contract, not popularity.