Weak Consistency

What is Weak Consistency?

Weak Consistency is a consistency model where there's no guarantee that subsequent reads will reflect a previous write. The system makes a "best effort" to propagate updates, but data may be lost or out of order.

Characteristics

How It Works

1. Client writes data to a node
2. The system acknowledges the write immediately
3. Data may or may not propagate to other nodes
4. Subsequent reads may return old, new, or no data

Client                  Server A              Server B
  │                        │                     │
  │──── Write X=1 ────────►│                     │
  │◄─── ACK ───────────────│                     │
  │                        │─ ─ ─(may or may not)─►│
  │                        │                     │
  │──── Read X ───────────►│                     │
  │◄─── X=1 ───────────────│                     │
  │                        │                     │
  │──── Read X ────────────────────────────────►│
  │◄─── X=? (old value, null, or 1) ────────────│

When to Use Weak Consistency

Ideal Use Cases

1. Real-time Gaming

   • Player positions updated frequently
   • Occasional glitches acceptable
   • Low latency critical

2. Live Streaming

   • Video/audio data flows continuously
   • Dropped frames acceptable
   • Real-time delivery prioritized

3. VoIP (Voice over IP)

   • Audio packets sent continuously
   • Some packet loss acceptable
   • Low latency essential for conversation

4. Telemetry/Metrics

   • High volume of data points
   • Individual data loss acceptable
   • Overall trends matter more

Example: Online Game

Player position updates:
- 60 updates per second per player
- 1000 concurrent players
- 60,000 updates/second

With weak consistency:
- Fire and forget updates
- Some positions may be slightly off
- Game remains playable and responsive

Weak vs Other Consistency Models

Implementation Patterns

UDP-Based Systems

Many weak consistency systems use UDP:

• No connection overhead
• No retransmission
• No ordering guarantee
• Perfect for real-time data

In-Memory Caches

Caches often use weak consistency:

• Cache invalidation is best-effort
• Stale data acceptable for short periods
• Performance prioritized

Trade-offs

Advantages

• Lowest latency: No waiting for confirmation
• Highest throughput: No coordination overhead
• Simplest implementation: No consensus needed
• Best availability: Works even with partitions

Disadvantages

• Data may be lost: No durability guarantee
• No ordering: Events may arrive out of order
• Hard to debug: Non-deterministic behavior
• Limited use cases: Not suitable for critical data

Real-World Examples

Video Conferencing (Zoom, Meet)

• Audio/video frames sent with weak consistency
• Dropped frames cause minor glitches
• Unacceptable latency would break the experience

Gaming (Fortnite, PUBG)

• Player movements use weak consistency
• Prediction and interpolation smooth out issues
• Hit registration may use stronger consistency

Sensor Networks (IoT)

• Temperature readings sent frequently
• Individual reading loss acceptable
• Battery life prioritized over reliability

Interview Tips

• Understand when it's appropriate: Real-time, high-frequency, loss-tolerant data
• Compare with other models: Know the spectrum from weak to strong
• Mention protocols: UDP is commonly associated with weak consistency
• Discuss trade-offs: Performance vs reliability
• Give examples: Gaming, streaming, VoIP

Summary

Weak consistency sacrifices data guarantees for maximum performance. It's the right choice when:

• Real-time delivery matters more than reliability
• Data is frequently updated (making old data less valuable)
• The application can tolerate occasional data loss
• Latency requirements are extremely strict

For most business applications, stronger consistency models are needed, but weak consistency powers much of the real-time infrastructure we use daily.