Consistency Models in Distributed Systems

Consistency Models: Practical Examples

1. Linearizability (Strong Consistency)

Banking Example: You withdraw $200 from an ATM. Your spouse checks the balance at another ATM immediately after and sees the updated amount.

• When needed: Financial transactions, inventory systems
• Systems: Google Spanner, traditional RDBMS with synchronous replication

2. Sequential Consistency

Document Editing Example: Alice adds paragraph A, Bob adds paragraph B. All users see both paragraphs in the same order (either A→B or B→A), but not necessarily matching real-time order.

• When needed: When operation order matters but real-time guarantees don’t
• Systems: Many distributed databases with strong consistency settings

3. Causal Consistency

Social Media Example: Post → Comment → Reply always appear in that order to all users, but unrelated posts may appear in different orders.

• When needed: Conversation threads, event dependencies
• Systems: ZooKeeper, COPS (Clusters of Order-Preserving Servers)

4. Session Guarantees

• Read Your Writes: Update your profile picture and immediately see the change yourself
• Monotonic Reads: Once you see posts 1-10, you’ll never see only posts 1-8 later
• Consistent Prefix: If you see “How are you?”, you’ll always see the preceding “Hi” message
• When needed: User-focused applications where personal consistency matters
• Systems: MongoDB, Redis, most modern databases

5. Eventual Consistency

DNS Example: Change your website’s IP address. Some users get the new address immediately, others temporarily get the old one, but eventually all see the new address.

• When needed: High availability requirements, geographically distributed systems
• Systems: Amazon DynamoDB, Cassandra, Riak

6. Conflict Resolution Approaches

Calendar Example: You and a colleague both schedule different meetings at 2 PM while offline. When reconnecting, the system must resolve this conflict.

• Last-write-wins: Most recent change prevails (based on timestamps)
• CRDTs: Mathematical structures that automatically merge (used in collaborative editors)
• Application-specific: Custom logic (e.g., meeting with more attendees wins)
• Systems: CouchDB, Figma, Google Docs

The right consistency model depends on your application’s requirements around correctness, availability, and performance. Stronger consistency typically means higher latency and reduced availability during network partitions.