CAP Theorem

How it works

1. Detect partition
2. Pick CP or AP per operation
3. Apply tactics (quorum/async)
4. Reconcile or fail safely

🎯 What is the CAP Theorem?

The CAP theorem states that in the presence of a network partition, a distributed system can provide only two of the following three guarantees:

⚖️ Trade-offs

Depending on the system design and requirements, different trade-offs can be made:

CP Systems (Consistency + Partition Tolerance)

• Sacrifice availability during network partitions
• Examples: MongoDB, Redis, HBase
• Use case: Financial systems, inventory management

AP Systems (Availability + Partition Tolerance)

• Sacrifice consistency during network partitions
• Examples: Cassandra, DynamoDB, CouchDB
• Use case: Social media feeds, content delivery

CA Systems (Consistency + Availability)

• Only possible without network partitions
• Examples: Traditional RDBMS in single location
• Not realistic for distributed systems

Real-world Implications

• Most systems accept Partition Tolerance (P) as networks fail.
• During a partition, choose Consistency (CP) or Availability (AP) per use case.
• Many stores expose tunable consistency to dial per operation.

Practical Guidance

• CP (Consistency + Partition tolerance)
• Choose for: payments, inventory counts, auth, uniqueness constraints.
• Tactics: leader/follower, quorum writes, synchronous replication.
• Cost: higher write latency, possible unavailability when partitioned.

• AP (Availability + Partition tolerance)
• Choose for: social feeds, analytics, logs, recommendations.
• Tactics: async replication, conflict resolution, CRDTs, last-write-wins.
• Cost: clients may read stale data; require reconciliation.

📐 Quick Diagram

      Network Partition
      A ───X─── B
      
      CP: Pause one side or error on conflicting ops.
      AP: Accept writes on both sides; reconcile later.
      

❓ Interview Q&A (concise)

• Q: Does CAP apply outside partitions? A: CAP concerns partitions; PACELC extends trade-offs to normal operation.
• Q: Can a system be CA at scale? A: Only if you assume no partitions (single node/zone), which is unrealistic.
• Q: How to reconcile in AP systems? A: Version vectors, CRDTs, idempotent merges, business-specific rules.