A network partition is a communication failure that prevents groups of distributed-system nodes from exchanging messages reliably. It makes consistency and availability trade-offs unavoidable because real systems must continue operating despite network failures. section 6.2
Interview scenario
When asked whether a system can guarantee consistency, availability, and partition tolerance simultaneously, explain that CAP permits only two guarantees at a time. Because partition tolerance is necessary in real distributed systems, the practical choice is usually between consistency and availability: CP systems such as HBase favor consistency, while AP systems such as DynamoDB favor availability and may provide eventual consistency. section 6.2
Table 6.3 Distributed computing fundamentals
|
Topic
|
Fundamentals
|
Applied skills
|
Advanced
|
|
CAP theorem
|
Understand that a distributed system can only guarantee two of: consistency, availability, and partition tolerance.
|
Classify real systems as CP, AP, or CA. e.g., DynamoDB is AP; HBase is CP. Explain trade-offs during network failures.
See appendix A.13 for details.
|
Evaluate architectural decisions under partition scenarios. Justify consistency/availability choices in critical systems.
|
|
Consistency models
|
Know the difference between strong consistency and eventual consistency. Recognize intermediate models like causal and read-your-writes.
|
Select appropriate models based on system requirements—Use strong consistency (e.g., PostgreSQL for financial applications, and eventual consistency (e.g., Cassandra or DynamoDB) for social media feeds. Implement tunable consistency settings when supported to balance performance and correctness.
|
Analyze and optimize consistency guarantees in distributed storage or multi-region deployments. Design around consistency anomalies.
|
|
Consensus
|
Understand the purpose of consensus in distributed systems (e.g., agreement despite failures). Know examples like Paxos and Raft.
|
Use consensus algorithms for leader election, quorum writes, or log replication (e.g., etcd —
https://etcd.io, ZooKeeper—
https://zookeeper.apache.org). Handle node failures and retries.
|
Evaluate and tune consensus performance under network partitions. Compare protocols (Raft vs. Paxos) in terms of safety, liveness, and complexity.
|
|
Constraints
|
Understand that distributed systems face inherent challenges such as partial failures (some nodes fail while others function), network latency, and messages arriving out of order or being lost.
|
Design applications to handle retries, timeouts, and idempotency. Implement patterns like heartbeats, acknowledgments, and sequence numbers to detect and mitigate message reordering or loss.
|
Architect fault-tolerant systems that maintain correctness under adverse conditions. Simulate failures (e.g., chaos engineering) to validate resilience. Optimize protocols for latency-sensitive workloads while tolerating network-level issues.
|
Compare CAP fundamentals, applied classification of systems as CP or AP, and architectural decisions under partition scenarios. section 6.2


