Each test is 5 questions with varying difficulty.
AI Prep covers AI Agents, Generative AI, ML Fundamentals, NLP & LLMs and a lot more, with adaptive tests and daily challenges. Fully offline on Android. Free to try, one-time unlock for lifetime access.
Distributed consensus is the foundational mechanism that enables multiple nodes in a distributed system to agree on a single data value or state transition, even in the presence of failures. In 2026, as systems scale to thousands of nodes and require strict consistency for financial transactions, metadata management, and coordination services, mastery of consensus protocols like Raft and Paxos is essential for senior backend and infrastructure engineers. Interviewers use this topic to assess a candidate's ability to reason about partial failures, network partitions, and the trade-offs between availability and consistency. Junior engineers are expected to understand the basic flow of leader election and log replication. Senior engineers must demonstrate deep knowledge of safety properties, liveness guarantees, performance bottlenecks, and the ability to implement or debug complex state machine replication scenarios.
Distributed consensus is the 'source of truth' engine for modern cloud infrastructure. Without it, distributed databases like CockroachDB or coordination services like etcd would be unable to guarantee linearizability, leading to data corruption or split-brain scenarios. In production, these protocols handle the complexity of node crashes, packet loss, and network delays, ensuring that the system remains consistent. For an interviewer, this topic is a high-signal indicator of a candidate's depth. A strong answer moves beyond definitions to discuss the 'why'—such as why Raft's strong leader model simplifies implementation compared to Paxos, or how Paxos handles multi-leader scenarios. In 2026, as we move toward more complex multi-region deployments, understanding how consensus latency scales with geographic distance is a critical skill. Candidates who can articulate the impact of RTT (Round Trip Time) on commit latency demonstrate the practical engineering maturity required for high-scale systems.
The consensus execution model centers on a Replicated State Machine (RSM) where every node processes the same sequence of inputs. The leader receives client requests, appends them to its local log, and broadcasts them to followers. Once a majority of followers acknowledge the log entry, the leader commits it and applies it to its state machine.
[Client Request]
↓
[Leader: Append Log]
↓ ↓
[Follower A] [Follower B]
↓ ↓
[Majority Acknowledged]
↓
[Leader: Commit Entry]
↓
[Apply to State Machine]Periodically save the full state machine to disk and discard the log entries before that point.
Trade-offs: Reduces recovery time and log size, but adds complexity to the state machine serialization.
A candidate node checks if it can win an election before incrementing its term, preventing disruptive re-elections.
Trade-offs: Increases election latency slightly but prevents unnecessary term increments during network instability.
The leader maintains a lease to serve reads without a full round-trip to followers, ensuring it is still the leader.
Trade-offs: Improves read throughput significantly but requires synchronized clocks or heartbeat-based timeouts.
| Reliability | Consensus protocols handle node failures by requiring a majority quorum. If a leader fails, a new election is triggered. |
| Scalability | Consensus is limited by the number of nodes in the quorum. Scaling is usually achieved by partitioning (sharding) the data. |
| Performance | Latency is dominated by the round-trip time between the leader and a majority of followers. |
| Cost | High cost due to the requirement for persistent storage and low-latency network connectivity. |
| Security | Requires mutual TLS (mTLS) between nodes to prevent unauthorized nodes from joining the cluster. |
| Monitoring | Key metrics include election count, log replication latency, and heartbeat failure frequency. |
Raft is designed for understandability and uses a strong leader-based model, making it easier to implement. Paxos is more flexible but notoriously difficult to implement correctly due to its multi-leader nature and complex state transitions.
Yes, but it is generally discouraged. An even number of nodes (e.g., 4) requires a majority of 3, which is the same as a 5-node cluster. This provides no extra fault tolerance while increasing the overhead of communication.
Consensus protocols use terms or epochs to detect stale information. If a partition occurs, the side with the majority can continue to make progress, while the minority side will stop committing entries to ensure consistency.
A quorum ensures that at least one node in the new majority has the most recent committed data, preventing data loss and ensuring that the system can always reach a single, consistent decision.
If all nodes fail, the system loses its state unless the logs were persisted to durable storage. Upon restart, nodes must recover their state from the last snapshot and the remaining log entries.
No. Replication is the process of copying data to multiple nodes. Consensus is the protocol used to ensure that all replicas agree on the order and content of those updates, maintaining consistency.
If a leader is slow, followers may time out and trigger a new election. The node with the most up-to-date log will typically win the election, replacing the slow leader.
The term is a logical clock that helps nodes identify stale leaders. If a node receives a message from a leader with a lower term, it knows the leader is outdated and rejects the message.
Consensus is generally not suitable for high-throughput streaming because it requires a round-trip to a majority of nodes for every commit. For streaming, use message queues like Kafka instead.
Linearizable consistency ensures that all operations appear to occur instantly at some point between their invocation and response. Eventual consistency allows temporary discrepancies between nodes, which will eventually resolve.
Monitor metrics like election frequency, log replication latency, heartbeat failure counts, and the number of active nodes in the quorum. High election counts often indicate network instability.
A split-brain occurs when a network partition causes a cluster to divide into two or more groups, each believing it is the leader. This can lead to conflicting writes and data corruption.
AI Prep covers AI Agents, Generative AI, ML Fundamentals, NLP & LLMs and a lot more, with adaptive tests and daily challenges. Fully offline on Android. Free to try, one-time unlock for lifetime access.