Course contentsShow
← System Design
Lesson 600 of 1,91917. Consensus AlgorithmsFree lesson

Why Consensus Is Hard

Learn the fundamental challenges: network partitions, node failures, message delays, and Byzantine faults.

Why Consensus Is Hard

What you'll learn: You'll discover the fundamental obstacles that make getting multiple nodes to agree on a value extremely challenging in distributed systems.

The Simple Question, The Complex Reality

Consensus sounds straightforward: "Can these nodes agree on a single value?" But in distributed systems, this becomes one of the hardest problems in computer science.

Imagine trying to organize a group decision where people can suddenly disappear, messages get lost in transit, some participants might lie, and you never know how long to wait for a response. That's distributed consensus.

The Core Challenges

Network Partitions

Networks split unpredictably. Node A might communicate with Node B but not Node C, while C can talk to B. Who has the "real" view of the system? How do you prevent split decisions?

Node Failures

Nodes crash without warning. Did a node fail, or is it just slow? If you wait for it, you sacrifice availability. If you proceed without it, you risk inconsistency.

Message Delays

Networks are asynchronous—messages arrive at unpredictable times, or not at all. You can't distinguish between a dead node and a really slow one. This is called the FLP impossibility: you cannot solve consensus in purely asynchronous systems with even one possible failure.

Byzantine Faults

Some nodes might be malicious or corrupted, sending different values to different peers. How do you reach agreement when participants actively lie?

Why This Matters

Without solving consensus, you can't safely replicate state machines, elect leaders, or coordinate distributed transactions. Every distributed system must either solve these problems or carefully work around them.

Key Takeaway: Consensus is hard because distributed systems face network partitions, node failures, unbounded message delays, and potentially malicious actors—all simultaneously, with no perfect way to distinguish failure from slowness.