Kubernetes stores cluster state in etcd, requiring a strict majority vote (quorum) to elect leaders and commit updates. During network partitions, isolated nodes might mistakenly declare themselves leaders, causing a catastrophic "split-brain" scenario where multiple nodes accept conflicting data writes simultaneously. Quorum forces minority nodes to safely step down, preventing data drift.
This consensus logic makes even-numbered clusters inefficient. A 3-node cluster needs 2 nodes for a majority, surviving 1 failure. A 4-node cluster requires a majority of 3 nodes, meaning it still only tolerates 1 failure. Adding even nodes increases hardware costs and overhead without adding fault tolerance, making odd numbers the golden standard.
#Kubernetes #DevOps #SystemDesign #etcd #CloudNative #SRE #Infrastructure"