What is etcd's role, and why does quorum matter?

Every object you create — pods, secrets, configmaps, deployments — is persisted in etcd via the apiserver. etcd replicates data across members using Raft, which requires a majority to agree before committing a write. With 3 members you tolerate 1 failure (2 is still a majority); with 5 you tolerate 2. An even number buys you nothing: 4 members still only tolerate 1 failure but add cost and latency, so odd counts are standard. If quorum is lost, etcd goes read-only and the cluster can't accept changes — pods keep running but you can't schedule or update anything. Practical implications: back up etcd regularly (snapshot), keep it on fast low-latency disks, and never let latency between members grow (don't stretch a single etcd cluster across distant regions).