Single Point of Failure

What It Is

Any component whose failure takes down the whole system. Load balancers, coordinators, and service-discovery nodes are common SPOFs precisely because their job is centralizing routing decisions — the thing that makes them useful also makes them dangerous.

Mitigations

• Run multiple instances of every component — including the load balancer itself (register redundant LBs via DNS, so the LB isn't itself the SPOF it's meant to remove).
• Master-backup with fast failover for stateful components (databases) — see [[Database Replication & Sharding]].
• Multi-region deployment for physical/regional failure (a data center outage, not just a single server dying).

The CAP-Theorem Caveat

You cannot fully eliminate SPOFs and guarantee perfect strong consistency — true redundancy requires accepting some coordination cost or some consistency relaxation. This is [[CAP Theorem & Consistency Models]] showing up again, not a separate law.

Interview Drill

Q: You've made every component redundant except the load balancer. Is the system SPOF-free? A: No — the load balancer is now the SPOF. Fix by running multiple LB instances registered through DNS (or a floating IP / anycast setup), so no single LB instance is load-bearing.

Cross-links

[[Service Discovery & Health Checks]] · [[Database Replication & Sharding]] · [[CAP Theorem & Consistency Models]]