Skip to content

FLP Impossibility

No deterministic consensus algorithm can guarantee agreement in an asynchronous system with even one process that might fail - a fundamental limit proving that consensus protocols must rely on timeouts or randomization.

Tension Addressed

Safety vs. Liveness

Recognizable As

Consensus protocols using timeouts; randomized algorithms; leader election mechanisms; "suspected failure" vs. "confirmed failure" distinctions

Canonical Sources

Fischer, Lynch & Paterson, "Impossibility of Distributed Consensus with One Faulty Process" (1985)

Transfers To

Consensus protocol design, blockchain systems, distributed databases, cluster management

The FLP result doesn't mean consensus is impossible in practice - it means any practical solution must sacrifice something (usually by using timeouts that might wrongly suspect healthy processes). This explains why all production consensus systems have configuration for timeouts and failure detection.