2PC

Two-Phase Commit (2PC) is a distributed algorithm used to ensure atomic commitment of a transaction across multiple nodes, or participants, in a distributed system. It guarantees that either the transaction is committed on all participants or it is aborted on all, preserving atomicity and consistency in the presence of failures. It is commonly used in distributed databases and resource managers that need cross-node transactional guarantees.

In 2PC, there are two phases. Phase one is the prepare (voting) phase. The coordinator sends a prepare request to each participant, and each participant writes a durable log entry and votes Yes if it can commit the transaction given its local state, or No if it cannot. If any participant votes No or a timeout occurs, the coordinator will abort the transaction. Phase two is the commit/abort phase. If all participants voted Yes, the coordinator sends a commit command; otherwise, it sends an abort command. Participants then perform the corresponding action locally and acknowledge completion.

The protocol relies on durable logging and reliable messaging. If the coordinator fails after the prepare messages have been sent but before the final decision, participants may remain in a prepared state, causing the system to block until a new coordinator recovers and issues a decision. This blocking behavior is a key drawback of 2PC. Recovery typically requires a durable log and a newer coordinator or recovery process to determine and enact the final decision.

Variants and limitations include three-phase commit (3PC), which attempts to reduce blocking at the cost of extra complexity, and optimizations such as timeouts and presumed-commit or presumed-abort strategies. In practice, 2PC is simple and reliable for short-lived transactions but can impact scalability and latency in large, distributed deployments.