Three-Phase Commit (3PC)
What you'll learn: How Three-Phase Commit extends 2PC with an extra phase to reduce blocking, and why it still struggles with network partitions.
The Core Problem with 2PC
Recall that Two-Phase Commit can block indefinitely if the coordinator crashes after sending PREPARE but before sending COMMIT/ABORT. Participants are stuck waiting, unsure whether to commit or abort, holding locks that freeze the system.
How 3PC Works
Three-Phase Commit adds a pre-commit phase between the voting phase and the final commit:
- Phase 1 (Prepare): Coordinator asks all participants "Can you commit?" Just like 2PC.
- Phase 2 (Pre-Commit): If all vote YES, the coordinator sends PRE-COMMIT to all participants. Participants acknowledge and know that everyone voted YES—a commit is coming.
- Phase 3 (Commit): Coordinator sends the final COMMIT message.
The Key Improvement
The pre-commit phase creates a non-blocking window. If the coordinator crashes after pre-commit but before commit, participants know:
- Everyone voted YES
- A commit is inevitable
- After a timeout, they can safely elect a new coordinator and complete the commit
This eliminates the indefinite blocking problem of 2PC in crash-failure scenarios.
The Critical Weakness: Network Partitions
3PC assumes a synchronous network model with known message delays and timeouts. In reality, networks experience partitions—groups of nodes become isolated from each other.
If a network partition occurs:
- Some nodes might timeout and elect a new coordinator, deciding to commit
- Other nodes (in the other partition) might timeout differently and abort
- Result: Inconsistent state across partitions, violating safety
3PC trades 2PC's blocking problem for partition vulnerability—a dangerous tradeoff in real-world asynchronous networks.
Key Takeaway: Three-Phase Commit reduces blocking by adding a pre-commit phase that allows recovery after coordinator failure, but it sacrifices safety during network partitions, making it rarely used in practice compared to 2PC.