Two Phase Commit Protocol

Context

• Distributed transaction updates data across multiple systems or databases.
• Data needs to be atomically stored on multiple cluster nodes
• Nodes cannot make the data accessible to clients until the decision of other cluster nodes is known.
• Each node needs to know if other nodes successfully stored the data or if they failed.
• Assume that each site logs actions at that site, but there is no global log.

Problem

• Keep data accurate across all systems (consistency)
• Make sure all parts of a transaction succeed or fail together (atomicity)

Solution

• To ensure atomicity property Transaction must either commit at all the sites, or it must abort at all sites.
• The protocol has two phases: Prepare and Commit/Abort
• Phase 1 - Coordinator sends a Prepare message to all the sites where the Transaction is executed. Each site must eventually send a response
• Phase 2 - Response abort or commit received by the Coordinator from all the sites that are collaboratively executing the transaction.

Concepts

• Transaction - Prepare, Commit, Abort
• Coordinator - First asks sites to prepare their part. Then asks to commit when all agreed, asks to abort when all disagreed. Waits for all to respond.
• Sites - Responds to prepare by lock resources, prepare data changes and check if they can finish.

Benefits

• Guarantees data consistency
• Guarantees data atomicity
• Handles failures

Drawbacks

• Coordinator site failure may result in blocking
• Impossible to determine what decision has been made if the Coordinator fails and the active sites keep no additional log-record
• Can negatively impact system performance

Challenges

• Blocking Problem - the locked data items remain inaccessible for other transactions until the Coordinator is restored or fixed
• Keeping data in sync across systems
• Dealing with network problems
• Handling partial failures
• Coordinating all systems involved

Design considerations

• Consider coordinator failure after prepare message - set a timeout, then check other sites.
• Consider site failure - should others continue?
• Set timeouts for responses
• Keep good logs
• Have a manual fix plan
• Use query messages for the status of coordinator and sites
• Handle system / network failures

References

• Two Phase Commit Protocol : Distributed Transaction Management
• Two-Phase Commit
• Two-Phase Commit Protocol Explained