System for first requesting changes be staged on one or more storage depots and then flipping those changes to be live after the data is secured.

Requires write ahead, undo/redo logs, need to handle acknowledgements going stale (ex. network partitions or crashes.)

Stage transaction

• Coordinator sends query to commit to all affected parties
• Coordinator waits until all parties respond
• Parties perform their component (writing undo/redo and write ahead logs)
• Parties inform coordinator if writing succeeded
• If any party failed coordinator issues abort to everyone

Commit transaction

• Coordinator sends commit to all affected parties
• Parties make the transaction changes permanent
• Parties send completion notice to coordinator

Abort transaction

• Coordinator sends abort to all affected parties
• Parties dump the staged transaction
• Parties send completion notice to coordinator