What is 3 Phase commit?
3 Phase commit is a protocol for distributed transaction management.
Note: 3 Phase commit protocol was created to address issues of 2 phase protocol.
The main motivation for the 3PC algorithm is to address the fact that two-phase commit protocol is blocking in case of a site (site means coordinator or participant) failure.
Like in 2 Phase commit, here also we need a transaction Coordinator Service to take care of the distributed transaction. The transaction coordinator manages the commits to database stores.
What are different phases of 3 Phase commit?
Can-Commit:
- Here Coordinator just check if slaves are ready to accept transactions, but no data is locked.
- Coordinator sends a Can-Commit message to all slaves.
- Slaves reply back with a READY or NOT READY message back:
- If a slave responds with a NOT READY message or does not respond at all, then the coordinator sends a ABORT message to all the other slaves.
Pre-commit:
- This is a prepare-to-commit stage.
- Coordinator sends a prepare-commit message to all nodes, essentially asking the nodes if they are prepared to commit
- And, if they are not, the commit is aborted.
- All participants acquire locks on data rows etc, but they don’t actually commit.
Do-Commit:
- Once the coordinator receives a YES from all nodes stating that they are prepared to commit, the coordinator will send out a commit message.
- The nodes will then each commit to the specified transaction.
Can you give example, why a Node will send Not Ready message back?
This may happen when a node has a conflicting concurrent transaction or Node has some internal issue going on.
[private]
How is 3 Phase commit different from 2 Phase commit?
The 2 Phase commit protocol is blocking protocol whereas the 3 Phase commit protocol is non-blocking.
In 2PC processing is blocked when the Prepare phase is completed but the Coordinator fails after that and becomes un-reachable. There is no way to finish with the commit until the coordinator is available. Similarly, if the participant fails and there are no timeouts in place, the Coordinator will send a commit statement and will keep waiting forever for a reply back from all participants.The three-phase commit protocol is said to be non-blocking. The blocking nature of the two-phase commit protocol is removed by introducing one more phase. 3PC splits the prepare state in two phases can-commit and pre-commit.
The reason behind this is that the reachable final state for the participant in prepare state in 2PC is commit and abort (both possibilities). Splitting state to waiting and pre-commit in 3PC we got to the situation that there is only one final state for each of them – abort from waiting, commit from pre-commit. This new state should avoid need of the external input – ie. from coordinator.If a participant is in pre-commit state we know that all participants acked the initial coordinator canCommit query by ack vote and they are waiting to commit. Thus the pre-commit state could be marked as “commitable”. Because of this structure participants can collectively decide the overall result of the transaction in case the coordinator fails.
3PC as non-blocking does not mean that the participants are not blocked in processing. The non-blocking means that protocol can proceed despite the existence of failures.
The states of the 3PC are depicted in the following state diagram.
[/private]
