Decentralized two-phase commit
A decentralized two-phase commit in a datastore or database includes determining pre-validating a transaction, writing first metadata to a leader partition, the first metadata including a state indicating preparing and mutations of the transaction, and writing second metadata of a follower partition to indicate the follower partition is locked by the leader partition. The decentralized two-phase commit further includes updating the state of the first metadata to indicate committed and performing the transaction on the leader partition, performing the transaction on the follower partition and unlocking the follower partition, and updating the first metadata to remove the mutations of the transaction and updating the state of the first metadata to indicate the transaction is completed. The decentralized two-phase commit may further include selecting a leader partition key and updating the state of the first metadata to indicate committed unapplied.
1. A method performed by a computing device, the method comprising:
pre-validating a transaction;
writing first metadata to a leader partition, the first metadata comprising a state indicating preparing and mutations of the transaction;
writing second metadata of a follower partition to indicate the follower partition is locked by the leader partition;
updating the state of the first metadata to indicate committed and performing the transaction on the leader partition;
performing the transaction on the follower partition and unlocking the follower partition; and
updating the first metadata to remove the mutations of the transaction and updating the state of the first metadata to indicate the transaction is completed, wherein the leader partition and the follower partition are part of a datastore or database, wherein the leader partition and the follower partition are different portions of a datastore or database.
2. The method of claim 1 further comprising:
before writing the first metadata to the leader partition, selecting a leader partition key.
3. The method of claim 1 further comprising:
after the performing the transaction on the follower partition, updating the state of the first metadata to indicate committed unapplied.
4. The method of claim 1 wherein the transaction is a two-phase transaction.
5. The method of claim 1 wherein the first metadata is stored in a row associated with a key of the leader partition and the second metadata is stored in a row associated with a key of the follower partition.
6. The method of claim 1 wherein the transaction comprises a two-phase transaction wherein a data field of the leader partition is changed and a data field of the follower partition is changed.
7. The method of claim 1 wherein pre-validating the transaction comprises:
determining that the leader partition and the follower partition are not locked.
8. The method of claim 1 wherein the updating the state of the first metadata to indicate committed and the performing the transaction on the leader partition are performed atomically and wherein the performing the transaction on the follower partition and unlocking the follower partition is performed atomically.
9. The method of claim 1 the updating the state of the first metadata to indicate committed and performing the transaction on the leader partition further comprises: updating a value of a version of the leader partition and wherein the performing the transaction on the follower partition and unlocking the follower partition further comprises: updating a value of a version of the follower partition.
10. The method of claim 1 wherein the pre-validating the transaction further comprises:
storing a version of the follower partition; and
storing a version of the leader partition.
11. The method of claim 10 wherein the writing the first metadata to the leader partition, the first metadata comprising the state indicating preparing and the mutations of the transaction further comprises:
accessing a version of the follower partition and aborting the method if the version is nota same version as the stored version of the follower partition.
12. The method of claim 10 wherein the writing second metadata of the follower partition to indicate the follower partition is locked by the leader partition further comprises:
accessing a version of the leader partition and aborting the method if the version is not a same version as the stored version of the leader partition.
13. The method of claim 1 wherein the performing the transaction on the leader partition comprises:
setting a field in the leader partition to a first new value, and wherein the performing the transaction on the follower partition further comprises:
setting a field in the follower partition to a second new value.
14. The method of claim 1 wherein the follower partition is a first follower partition and where the method further comprises:
writing third metadata of a second follower partition to indicate the second follower partition is locked by the leader partition; and
performing the transaction on the second follower partition and unlocking the second follower partition.
15. The method of claim 1 further comprising:
accessing metadata of a partition;
determining the metadata of the partition indicates that the partition is locked with a state of committed;
performing a second transaction indicated in the metadata of the partition; and
unlocking the partition.
16. A first computing device comprising:
a processor; and
a memory storing instructions that, when executed by the processor, configure the first computing device to perform operations comprising:
pre-validating a transaction;
writing first metadata to a leader partition, the first metadata comprising a state indicating preparing and mutations of the transaction;
writing second metadata of a follower partition to indicate the follower partition is locked by the leader partition;
updating the state of the first metadata to indicate committed and performing the transaction on the leader partition;
performing the transaction on the follower partition and unlocking the follower partition; and
updating the first metadata to remove the mutations of the transaction and updating the state of the first metadata to indicate transaction completed, wherein the leader partition and the follower partition are different portions of a datastore or database.
17. The first computing device of claim 16 wherein the operations further comprise:
before writing the first metadata to the leader partition, selecting a leader partition key; and
after the performing the transaction on the follower partition, updating the state of the first metadata to indicate committed unapplied, and wherein the transaction is a two-phase transaction.
18. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that, when executed by a first computing device, cause the computer to perform operations comprising:
pre-validating a transaction;
writing first metadata to a leader partition, the first metadata comprising a state indicating preparing and mutations of the transaction;
writing second metadata of a follower partition to indicate the follower partition is locked by the leader partition;
updating the state of the first metadata to indicate committed and performing the transaction on the leader partition;
performing the transaction on the follower partition and unlocking the follower partition; and
updating the first metadata to remove the mutations of the transaction and updating the state of the first metadata to indicate transaction completed, wherein the leader partition and the follower partition are different portions of a datastore or database.
19. The non-transitory computer-readable storage medium of claim 18 wherein the operations further comprise:
before writing the first metadata to the leader partition, selecting a leader partition key; and
after the performing the transaction on the follower partition, updating the state of the first metadata to indicate committed unapplied, and wherein the transaction is a two-phase transaction.