Incremental updates to user transaction state at read-only nodes of a distributed database

• US 9,842,031 B1
• Filed: 12/08/2014
• Issued: 12/12/2017
• Est. Priority Date: 12/08/2014
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Kharatishvili, Tengiz
• Primary Examiner(s): Pham, Hung Q
• Application Number:
  US14/563,756
• Time in Patent Office: 1,100 Days
• CPC Class Codes: G06F 11/1474 in transactions G06F16/20 takes precedence
• CPC Class Codes: G06F 16/27 Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor

Abstract

A distributed database system may implement incremental updates to user transaction state at read-only nodes of the distributed database. A read-write node of the distributed database may determine a change to transaction state for a user transaction initiated at the read-write node. The read-write node may send a notification of the change to the transaction state for the user transaction to a read-only node in order to incrementally update the transaction state for the user transaction at the read-only node. The read-only node may receive the notification of the change to the transaction state and update in-memory system metadata to apply the change to the transaction state for the user transaction. A view of the distributed database may be made available at the read-only node for read requests according to the updated transaction state for the user transaction.

Claims

1. A system, comprising:

a plurality of database nodes that implement a distributed database;

a common data store that persistently maintains data for the distributed database that is accessible to the plurality of database nodes;

a read-only node of the plurality of database nodes, configured to;

maintain in-memory system metadata that describes transaction state for a user transaction initiated at a read-write node of the plurality of database nodes;

receive an indication of a change to the transaction state for the user transaction from the read-write node, wherein the initiated user transaction performs one or more updates to the data persistently maintained for the distributed database at the common data store;

in response to the receipt of the indication of the change to the transaction state for the user transaction;

update in-memory system metadata to apply the change to the transaction state for the user transaction in the in-memory system metadata; and

make a view of the distributed database available for read requests according to the transaction state for the user transaction initiated at the read-write node.

2. The system of claim 1, further comprising:

the read-write node, configured to;

determine the change to the transaction state for the user transaction initiated at the read-write node; and

in response to the determination of the change to the transaction state for the user transaction, send a notification of the change to the transaction state for the user transaction to the at least one read-only node of the plurality of database nodes.

3. The system of claim 1, wherein the change to the transaction state for the user transaction changes the transaction state to committed, and wherein the view of the distributed database available for read requests includes the one or more updates to the data of the user transaction.

4. The system of claim 1, wherein the distributed database and the common data store are respective network-based services implemented as part of a network-based service platform.

5. A method, comprising:

performing, by one or more computing devices implementing at least one read-only node of a plurality of database nodes, wherein the plurality of database nodes implement a distributed database;

receiving an indication of a change to transaction state for a user transaction initiated at a read-write node of the plurality of database nodes from the read-write node, wherein the initiated user transaction performs one or more updates to data persistently maintained for the distributed database at a common data store that is accessible to the read-write node and the at least one read-only node;

in response to receiving the indication of the change to the transaction state for the user transaction;

updating in-memory system metadata to apply the change to the transaction state for the user transaction in the in-memory system metadata; and

making a view of the distributed database available for read requests according to the transaction state for the user transaction initiated at the read-write node.

6. The method of claim 5, wherein the change to the transaction state for the user transaction changes the transaction state to committed, and wherein the view of the distributed database available for read requests includes the one or more updates to the data of the user transaction.

7. The method of claim 5, wherein the change to the transaction state for the user transaction changes the transaction state to rollback, and wherein the view of the distributed database available for read requests does not include the one or more updates to the data of the user transaction.

8. The method of claim 5, further comprising:

initializing the in-memory system metadata at the read-only node from a view of transaction history maintained for the distributed database at the common data store; and

wherein the receiving the indication of the change to the transaction state and the updating of the in-memory system metadata are performed during the initializing of the in-memory system metadata, wherein the change to the transaction state indicates a new user transaction initiated subsequent to the view of the transaction history from which the read-only node is initializing.

9. The method of claim 5, further comprising:

performing, by another one or more computing devices implementing the read-write node of the plurality of database nodes;

determining, at the read-write node, the change to the transaction state for the user transaction initiated at the read-write node; and

in response to determining the change to transaction state for the user transaction, sending a notification of the change to the transaction state for the user transaction to the at least one read-only node of the plurality of database nodes.

10. The method of claim 9, further comprising:

performing, by the other one or more computing devices implementing the read-write node of the plurality of database nodes;

sending the change to the transaction state for the user transaction to the common data store to be maintained as part of a transaction history for the distributed database.

11. The method of claim 10,

wherein the determining the change to the transaction state for the user transaction and the sending the change to the common data store are performed for another user transaction initiated at the read-write node;

wherein the method further comprises performing, by the one or more computing devices implementing the read-only node;

obtaining the change to the transaction state for the other user transaction from the transaction history maintained at the common data store;

updating the in-memory system metadata to apply the change to the transaction state for the other user transaction in the in-memory system metadata.

12. The method of claim 11,

wherein the common data store is a log-structured data store such that the change to the transaction state for the other user transaction is stored according to a sequence for the transaction history;

and

wherein obtaining the change to the transaction state for the other user transaction from the transaction history comprises evaluating the transaction history according to the sequence for the transaction history in order to identify the change to the transaction state for the user transaction.

13. The method of claim 5, wherein the distributed database and the common data store are respective network-based services implemented as part of a network-based service platform.

14. A non-transitory, computer-readable storage medium, storing program instructions that when executed by one or more computing devices cause the one or more computing devices to implement:

determining, at a read-write node, a change to transaction state for a user transaction initiated at the read-write node, wherein the read-write node is one of a plurality of database nodes, wherein the plurality of database nodes implement a distributed database, wherein the initiated user transaction performs one or more updates to data persistently maintained for the distributed database at a common data store that is accessible to the plurality of database nodes;

in response to determining the change to transaction state for the user transaction, sending a notification of the change to the transaction state for the user transaction to at least one read-only node of the plurality of database nodes;

receiving, at the at least one read-only node, the notification of the change to the transaction state for the user transaction initiated at the read-write node; and

in response to receiving the indication of the change to the transaction state for the transaction, updating in-memory system metadata to apply the change to the transaction state for the user transaction in the in-memory system metadata.

15. The non-transitory, computer-readable storage medium of claim 14, wherein the change to the transaction state for the user transaction changes the transaction state to committed, and wherein the view of the distributed database available for read requests includes the one or more updates to the data of the user transaction.

16. The non-transitory, computer-readable storage medium of claim 14, wherein the program instructions cause the one or more computing devices to further implement:

sending, from the read-write node, the change to the transaction state for the user transaction to the common data store to be maintained as part of a transaction history for the distributed database.

17. The non-transitory, computer-readable storage medium of claim 16, wherein the change to the transaction state is stored according to a layout as part of the transaction history in the common data store that is different from a layout for in-memory system metadata at the read-only node.

18. The non-transitory, computer-readable storage medium of claim 16,

wherein the determining the change to the transaction state for the user transaction and the sending the change to the common data store are performed for another user transaction initiated at the read-write node, wherein the change to the transaction state for the other user transaction indicates that the other user transaction is a new user transaction;

wherein the program instructions further cause the one or more computing devices to implement;

obtaining the change to the transaction state for the other user transaction from the transaction history maintained at the common data store;

updating the in-memory system metadata to include the new user transaction indicated in the change to the other user transaction in the in-memory system metadata.

19. The non-transitory, computer-readable storage medium of claim 14, wherein the program instructions further cause the one or more computing devices to implement:

in response to receiving the indication of the change to the transaction state for the transaction, making a view of the distributed database available for read requests received at the read-only node according to the updated transaction state for the user transaction initiated at the read-write node.

20. The non-transitory, computer-readable storage medium of claim 14, wherein the distributed database and the common data store are respective network-based services implemented as part of a network-based service platform.