EFFICIENT READ REPLICAS

• US 20140324785A1
• Filed: 04/30/2013
• Published: 10/30/2014
• Est. Priority Date: 04/30/2013
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: GUPTA, ANURAG WINDLASS, MADHAVARAPU, PRADEEP JNANA, MCKELVIE, SAMUEL JAMES, LESHINSKY, YAN VALERIE
• Granted Patent: US 10,747,746 B2
• Time in Patent Office: Days
• US Class Current: 707/689
• CPC Class Codes: G06F 16/2358 Change logging, detection, and notification replication G06F16/27
• CPC Class Codes: G06F 16/2365 Ensuring data consistency and integrity
• CPC Class Codes: G06F 16/273 Asynchronous replication or reconciliation

Abstract

A database system may receive a write request that specifies a modification to be made to a particular data record stored by the database system. A log record representing the modification to be made to the particular data record may be sent to a storage service of the database system. An indication (e.g., log record or other indication) that indicates a cached version of the particular data record stored in a read replica'"'"'s cache is stale may be sent to a read replica. For a subsequent read of the particular data record received by the read replica, the read replica may request the particular data record from the storage service.

Claims

1. A method, comprising:

performing, by multiple nodes of a database service that includes a primary node, a plurality of read replicas each having a cache, and a distributed storage service;

receiving, from a client of the database service, a write request directed to a given data record in a database table, wherein the write request specifies a modification to be made to the given data record;

generating a redo log record representing the modification to be made to the given data record;

sending the redo log record to a particular server node of the distributed storage service that stores a version of the data page comprising the given data record; and

sending the redo log record to the plurality of read replicas, wherein the redo log record indicates that a cached version of the given data record stored in respective caches of the plurality of read replicas is stale;

wherein a subsequent request for the given data record received by a read replica of the plurality of read replicas includes the read replica requesting a current version of the given data record from the distributed storage service.

2. The method of claim 1, further comprising:

receiving, by the read replica from a client of the database service, a request to read the given data record; and

determining that the cached version of the given data record stored in the cache of the read replica is stale;

requesting, from the distributed storage service, the current version of the given data record; and

sending the current version of the given data record to the client.

3. The method of claim 1, further comprising:

sending, to the plurality of read replicas, an indication of which data pages are stored in a cache of the primary node; and

updating the respective caches of the plurality of read replicas with versions of the data pages stored in the cache of the primary node.

4. The method of claim 1, further comprising:

converting one of the plurality of read replicas into a new primary node to replace the primary node without a loss of data from an update not reflected in the cache of the one read replica at a time of conversion.

5. A system, comprising:

one or more computing nodes, each of which comprises at least one processor and a memory, wherein the one or more computing nodes are configured to collectively implement a database service, and where the database service comprises a primary node, a read replica having a cache, and a distributed storage service;

wherein the primary node is configured to;

receive a write request that specifies a modification to be made to a particular data record stored by the database service,

send a redo log record representing the modification to be made to the particular data record to a server node of the distributed storage service that stores a version of a data page comprising the particular data record, and

send an indication to the read replica indicating that a cached version of the particular data record in the cache of the read replica is stale.

6. The system of claim 5, wherein the indication is a cache invalidation notification.

7. The system of claim 5, wherein the indication is the redo log record, wherein the read replica is configured to apply the redo log record to the cached version of the particular data record stored in the cache of the read replica.

8. The system of claim 7, wherein the redo log record is associated with a temporal identifier, wherein the read replica is configured to:

receive, from the distributed server node, one or more log records having respective temporal identifiers representing a later point in time than an earlier temporal identifier associated with the cached version of the particular data record; and

apply the one or more log records and the redo log record to the cached version of the particular data record.

9. The system of claim 5, wherein the read replica is configured to:

receive, from a client of the database service, a request to read the data page that includes the particular data record; and

in response to receiving the indication that the cached version of the particular data record in the cache of the read replica is stale, send a request for a current version of the data page to the distributed storage service.

10. The system of claim 9, wherein the read replica is further configured to:

receive the current version of the data page from the distributed storage service;

replace the cached version of the particular data record with a current version of the particular data record from the current version of the data page; and

provide, to the client of the database service, the current version of the data page.

11. The system of claim 5, wherein the read replica is configured to:

receive, from a client of the database service, a request to read a different data record stored by the database service, wherein a cached version of the different data record has not been indicated as stale; and

provide, to the client of the database service, the cached version of the different data record.

12. The system of claim 5, wherein the read replica is configured to:

receive, from the primary node, an indication of a plurality of data pages stored in a cache of the primary node;

retrieve versions of the plurality of data pages from the distributed storage service, and

store the retrieved versions of the plurality of data pages in the cache of the read replica.

13. The system of claim 5, wherein the read replica is configured to:

in response to a failure of the primary node, convert into a new primary node without a loss of data from an update not reflected in the cache of the read replica at a time of conversion.

14. The system of claim 13, wherein the read replica is configured to:

determine, from the distributed storage service, which data in the cache of the read replica is stale; and

indicate the determined stale data as stale.

15. The system of claim 13, wherein the read replica is configured to:

receive, from the distributed storage service, one or more transactions to data that the distributed storage service received from the primary node but that were not previously received by the read replica; and

update a transaction table of the new primary node with the one or more transactions.

16. The system of claim 15, wherein the read replica is configured to:

determine that a particular transaction of the one or more transactions was related to the failure of the primary node; and

roll back a change of the particular transaction.

17. A non-transitory computer-readable storage medium storing program instructions, wherein the program instructions are computer-executable to implement a read replica of a database service, wherein the read replica is configured to:

receive a log record, from a primary node, that indicates that a cached version of particular data in a cache of the read replica is stale, wherein the log record is indicative of an update to the particular data stored by the database service;

receive, from a client of the database service, a request to read the particular data; and

send a request for a current version of the particular data to a distributed storage service of the database service.

18. The non-transitory computer-readable storage medium of claim 17, wherein the log record is associated with a log sequence number, and wherein the read replica is further configured to:

receive, from the distributed storage service, one or more log records having respective log sequence numbers indicative of later points in time than a log sequence number for the cached version of the particular data;

apply the one or more log records from the distributed storage service and the log record from the primary node to the cached version of the particular data to create a current version of the particular data; and

provide, to the client of the database service, the current version of the particular data.

19. The non-transitory computer-readable storage medium of claim 17, wherein the read replica is further configured to:

convert into a new primary node after failure of the primary node without a loss of data from an update not reflected in the cache of the read replica at a time of conversion.

20. The non-transitory computer-readable storage medium of claim 17, wherein the read replica is further configured to:

determine, from the distributed storage service, which data in the cache of the read replica is stale.

21. The non-transitory computer-readable storage medium of claim 17, wherein the read replica is further configured to:

update a transaction table of the read replica with one or more transactions received from the distributed storage service, wherein the one or more transactions were not previously received by the read replica.