Global database transaction management service

Patent metadata

Field	Value
Publication number	US-11080271-B2
Application number	US-201615261582-A
Country	US
Kind code	B2
Filing date	Sep 9, 2016
Priority date	Sep 9, 2016
Publication date	Aug 3, 2021
Grant date	Aug 3, 2021

Abstract

A transaction management service layer monitors a plurality of read and write transactions being executed by the data engines by way of client libraries at each of a plurality of data engines. In addition, the transaction management server layer maintains and coordinates transaction timestamp and sequence numbers that are shared by the plurality of data engines. The transaction management service layer identifies and resolves conflicts among the plurality of transactions. The transaction management service layer is a separate and independent service that is isolated from other components of the plurality of data engines. Related apparatus, systems, techniques and articles are also described.

First claim

What is claimed is:

1. A method comprising:
monitoring, by a transaction management service layer executed by a transaction broker server and a lock manager server and via client libraries at each of a plurality of data engines, a plurality of read and write transactions being executed by the data engines, at least a subset of the data engines being heterogeneous;
maintaining and coordinating, by the transaction management service layer, transaction timestamp and sequence numbers that are shared by the plurality of data engines;
identifying, by the transaction management service layer, conflicts among the plurality of transactions; and
resolving, by the transaction management service layer, the identified conflicts among the plurality of transactions;
wherein:
the transaction management service layer is a separate and independent service that is isolated from other components of the plurality of data engines to provide a single transaction domain across the heterogeneous data engines;
the transaction management service layer is executed, at least in part, by a transaction broker server separate from the data engines that maintains a global commit timestamp that is incremented when write transactions commit;
each data engine comprises a different client library embedded therein that wraps access methods associated with the transaction broker server and the lock manager executing the transaction management service layer, the access methods being called by the respective data engine such that the corresponding client library processes such calls and communicates with the transaction broker server and the lock manager server;
the transaction broker server provides snapshot isolation by assigning commit identifiers to committing write transactions and snapshot timestamps to starting read transactions;
the lock manager server maintains a hierarchy of multi-granular lock tables used to detect transactional conflicts among concurrently executing transactions;
the transaction broker server provides a minimum snapshot timestamp which is used to determine which data across all of the data engines is to be subject to garbage collection.

2. The method of claim 1 , wherein the transaction broker server maintains a transaction identification generator which assigns transaction identifications to each transaction as it starts.

3. The method of claim 1 , wherein the transaction broker server provides a minimum snapshot timestamp which is used to determine which data in the data engines is to be subject to garbage collection.

4. The method of claim 1 , wherein the hierarchy of multi-granular lock tables comprise: table locks, partition locks, and record locks.

5. The method of claim 1 , wherein the conflicts are resolved by the lock manager implementing a hybrid of optimistic locking and pessimistic locking such that database transactions are committed prior to corresponding lock requests being granted and such transactions are rolled back if such lock requests are not granted.

6. The method of claim 1 , wherein the client libraries group multiple requests originating from concurrently executing transactions.

7. The method of claim 5 , wherein the client libraries group and buffer multiple requests from a same transaction.

8. The method of claim 1 , wherein the client libraries are embedded to nodes of the data engines.

9. The method of claim 8 , wherein if a first of the data engines, which embeds a first client library within a same process boundary, calls an access method provided by a second client library, the second client library processes the call and communicates with at least one of the transaction broker server and a lock server.

10. A system comprising:
a plurality of data engines each storing or otherwise making data available to remote client computing systems, at least a portion of the data engines being heterogeneous, each data engine comprising at least one programmable data processor and memory,
a transaction broker server and a lock manager server forming a transaction management service layer;
wherein the transaction management service layer:
monitors, via client libraries at each of the plurality of data engines, a plurality of read and write transactions being executed by the data engines;
maintains and coordinates transaction timestamp and sequence numbers that are shared by the plurality of data engines;
identifies conflicts among the plurality of transactions;
resolves the identified conflicts among the plurality of transactions using optimistic concurrency control, hybrid locking such that database transactions are committed prior to corresponding lock requests being granted and such transactions are rolled back if such lock requests are not granted; and
is a separate and independent service that is isolated from other components of the plurality of data engines to provide a single transaction domain across the heterogeneous data engines;

wherein the transaction broker server provides a minimum snapshot timestamp which is used to determine which data across all of the data engines is to be subject to garbage collection.

11. The system of claim 10 , wherein the transaction broker server:
provides snapshot isolation by assigning commit identifiers to committing write transactions and snapshot timestamps to starting read transactions; and
maintains a transaction identification generator which assigns transaction identifications to each transaction as it starts.

12. The system of claim 11 , wherein the a lock manager server maintains a hierarchy of multi-granular lock tables used to detect transactional conflicts among concurrently executing transactions.

13. The system of claim 12 , wherein the hierarchy of multi-granular lock tables comprise: table locks, partition locks, and record locks.

14. A non-transitory computer program product storing instructions which, when executed by at least one data processor forming part of at least one computing device, perform operations comprising:
monitoring, by a transaction management service layer executed by a transaction broker server and a lock manager server and via client libraries at each of a plurality of data engines, a plurality of read and write transactions being executed by the data engines, at least a portion of the data engines being heterogeneous;
maintaining and coordinating, by the transaction management service layer, transaction timestamp and sequence numbers that are shared by the plurality of data engines;
identifying, by the transaction management service layer, conflicts among the plurality of transactions; and
resolving, by the transaction management service layer, the identified conflicts among the plurality of transactions;
wherein:
the transaction management service layer is a separate and independent service that is isolated from other components of the plurality of data engines;
each data engine comprises a different client library embedded therein that wraps access methods associated with both of a transaction broker server and a lock manager server that together execute the transaction management service layer;
the transaction broker server maintaining a global commit timestamp which is incremented when a global write transaction commits across one or more of the data engines;
the lock manager resolves conflicts using a hybrid of optimistic locking and pessimistic locking such that database transactions are committed prior to corresponding lock requests being granted and such transactions are rolled back if such lock requests are not granted; and

the transaction broker server provides a minimum snapshot timestamp which is used

Assignees

• Sap Se

Inventors

• Lee Juchang
• Park Chang Gyoo
• Gropengiesser Francis
• Goel Anil Kumar
• Kim Deok Koo
• Lee Sunho

Classifications

• G06F16/2365

  Ensuring data consistency and integrity · CPC title

• G06F16/2343

  Locking methods, e.g. distributed locking or locking implementation details · CPC title

• G06F16/2379 Primary

  Updates performed during online database operations; commit processing · CPC title

• G06F9/466

  Transaction processing · CPC title