Failure tolerant transaction processing system

• US 8,166,007 B2
• Filed: 04/03/2008
• Issued: 04/24/2012
• Est. Priority Date: 11/30/2005
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Yamamoto, Guku, Tai, Hideki, Hori, Hiroshi
• Primary Examiner(s): Lie, Angela
• Assistant Examiner(s): MINA, FATIMA P
• Application Number:
  US12/061,702
• Publication Number:
  US 20090157757A1
• Time in Patent Office: 1,482 Days
• Field of Search: 707/684, 707/690, 707/699, 707/703, 707/704
• US Class Current: 707/703
• CPC Class Codes: G06F 11/1443 Transmit or communication errors
• CPC Class Codes: G06F 11/1474 in transactions G06F16/20 takes precedence
• CPC Class Codes: G06F 11/183 by voting, the voting not being performed by the redundant components
• CPC Class Codes: G06F 11/2023 Failover techniques
• CPC Class Codes: G06F 9/466 Transaction processing

Abstract

The present invention proposes a system for resending a process to a backup server farm from a client without waiting for the failure-detection, if no reply is received for a certain time. The transaction processing mechanism of the present invention has a transaction start processing mechanism in which an exclusive control using a valid processing authority token and data consistency are combined, and a commit processing mechanism in which determination on whether a commit is available or not based on a distributed agreement and replication of updated data. With the mechanisms, a system for shortening a service halt time when a failure occurs to a time as such it appears to a client that the service does not stop is provided.

Claims

1. A computer-implemented method of transaction processing comprising a plurality of clients and a plurality of server farms connected to a network, each of said plurality of server farms comprising a database (DB), a DB server for managing the database, and an application server for communicating with said DB server and said plurality of clients, the method comprising:

determining, by a computer, whether a transaction with a same ID is being executed or completes a commit by referencing a transaction ID according to a transaction start request from said client via said application server, and when said transaction with the same ID is being executed or completes a commit, returning rollback to the transaction start request;

transferring, by said computer, said transaction ID of an objective transaction, data updated by the transaction, and a processing result to be returned to all of said server farms, and checking, by said computer, whether said data is copied to said database in a majority of server farms;

when said checking of said data being copied is successful, sending, by said computer, a message of a commit request to all of said other server farms, and when a commit agreement from a majority of said other server farms is received, determining, by said computer, that said commit request is successful;

sending, by said computer, a result of said commit established by said determining to all of said other server farms;

checking, by said computer, whether a self server farm keeps a valid time-limited processing authority token, and when said self server farm does not keep said valid time-limited processing authority token, issuing an acquisition request of said valid time-limited processing authority token to all the other server farms, other than said self server farm, and waiting for completion of the acquisition request from a majority of said other server farms;

when said commit request is received while said valid time-limited processing authority token acquisition is being processed, not replying, by said computer, to the commit request until said valid time-limited processing authority token acquisition process is completed, or performing, by said computer, a resending request of a message of said commit request; and

said self server farm, of said computer, performing data consistency among all of said other server farms when said valid time-limited processing authority token is acquired.

2. The transaction processing method according to the claim 1, further comprising:

sending, by said computer, a message of said commit request to all of said other server farms, and when an acquisition message of said valid time-limited processing authority token is received from when a reply for the commit request is received until a commit is established, not replying, by said computer, to the acquisition message of said valid time-limited processing authority token until the commit is established, or performing, by said computer, a resending request of an acquisition message of said valid time-limited processing authority token.

3. The transaction processing method according to the claim 1, further comprising:

receiving, by said computer, unestablished transaction logs from said other server farms and bringing a state of said unestablished transaction logs into synchrony before said valid time-limited processing authority token acquisition is completed.

4. The transaction processing method according to the claim 1, further comprising:

when transaction information with an unknown commit is included in said message when the acquisition request message of said valid time-limited processing authority token is received and when the acquisition request responding message of the valid time-limited processing authority token is received, then referencing, by said computer, said transaction information, and when a commit or rollback of the transaction has been established, returning, by said computer, the result.

5. The transaction processing method according to the claim 1, further comprising:

determining, by said computer, whether those second other server farms, other than that sent a message of the commit request, kept said valid time-limited processing authority token to said commit request message, and when said second other server farms keep said valid time-limited processing authority token, returning, by said computer, a rejection to the commit request, and if they did not keep said valid processing authority token, returning, by said computer, approval.

6. The transaction processing method according to the claim 1, further comprising:

causing, by said computer, said client to set a transaction resending time which sends said transaction start request and a valid period of said valid time-limited processing authority token.

7. The transaction processing method according to the claim 1, said valid time-limited processing authority token including an expiration time.

8. A computer system for transaction processing, said computer system comprising a plurality of clients and a plurality of server farms connected to a network, each of said plurality of server farms comprising a database (DB), a DB server for managing the database, and one or more application servers for communicating with said DB server, the system comprising:

a memory, for each said database, for storing data; and

a processor, for each said DB server, configured to;

determine whether a transaction with a same ID is being executed or completes a commit by referencing a transaction ID according to a transaction start request from said client via said application server, and when the transaction with said same ID is being executed or completes a commit, returning rollback to the transaction start request;

check whether the self server farm keeps a valid time-limited processing authority token, and when said self server farm does not keep said valid time-limited processing authority token, issuing an acquisition request of said valid time-limited processing authority token to all the server farms other than said self server farm, and waiting for completion of the acquisition request from a majority of server farms;

when said commit request is received while said valid time-limited processing authority token is being processed, not replying to the commit request until said valid time-limited processing authority token acquisition process is completed, or performing a sending request of a message of said commit request;

perform data consistency, by said self server farm, among all of said server farms when said valid time-limited processing authority token is acquired, and

said computer system in a commit process of said transaction being further configured to;

transfer said transaction ID of an objective transaction, data updated by the transaction, and a processing result to be returned to all of said other server farms, and check whether said data is copied to said database in at least a majority of server farms;

when checking of said data being copied is successful, sending a message of a commit request to all of said other server farms, and when a commit agreement from a majority of server farms is received, determining that said commit request is successful; and

send a result of a commit established by commit establishing to all of said other server farms.

9. The computer system according to the claim 8, being further configured to:

cause the client to set a transaction resending time of the client which sends said transaction start request and a valid period of said valid time-limited processing authority token.

10. A non-transitory storage medium for a computer program readable by a computer system, tangibly embodying a program of instructions executable by said computer system, said computer system comprising a plurality of clients and a plurality of server farms connected to a network, said server farms comprising a database (DB), a DB server for managing the database, and one or more application servers for communicating with said DB server and said clients, to perform a method of transaction processing, said method comprising:

determining whether a transaction with a same ID is being executed or completes a commit by referencing a transaction ID according to a transaction start request from said client via said application server, and when said transaction with the same ID is being executed or completes a commit, returning rollback to the transaction start request;

transferring said transaction ID of an objective transaction, data updated by the transaction, and a processing result to be returned to all of said server farms, and checking whether said data is copied to said database in a majority of server farms;

when said checking of said data being copied is successful, sending a message of a commit request to all of said other server farms, and when a commit agreement from a majority of said other server farms is received, determining that said commit request is successful;

sending a result of said commit established by said determining to all of said other server farms.

checking whether a self server farm keeps a valid time-limited processing authority token, and when said self server farm does not keep said valid time-limited processing authority token, issuing an acquisition request of said valid time-limited processing authority token to all the other server farms, other than said self server farm, and waiting for completion of the acquisition request from a majority of said other server farms;

when said commit request is received while said valid time-limited processing authority token acquisition is being processed, not replying to the commit request until said valid time-limited processing authority token acquisition process is completed, or performing a resending request of a message of said commit request; and

said self server farm performing data consistency among all of said other server farms when said valid time-limited processing authority token is acquired.

11. The method, stored on the non-transitory storage medium, according to the claim 10, further comprising:

causing said client to set a transaction resending time of a client which sends said transaction start request and a valid period of said valid time-limited processing authority token.

12. The method, stored on the non-transitory storage medium, according to the claim 11, said valid time-limited processing authority token including an expiration time.

13. The method, stored on the non-transitory storage medium, according to the claim 10, further comprising:

sending a message of said commit request to all of said other server farms, and when an acquisition message of said valid time-limited processing authority token is received from when a reply for the commit request is received until a commit is established, not replying to the acquisition message of said valid time-limited processing authority token until the commit is established, or performing a resending request of an acquisition message of said valid time-limited processing authority token.