JP5515286B2 - Distributed transaction processing system, server apparatus, and distributed transaction failure recovery method used therefor - Google Patents

Description

  The present invention relates to a distributed transaction processing system, a server apparatus, a distributed transaction failure recovery method used therefor, and a program thereof, and more particularly to a distributed transaction failure recovery when a failure occurs.

  In the distributed transaction processing system related to the present invention, the transaction manager has a plurality of resource managers, and the transaction manager performs a two-layer commitment procedure in order to correctly perform the commitments of the plurality of resource managers.

  The outline of the two-layer commitment process will be described below. The two-layer commitment process includes a first layer process and a second layer process. In the transaction update process, the transaction manager issues a prepare request to each resource manager as the first layer process. Each resource manager returns to the transaction manager whether the commitment is possible.

  When the transaction manager receives replies from all resource managers, the transaction manager issues a commitment request as a second layer process when replies that can be committed are received from all resource managers. The transaction manager completes a series of transactions when receiving the response of the second layer processing from all the resource managers.

  In the first layer processing, when a response that cannot be committed is received from one or more resource managers, the transaction manager issues a rollback request as the second layer processing, and when a response is received from all resource managers, Complete the transaction.

  Transaction processing becomes incomplete when the server device on which the transaction manager operates stops during transaction execution, and the resource manager associated with the transaction performs updates from other sources until the database update processing is completed. It is locked to prevent it from being broken, so it is in an unusable state. Transaction processing is resumed when the server device on which the transaction manager operates is started.

  In order to increase the reliability of the application, a cluster including a pair of an active server device and a standby server device is configured. In this configuration, when the active server device stops, the transaction manager immediately starts up the standby server device, resumes transaction processing, and releases the resource manager lock.

Other distributed transaction failure recovery methods related to the present invention include a method of preparing a device for recovering transactions in the same node (see, for example, Patent Document 1), and determination of a transaction in progress when a failure occurs. A method of performing processing in a short time (for example, see Patent Document 2) has been proposed.
JP 2005-250998 A JP 2004-234203 A

  However, in recent years, in order to increase the reliability of applications, as shown in FIG. 1, processing is distributed to a plurality of server devices using a load balancer. In this system, a configuration is adopted in which a standby system is not arranged in the server device.

  For this reason, when a certain server device is stopped, a transaction that was operating on the server device is not started on the alternative device. Therefore, since the area being updated by the resource manager cannot be referenced by other transactions, if the transaction is processed from another server device to the resource manager before the transaction is recovered, the entire system will stop. There is.

  A problem of the distributed transaction processing system related to the present invention is that it takes time to recover a transaction when a server failure or the like occurs.

  The reason is that in the case of the above-described method of Patent Document 1, the transaction recovery means, the transaction information management table, and the transaction execution means operate on the same server, so that recovery cannot be performed. This is because if the server including the recovery means goes down, there is no transaction recovery means.

  Therefore, an object of the present invention is to solve the above-mentioned problems, and when a failure occurs, a distributed transaction processing that can minimize the lock time of the resource manager and minimize the stop time of the entire system. It is an object to provide a system, a server device, a failure recovery method for a distributed transaction used for them, and a program therefor.

A distributed transaction processing system according to the present invention includes a transaction manager and a relay device having a resource manager , and the resource manager performs a two-layer commitment process of a transaction based on an instruction from the transaction manager,
The relay device receives and transmits the instruction from the transaction manager and transmits to the resource manager, detection means for detecting an abnormal end of the transaction manager, and when the detection means detects the abnormal end Execution means for executing the second layer processing of the two-layer commitment processing of the transaction on the resource manager based on the elapsed time from the completion of the first layer processing of the two-layer commitment processing.

A server device according to the present invention includes a transaction manager and a relay device having a resource manager , and the server device is used in a distributed transaction processing system in which a resource manager performs a two-layer commitment process of a transaction based on an instruction from the transaction manager. ,
Transmission / reception means for receiving the instruction from the transaction manager and transmitting it to the resource manager, detection means for detecting abnormal termination of the transaction manager, and the two-layer commitment processing when the detection means detects the abnormal termination Execution means for executing the second layer processing of the two-layer commitment processing of the transaction on the resource manager based on the elapsed time from the completion of the first layer processing.

Error recovery method for distributed transactions according to the invention comprises a transaction manager, and a relay device having a resource manager, the resource manager on the basis of the instructions of the transaction manager uses the distributed transaction processing system that performs two layers commitment of a transaction distributed A transaction recovery method,
The relay device receives the instruction from the transaction manager and transmits it to the resource manager, a detection process for detecting abnormal termination of the transaction manager, and the abnormal termination is detected in the detection process In this case, execution processing for executing the second layer processing of the two-layer commitment processing of the transaction is executed for the resource manager based on the elapsed time from the completion of the first layer processing of the two-layer commitment processing.

A program according to the present invention includes a transaction manager and a relay device having a resource manager, and a central processing in the relay device in a distributed transaction processing system in which a resource manager performs a two-layer commitment process of a transaction based on an instruction from the transaction manager. A program executed by the device,
A transmission / reception process for receiving the instruction from the transaction manager and transmitting it to the resource manager, a detection process for detecting an abnormal end of the transaction manager, and the two layers when the abnormal end is detected in the detection process And an execution process for executing the second layer process of the two-layer commitment process of the transaction on the resource manager based on the elapsed time from the completion of the first layer process of the commitment process.

  The present invention is configured and operated as described above, so that when a failure occurs, the lock time of the resource manager can be minimized, and the stop time of the entire system can be minimized. Is obtained.

  Next, embodiments of the present invention will be described with reference to the drawings. First, an outline of a distributed transaction processing system according to the present invention will be described. A distributed transaction processing system according to the present invention is a system in which a terminal device, a server device, and a load distribution device are connected by a LAN (Local Area Network), and data is updated from the terminal device to the server device.

  In the system described above, the server device includes a resource manager and a relay device. The relay device includes a disconnection detection unit, a reception unit, a transmission unit, a storage unit, and a progress determination unit. When it becomes difficult to continue the transaction, the relay device determines the transaction, thereby minimizing the lock time of the resource manager and minimizing the downtime of the entire system.

  The distributed transaction processing system according to the present invention has a plurality of resource managers as in the background art described above, and the transaction manager executes a two-layer commitment procedure in order to correctly perform the commitments of the plurality of resource managers. Suppose you are.

  In the distributed transaction processing system according to the present invention, when a failure occurs in the server device, the state of the transaction is determined based on the elapsed time from the first layer processing, thereby continuing the processing in the entire system. Thus, in the distributed transaction processing system according to the present invention, the resource manager lock time can be minimized by determining the transaction to be solved by the transaction manager on the resource manager side, thereby minimizing the downtime of the entire system. can do.

  FIG. 1 is a block diagram showing a configuration example of a distributed transaction processing system according to the first exemplary embodiment of the present invention. In FIG. 1, the distributed transaction processing system according to the first embodiment of the present invention operates two server apparatuses 1-1 and 1-2 connected to a load distribution apparatus 5 and a resource manager such as a database. The server device 2 and the server device 3, terminal devices 4-1 and 4-2, and a network connecting them are configured.

  The server device 1-1 includes an application server 11-1, and the application server 11-1 includes a transaction manager 111-1 and an application 112-1. The server apparatus 1-2 includes an application server 11-2, and the application server 11-2 includes a transaction manager 111-2 and an application 112-2.

  The server device 2 includes a relay device 21 and a resource manager 22, and the relay device 21 includes a disconnection detection unit 211, a reception unit 212, a transmission unit 213, a storage unit 214, and a progress determination unit 215. It is out. The server device 3 includes a resource manager 31.

  The transaction manager 111-1 operates on the application server 11-1 of the server device 1-1 and manages distributed transactions between the resource manager 22 and the resource manager 31.

  The application 112-1 receives a request from the terminal device 4-1 and the terminal device 4-2, updates information of the resource manager 22 and the resource manager 31, and commits or rolls back the transaction manager 111-1. Make a request.

  The disconnection detection unit 211 operates on the relay device 21 of the server device 2 and detects the operation stop of the transaction managers 111-1 and 111-2. The receiving unit 212 operates on the relay device 21 of the server device 2 and receives communications from the applications 112-1 and 112-2 and the transaction managers 111-1 and 111-2 to the resource manager 22.

  The transmission unit 213 operates on the relay device 21 of the server device 2, transmits the communication content received by the reception unit 212 to the resource manager 22, and transmits a transaction completion notification to the resource manager 22 when the resource manager 22 stops. To do.

  The storage unit 214 operates on the relay device 21 of the server device 2 and holds the transaction state determined by the relay device 21.

  2 and 3 are sequence charts showing the flow of operations when a transaction ends normally in the distributed transaction processing system according to the first exemplary embodiment of the present invention. FIG. 4 is a sequence chart illustrating the first exemplary embodiment of the present invention. It is a flowchart which shows the flow of operation | movement when abnormality has generate | occur | produced in the server apparatus in the distributed transaction processing system by a form.

  FIG. 5 is a sequence chart showing an operation flow when the server apparatus recovers from the failure in the distributed transaction processing system according to the first exemplary embodiment of the present invention. The overall operation of the distributed transaction processing system according to the first embodiment of the present invention will be described with reference to FIGS. 2 to 5, the processing executed by the relay device 21 of the server device 2 is realized by a CPU (Central Processing Unit) (not shown) executing a program.

  In the following description, it is assumed that processing is performed from the terminal device 4-1 to the server device 1-1 via the load distribution device 5. Further, the request from the terminal device 4-1 updates the resource manager 22 and the resource manager 31 on the server device 2 and the server device 3.

Less than,
-Operation when it ends normally-Operation when an abnormality occurs (Case 1: An abnormality occurs in the server device 1-1 when the first layer processing is completed)
Operation when an abnormality occurs (Case 2: An abnormality occurs in the server device 1-1 when the first layer processing is not completed)
-Operation when an abnormality occurs (Case 3: An abnormality occurs in the server device 1-1 when the second layer processing is completed)
・ Operation when recovering after an error occurs (server device 1-1 is restarted)
The operation of each case will be described.

  First, a description will be given of processing when the processing ends normally, that is, when the two-layer commitment processing from the transaction manager 111-1 to the resource manager 22 and the resource manager 31 ends normally.

  2 and 3, first, the terminal device 4-1 issues a transaction processing request to the load balancer 5 (a1 in FIG. 2). The load distribution device 5 transfers the transaction processing request received from the terminal device 4-1 to the server device 1-1 connected (a 2 in FIG. 2).

  When receiving the transaction processing request, the application server 11 of the server apparatus 1-1 starts a transaction using the transaction manager 111-1. The transaction started here is [T1] (a3 in FIG. 2).

  The application 112-1 on the application server 11 requests the relay device 21 to perform update processing in order to update the resource manager 22 (a 4 in FIG. 2).

  When receiving the update request from the application 112-1, the reception unit 212 on the relay device 21 transmits the update request to the resource manager 22 using the transmission unit 213 (a5 in FIG. 2). Thereafter, the resource manager 22 updates the data. After the resource manager 22 is updated, the storage unit 214 of the relay device 21 records the current time, transaction information, and information of the server device 1-1 (a6 in FIG. 2).

  The application 112-1 on the application server 11 updates the resource manager 31 (a7 in FIG. 2). After completing the process of the application 112-1, the application server 11 performs the commitment process for the transaction [T1] using the transaction manager 111-1 (a8 in FIG. 2).

  The transaction manager 111-1 starts the first layer processing of the resource manager 22, and issues a first layer processing start request to the relay device 21 (a9 in FIG. 2).

  When receiving the first layer processing start request, the reception unit 212 on the relay device 21 transmits the first layer processing start request to the resource manager 22 using the transmission unit 213 (a10 in FIG. 2). The resource manager 22 executes the first layer process.

  After the completion of the first layer process, the storage unit 214 on the relay device 21 updates the time stored in the process a6 described above with the current time, and stores the completion state of the first layer process (see FIG. 3 a11).

  After the completion of the first layer processing of the resource manager 22, the transaction manager 111-1 requests the first layer processing from the resource manager 31 (a12 in FIG. 3). The resource manager 31 executes the first layer process.

  After the first layer processing is completed in the resource manager 22 and the resource manager 31, the transaction manager 111-1 starts the second layer processing. Similar to the above first layer processing, the transaction manager 111-1 issues a second layer processing request from the resource manager to the relay device 21 (a13 in FIG. 3).

  The receiving unit 212 on the relay device 21 that has received the request for starting the second layer processing transmits the start of the second layer processing request to the resource manager 22 using the transmitting unit 213 (a14 in FIG. 3). The resource manager 22 executes the second layer process.

  After the completion of the second layer process, the storage unit 214 on the relay device 21 deletes the information stored in the process a6 and the process a11 (a15 in FIG. 3).

  The transaction manager 111-1 requests the resource manager 31 to start the second layer process (a16 in FIG. 3). The resource manager 31 executes the second layer process.

When the resource manager 22 and the resource manager 31 complete the second layer process, the transaction [T1] is completed (a17 in FIG. 3).
Next, processing when an abnormality occurs (case 1), that is, when a failure occurs in the server device 1-1 when the first layer processing is completed will be described.

  Referring to FIG. 4, first, when the disconnection detection unit 211 of the relay device 21 detects the stop of the server device 1-1 (step S1 in FIG. 4), the progress determination unit 215 acquires the transaction state from the storage unit 214 ( FIG. 4 step S2).

  The progress determination unit 215 determines a transaction based on the difference between the current time and the time recorded in the storage unit 214 (step S3 in FIG. 4). The transmission unit 213 requests the resource manager 22 for a list of transactions that have been processed by the first layer process (step S4 in FIG. 4). The resource manager 22 returns the result of the first layer process to the relay device 21.

  Based on the transaction determination, the relay device 21 transmits the second layer process to the transaction registered in the storage unit 214 (step S5 in FIG. 4). The recording unit 214 updates the registered transaction information (step S6 in FIG. 4).

  With these processes, the update process of the resource manager 22 is normally completed.

  Next, processing when an abnormality occurs (case 2), that is, when a failure occurs in the server device 1-1 when the first layer processing is not completed will be described.

  First, when the disconnection detection unit 211 of the relay device 21 detects the stop of the server device 1-1, the progress determination unit 215 acquires the transaction state from the storage unit 214. In this case, since the transaction state is not recorded in the storage unit 214, the progress determination unit 215 determines to roll back as a transaction.

  The relay device 21 requests a rollback process for the transaction registered in the storage unit 214 based on the determination of the transaction. The recording unit 214 deletes the registered transaction information.

  By these processes, all transactions for which the first layer process has not been completed complete the rollback process.

  Furthermore, a process when an abnormality occurs (Case 3), that is, when a failure occurs in the server device 1-1 when the second layer process is completed will be described.

  First, when the disconnection detection unit 211 of the relay device 21 detects the stop of the server device 1-1, the progress determination unit 215 acquires the transaction state from the storage unit 214. In this case, since the transaction state is not recorded in the storage unit 214, the progress determination unit 215 does not perform the subsequent processing. Since the resource manager 22 has completed the processing, the transaction processing from the server device 1-2 is not hindered.

  Next, a case where the server apparatus 1-1 recovers from a failure, that is, a case where the transaction manager 111-1 resumes processing will be described.

  The transaction manager 111-1 executes transaction recovery processing when a transaction is determined, that is, when the first layer processing for all the resource managers 22 and 31 is completed.

  Referring to FIG. 5, after the server apparatus 1-1 is restarted, the transaction manager 111-1 issues a transaction confirmation request in order to confirm the transaction state of the resource manager 22 (b1 in FIG. 5).

  When receiving the transaction status confirmation request, the receiving unit 212 on the relay device 21 returns the transaction status recorded in the recording unit 214 (b2 in FIG. 5).

  The transaction manager 111-1 matches the transaction state, and resumes transaction processing if the transaction of the resource manager 31 is not completed (b3 in FIG. 5). At this time, if there is a discrepancy between the status returned from the transaction manager 111-1 and the relay device 21, the transaction manager 111-1 outputs the status to a log or the like.

  Through the above processing, the transaction in progress is completed or is in a state inconsistent state and is completed. For transactions that do not match the status, the operation administrator manually matches the status.

  As described above, in the present embodiment, the relay device 21 detects the abnormal end of the server device 1-1 including the transaction manager 111-1, and independently determines the transaction, so the recovery of the transaction manager 111-1 is awaited. Instead, the resource managers 22 and 31 are updated.

  In the present embodiment, since the recovery of the transaction manager 111-1 is not requested, the server apparatus 1-1 including the application server 11 does not need to take a special configuration such as a cluster configuration, and the apparatus is restarted. Even in such a situation, processing from other server devices is not interrupted by a lock for updating the resource managers 22 and 31 or the like.

  Therefore, in this embodiment, since the relay device 21 determines the transaction and updates the resource managers 22 and 31, the resource can be quickly acquired regardless of the state of the transaction manager 111-1 when the transaction manager 111-1 fails. The locks of the managers 22 and 31 can be released.

  Therefore, in this embodiment, when a failure occurs in the server device 1-1, the lock time of the resource managers 22 and 31 can be minimized, and the stop time of the entire system can be minimized.

  In the above embodiment, the case of updating data using two server devices has been described. However, the number of server machines, the number of terminal devices, and the number of resource managers are not limited. In addition, the lease manager has a function to cooperate with the transaction manager, such as a database and a message queue system, and has no functional limitation.

  The present invention can be applied to applications such as an information updating apparatus that refers to a plurality of databases and a program for realizing the information updating apparatus in a computer.

It is a block diagram which shows the structural example of the distributed transaction processing system by the 1st Embodiment of this invention. It is a sequence chart which shows the flow of operation | movement when the transaction in the distributed transaction processing system by the 1st Embodiment of this invention is complete | finished normally. It is a sequence chart which shows the flow of operation | movement when the transaction in the distributed transaction processing system by the 1st Embodiment of this invention is complete | finished normally. It is a flowchart which shows the flow of operation | movement when abnormality has generate | occur | produced in the server apparatus in the distributed transaction processing system by the 1st Embodiment of this invention. It is a sequence chart which shows the flow of operation | movement when a server apparatus recovers from a failure in the distributed transaction processing system by the 1st Embodiment of this invention.

Explanation of symbols

1-1, 1-2,2,3 server device
4-1, 4-2 Terminal equipment
5 Load balancer 11-1, 11-2 Application server
21 Relay device
22, 31 Resource manager 111-1, 111-2 Transaction manager 112-1, 112-2 Application
211 Cutting detection unit
212 Receiver
213 Transmitter
214 Storage unit
215 progress judgment part

Claims (19)

A distributed transaction processing system including a transaction manager and a relay device having a resource manager , wherein the resource manager performs a two-layer commitment process of a transaction based on an instruction of the transaction manager,
The relay device receives and transmits the instruction from the transaction manager and transmits to the resource manager, detection means for detecting an abnormal end of the transaction manager, and when the detection means detects the abnormal end And a means for executing a second layer process of the two-layer commitment process of a transaction on the resource manager based on an elapsed time from the completion of the first layer process of the two-layer commitment process. Transaction processing system.   2. The distributed transaction processing system according to claim 1, wherein when the abnormal termination is detected, the resource manager is updated without waiting for the recovery of the transaction manager. The relay device includes storage means for holding a transaction state determined by the relay device,
Whether an abnormality has occurred in the transaction manager when the first layer processing has been completed, or has an abnormality occurred in the transaction manager in a state where the first layer processing has not been completed, based on the storage contents of the storage means 3. The distributed transaction processing system according to claim 1, wherein it is determined whether an abnormality has occurred in the transaction manager in a state where the second layer processing is completed.   4. The distributed transaction processing system according to claim 3, wherein the relay apparatus determines the transaction based on the storage contents of the storage unit and updates the resource manager.   The transaction information on which the first layer processing has been completed is obtained from the resource manager with reference to the storage means, and the transaction that has completed the first layer processing is determined in an appropriate state. The distributed transaction processing system according to claim 3 or claim 4.   6. The distributed transaction processing system according to claim 1, wherein a transaction consistency check is performed between the transaction manager and the relay device when recovery from the abnormal termination. A server device used in a distributed transaction processing system including a transaction manager and a relay device having a resource manager , wherein the resource manager performs a two-layer commitment process of a transaction based on an instruction of the transaction manager,
Transmission / reception means for receiving the instruction from the transaction manager and transmitting it to the resource manager, detection means for detecting abnormal termination of the transaction manager, and the two-layer commitment processing when the detection means detects the abnormal termination A server apparatus comprising: an execution unit configured to execute a second layer process of the two-layer commitment process of a transaction on the resource manager based on an elapsed time from the completion of the first layer process.   8. The server apparatus according to claim 7, wherein when the abnormal end is detected, the resource manager is updated without waiting for the transaction manager to recover. Storage means for holding the state of the determined transaction;
Whether an abnormality has occurred in the transaction manager when the first layer processing has been completed, or has an abnormality occurred in the transaction manager in a state where the first layer processing has not been completed, based on the storage contents of the storage means 9. The server device according to claim 7, wherein whether or not an abnormality has occurred in the transaction manager in a state where the second layer processing has been completed is determined.   The server apparatus according to claim 9, wherein the resource manager is updated by determining the transaction based on a storage content of the storage unit.   The transaction information on which the first layer processing has been completed is obtained from the resource manager with reference to the storage means, and the transaction that has completed the first layer processing is determined in an appropriate state. The server device according to claim 9 or claim 10.   12. The server apparatus according to claim 7, wherein a transaction consistency check is performed with the transaction manager when the abnormal end is recovered. A distributed transaction failure recovery method used in a distributed transaction processing system including a transaction manager and a relay device having a resource manager , wherein the resource manager performs two-layer commitment processing of transactions based on an instruction of the transaction manager,
The relay device receives the instruction from the transaction manager and transmits it to the resource manager, a detection process for detecting abnormal termination of the transaction manager, and the abnormal termination is detected in the detection process And executing an execution process for executing a second layer process of the two-layer commitment process of a transaction on the resource manager based on an elapsed time from the completion of the first layer process of the two-layer commitment process. A failure recovery method for distributed transactions.   14. The distributed transaction failure recovery method according to claim 13, wherein the resource manager is updated without waiting for recovery of the transaction manager when the abnormal termination is detected. In the relay device, the state of the transaction determined by the relay device is held in the storage means,
Whether an abnormality has occurred in the transaction manager when the first layer processing has been completed, or has an abnormality occurred in the transaction manager in a state where the first layer processing has not been completed, based on the storage contents of the storage means 15. The distributed transaction failure recovery method according to claim 13 or 14, further comprising: determining whether an abnormality has occurred in the transaction manager in a state where the second layer processing has been completed.   16. The distributed transaction failure recovery method according to claim 15, wherein the relay device determines the transaction based on the storage contents of the storage means and updates the resource manager.   The relay device refers to the storage means, obtains transaction information that the first layer processing has been completed from the resource manager, and determines that the transaction that has completed the first layer processing is in an appropriate state. The distributed transaction failure recovery method according to claim 15 or 16, characterized in that:   18. The distributed transaction failure recovery method according to claim 13, wherein a transaction consistency check is performed between the transaction manager and the relay device when recovered from the abnormal termination. . A program executed by a central processing unit in the relay device in a distributed transaction processing system including a transaction manager and a relay device having a resource manager , wherein the resource manager performs two-layer commitment processing of transactions based on instructions from the transaction manager. There,
A transmission / reception process for receiving the instruction from the transaction manager and transmitting it to the resource manager, a detection process for detecting an abnormal end of the transaction manager, and the two layers when the abnormal end is detected in the detection process An execution process for executing a second layer process of the two-layer commitment process of a transaction on the resource manager based on an elapsed time from the completion of the first layer process of the commitment process.

Images