JP2014137798A - Database system and control method for database system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce an influence when system switching related to abnormality occurs in a database client having a redundant data processing part.SOLUTION: The database system includes: a database server; and a database client formed with redundant configurations. Then, the database system includes: means for detecting the abnormality of a first data processing part which operates as a current system; means for, when the abnormality of the first data processing part is detected, performing processing of switching a second data processing part which operates as a standby system to a current system; means for collecting session information related to a connection state session; means for specifying the connection state session on the basis of the collected session information, and for allowing a database server to execute the disconnection of the connection state session; and means for executing the rollback of data related to the connection state session in the case of disconnecting the connection state session.

Description

  The present invention relates to a database system and a database system control method, and can be applied to, for example, a database system including a redundantly configured database client.

  In the database system, a database server and a database client that uses data of the database server are arranged. When a conventional database client updates data on the database server, the data update is usually performed by requesting the database server to update the data and executing a procedure for committing the updated data as one transaction. Do. In this specification, data update on the database server is a term including registration of new data, update of arbitrary data recorded in the database, deletion of arbitrary data recorded in the database, and the like. To do.

  In the conventional database system, when an abnormality occurs in the database server or the database client, the database server performs the process of rolling back the data that has not been committed after the update, if necessary. Some maintain consistency.

  As a conventional database system corresponding to the rollback function, there is a technique described in Patent Document 1. In the system described in Patent Document 1, when a system restart or the like is required, a recovery log (operation history log) is used in order to execute rollback processing that efficiently recovers data consistency. It describes that recording is performed using a plurality of types of data recording media (low-speed nonvolatile storage area and high-speed volatile storage area).

Japanese Patent Laid-Open No. 07-084815

  Incidentally, in some conventional database systems, a data processing unit (computer such as a server) that performs data processing is made redundant on the database client side that uses the database server. In such a redundant database client, when a redundant configuration of clustering is built in the data processing unit, a failure (in the first data processing unit of the active system (hereinafter also referred to as “ACT system”) ( When an abnormality occurs, the second data processing unit of the standby system (hereinafter also referred to as “standby system” or “SBY system”) operates as the ACT system by system switching.

  However, in the session that was held when an abnormality occurred in the old ACT computer, the data being updated on the database (data that was not committed after the update) was held (data update in other sessions) If the data is not released, the state where the data is held is continued. As a result, the second data processing unit newly operating as the ACT system cannot immediately update the data held in the first data processing unit, which hinders the operation immediately after system switching. Will occur.

  Therefore, in view of the problems as described above, a database system in which a data processing unit that accesses data of a database server is made redundant, and a database system that can reduce the influence when system switching occurs due to an abnormality, and A method for controlling a database system is desired.

  The first aspect of the present invention is a database system including a database server and a database client in which a redundant configuration is formed by a plurality of data processing units. (1) The first data processing unit operates as an active system. When the abnormality detection unit detects that an abnormality has occurred in the first data processing unit, and (2) when the abnormality detection unit detects an abnormality in the first data processing unit, System switching means for performing processing for switching the operating second data processing unit to the active system; and (3) a connection state session that was in connection with the database server when an abnormality occurred in the first data processing unit. Session information collecting means for collecting the session information according to (4) the database based on the session information collected by the session information collecting means. Session disconnection means for identifying a connection state session related to the first data processing unit on the server and causing the database server to disconnect the connection state session related to the first data processing unit; (5) the above When the session disconnecting means disconnects the connection state session, the database server has a rollback means for rolling back the data updated in the connection state session to the state before the update by the connection state session. It is characterized by.

  According to a second aspect of the present invention, there is provided a database system control method comprising a database server and a database client in which a redundant configuration is formed by a plurality of data processing units. (1) Abnormality detection means, system switching means, session information collection Means, a session disconnecting means, and a rollback means. (2) The abnormality detecting means has an abnormality in the first data processing unit when the first data processing unit is operating as an active system. (3) The system switching means sets the second data processing section operating as a standby system to the active system when the abnormality detection means detects an abnormality of the first data processing section. (4) The session information collection means is connected to the database server when the first data processing unit is abnormal. (5) The session disconnecting means is connected to the first data processing unit on the database server based on the session information collected by the session information collecting means. And the database server executes disconnection of the connection state session related to the first data processing unit. (6) When the session disconnection unit disconnects the connection state session, On the database server, the data updated in the connection state session is rolled back to the state before the update by the connection state session.

  ADVANTAGE OF THE INVENTION According to this invention, the database system which can reduce the influence at the time of system switching can be provided by the database client with which the data processing part which accesses the data of a database server was made redundant.

It is the block diagram shown about the whole structure of the database system which concerns on embodiment. It is the sequence diagram shown about the basic operation | movement of the record update in the database system which concerns on embodiment. In the database system according to the embodiment, it is an explanatory diagram showing a state transition when an abnormality occurs in an ACT-based application process. It is the flowchart shown about the operation | movement when system switching generate | occur | produces in the database client which concerns on embodiment. It is the flowchart shown about the operation | movement when the session cutting | disconnection (including forced rollback process) accompanying system switching generate | occur | produces in the database client which concerns on embodiment. In the database system which concerns on embodiment, it is explanatory drawing (the 1) shown about the transition of a state when system switching generate | occur | produces. In the database system which concerns on embodiment, it is explanatory drawing (the 2) shown about the transition of a state when system switching generate | occur | produces.

(A) Main Embodiment Hereinafter, an embodiment of a database system and a database system control method according to the present invention will be described in detail with reference to the drawings.

(A-1) Configuration of Embodiment FIG. 1 is a block diagram showing the overall configuration of the database system 1 of this embodiment.

  It is assumed that the database server 10 and the database client 20 are arranged in the database system 1. Note that a network connection N1 (the connection means is not limited) is set between the database server 10 and the database client 20, and the number of database clients 20 is not limited.

  The database server 10 includes a server 11 and a file system 12.

  The file system 12 is a storage for storing data files. In FIG. 1, the file system 12 is illustrated as an independent device, but may be constructed (configured on a built-in storage) as a device integrated with the server 11.

  The server 11 is a device (computer) that processes data stored in the file system 12 and provides a database service to the database client 20. On the server 11, a server process 111 operates as a database service application. That is, a database service program is installed in the server 11, and the server process 111 is started by executing the program on the server 11.

  It is assumed that the file system 12 has a data storage area 121 for storing data to be managed itself and a log storage area 122 for realizing data update rollback and the like. The server process 111 manages the status of each piece of data in the data storage area 121 for each processing unit (hereinafter, this processing unit is referred to as “record”). The server process 111 also records each record in the data storage area 121. When the update is performed, the history information related to the update is also stored in the log storage area 122 in a rollable format.

  The database client 20 is a device (computer) that receives a database service from the database server 10. It should be noted that the database client 20 is not a client terminal that simply uses a database, but may provide an application service (for example, a function of a web server or a call center server) using another database to other terminals. Of course. Further, it is assumed that the database client 20 has a redundant configuration with a plurality of servers 30 (computers).

  In this embodiment, the database client 20 is assumed to have a redundant configuration with two servers 30 (30-1 and 30-2). The number of redundant systems (servers 30) in the database client 20 is not limited.

  The database client 20 has a redundant configuration (clustering configuration) in which one of the plurality of servers 30 operates as an ACT system and the other servers 30 operate as an SBY system.

  That is, the database client 20 has a clustering configuration in which a plurality of hardware (servers 30-1 and 30-2) operate logically as one server. A network connection N2 (communication path) capable of mutual communication is set between the servers 30 in order to realize a clustering configuration. The communication method of the network connection N2 between the servers 30 is not limited. For example, various communication methods (interfaces) such as various Ethernet (registered trademark) and fiber channel can be applied. In this embodiment, the network connections N1 and N2 are described as separate communication paths. However, the same communication path (communication medium) may be shared.

  Next, the internal configuration of each server 30 will be described. In this embodiment, each server 30 is described as having the same configuration.

  Each server 30 has a configuration in which an OS and various types of software are installed on hardware 31 (computer), and the functional configuration in the activated state can be illustrated as shown in FIG. In FIG. 1, a failure management 33 and a system configuration management 34 operate as middleware for managing a redundant configuration system and the like on each server 30 on a kernel 32 activated by an OS function. In each server 30, an application process 35 operates as a database client on the above-described middleware.

  As described above, there may be one or more database clients 20, and it is possible to divide a server for each service to be provided or to provide the same service by load balancing.

  In the database client 20, the application process 35 that provides a service is normally an ACT system process, and the SBY application process 35 is in an idle (or unstarted) state. In the database client 20, such clustering configuration control (for example, system switching, start / stop control of the application process 35, etc.) is performed by mutual communication between the middleware (failure management 33 and system configuration management 34) of the server 30. Done.

  In FIG. 1, the logical configuration is illustrated by one line as the network connection N <b> 1 between the database client 20 and the database server 10. However, actually, the network connection N1 needs to have a connection configuration in which each server 30 can communicate with the database server 10 regardless of the operation of the other servers 30. For example, each server 30 may be configured to have a network interface and be connected to the network (for example, VRRP (Virtual Router Redundancy Protocol)), or a communication control device (for example, layer 2) not shown in the database client 20. A switch device such as a switch) may be provided so that each server 30 is connected to the network. Further, in FIG. 1, the database server 10 side is not shown as a redundant configuration, but it is a matter of course that a redundant configuration may be used.

  Next, data processing performed in cooperation between the database client 20 (application process 35) and the database server 10 (server process 111) will be described.

  It is assumed that data processing between the application process 35 and the server process 111 is executed in units of transactions. Then, between the application process 35 and the server process 111, the processing performed from the start to the end of one session will be described as being managed as one transaction. For example, a unique ID is assigned to one session (transaction) between the application process 35 and the server process 111.

  The server process 111 executes transaction processing while managing the state of each record (access management) in order to maintain the consistency of each data. In this embodiment, as an example, the server process 111 can access each record from a “contention lock state” that prohibits access from other than the session (transaction) currently being processed, or from any session (transaction). It shall be managed in any of the “released state”. When the server process 111 updates the record of the record in the data storage area 121 in response to a request from the database client 20, the server process 111 manages the record as a “contention lock state”. After that, the server process 111 commits the record in response to a request from the database client 20 (updated application process 35), and manages the record as a released state. The server process 111 manages the data update from other transactions as a waiting state for the record in the contention lock state, and updates the record after entering the release state.

(A-2) Operation of Embodiment Next, the operation of the database system 1 of this embodiment having the above configuration (the database system control method of the embodiment) will be described.

(A-2-1) Basic Sequence of Data Update FIG. 2 shows a transaction when updating arbitrary data (hereinafter referred to as “record A”) on the database server 10 (data storage area 121 of the file system 12). It is the sequence diagram shown about the example of the process.

  Here, it is assumed that the database client 20 is initially operating as the server 30-1 as an ACT system and the server 30-2 as an SBY system.

  First, the application process 35 of the ACT system (server 30-1) in the database client 20 transmits a session connection request (session connection request related to database processing) to the database server 10 (server process 111 of the server 11) (S101). Assume that the database server 10 (server process 111) responds (connects OK) to the transmission request (S102).

  The ACT application process 35 may generate a session each time access to the database server 10 is required, or may exclusively use a connection session prepared in advance.

  Then, the ACT application process 35 holds a session with the database server 10 (server process 111) and then transmits an update request for changing the content of an arbitrary record A (S103). ).

  Then, it is assumed that the database server 10 (server process 111) that has received the update request for record A has performed update processing on record A in the file system 12 (data storage area 121) (S104). Here, it is assumed that the initial state of the record A is a released state, and the update of the record A by the database server 10 (server process 111) is successful. Then, the database server 10 (server process 111) manages the record A as a contention lock state. Thereafter, the database server 10 (server process 111) manages an update request from another connection session (transaction) as a waiting state for the record A until the update is confirmed by the commit. Further, the database server 10 (server process 111) stores, in the log storage area 122, log information for enabling the rollback of the record A in the data storage area 121 to the state before the update.

  Then, the database server 10 (server process 111) responds to the update result (update success) of the record A to the ACT system application process 35 (S105).

  The ACT application process 35 that has received the update result indicating the successful update transmits a commit request for the record A to the database server 10 (server process 111) (S106).

  The database server 10 (server process 111) that has received the commit request for record A executes the commit process for record A (the process for releasing the contention lock state and releasing it) (S107), and the result of the commit process (Commit success) is returned to the ACT application process 35 (S108).

  As described above, the database server 10 (server process 111) executes update record confirmation by commit in units of transactions. Further, in the database server 10 (server process 111), when the commit process is performed, the contention lock state of the record A is released, and data is updated from another transaction to the record A (such as a transaction in a waiting state). Is possible. As a result, the database server 10 (server process 111) ensures that no content conflict occurs (maintains consistency) in updating related records.

  Then, the ACT application process 35 requests the database server 10 (server process 111) to disconnect the session (S109), and a series of transaction (session) processing ends.

  In the ACT application process 35, it is necessary to manage details of a series of processes (from connection session generation to disconnection) shown in FIG. 1 on a memory (this procedure unit is referred to as “context” below). Or “handler”).

(A-2-2) When an error occurs after record update and before commit processing Next, in the sequence of FIG. 2 described above, the database system 1 (server process 111) executes update of record A and then commits. An operation when the ACT application process 35 enters an abnormal state (for example, a state in which a core dump is output) before processing (after the processing in step S103 to before the processing in step S107) will be described.
FIG. 3 shows a state in which the ACT-related application process 35 becomes abnormal between the process of step S103 and before the process of step S107 in the sequence of FIG. That is, at this time, since the commit process for the record A has not been performed, the record A is in the contention lock state.

  In FIG. 3, at this time, the ACT system application process 35 (the application process 35 of the server 30-1) becomes an abnormal state, the failure management 33 detects the abnormal state, and the system configuration management 34 switches the system. It shall be started. At this time, the operation when system switching is executed by the database client 20 will be described with reference to the flowcharts of FIGS.

  4 and 5 show an operation in which the server 30-1 operating as the ACT system is switched to the SBY system and the server 30-2 operating as the SBY system is switched to the ACT system. 4 and 5, the server 30-1 operating as the ACT system is illustrated as “old ACT system” or “new SBY system”. 4 and 5, the server 30-2 operating as the SBY system is illustrated as “new ACT system” or “old SBY system”.

  First, system switching is determined by the system configuration management 34 of the old ACT system (server 30-1), and a system switching request is notified to the system configuration management 34 of the old SBY system (server 30-2) through communication. Shall be. In the old SBY system (server 30-2), it is assumed that the system configuration management 34 has requested the application process 35 to start as a new ACT system (restart activation request). Then, it is assumed that the application process 35 of the old SBY system (server 30-2) has accepted the restart activation request (S201).

  And the application process 35 of the server 30-2 will start the process which changes to the new ACT type | system | group from the old SBY type | system | group, if a restart starting request | requirement is received. Specifically, the application process 35 of the server 30-2 is incomplete in the old ACT system application process 35 in cooperation with the database server 10 when transitioning from the old SBY system to the new ACT system. Session disconnection processing that forcibly disconnects the session (session session) is performed. Along with this session disconnection process, the database server 10 forcibly rolls back a record in a contention lock state related to a session that has not been completed in the old ACT application process 35 (hereinafter, “ (Referred to as “forced rollback process”) (S202).

  The session related to the transaction that has not been completed in the old ACT system application process 35 continues on the database server 10 side, and the state of the record A remains in the contention lock state. Therefore, when the application process 35 of the server 30-2 transitions from the old SBY system to the new ACT system, by executing session disconnection processing (including forced rollback processing), the state on the database server 10 side is changed. The old ACT system application process 35 is restored to the state before starting the session (transaction). That is, by session disconnection processing (including forced rollback processing), record A is rolled back to the state before update, released from the contention lock state, returned to the release state, and data update by another session (transaction) is also performed. It becomes possible.

  When the session disconnection process ends, the application process 35 of the server 30-2 executes an initial setting process for transitioning from the old SBY system to the new ACT system (S203), restarts the service (S204), and Thus, a new update request can be sent.

  Next, details of session disconnection processing (including forced rollback processing) by the application process 35 of the server 30-2 will be described with reference to the flowchart of FIG.

  When the application process 35 of the server 30-2 starts session disconnection processing, it connects a session with the database server 10 (server process 111) (S301), and continues session information (transaction) related to the old ACT system application process 35. (Information) is collected (S302).

  Here, the format of the session information collected by the application process 35 of the server 30-2 is not limited. For example, a list of IDs of sessions currently connected by the server process 111 may be used.

  Then, the application process 35 of the server 30-2 confirms whether or not there is an ongoing session related to the old ACT application process 35 from the collected session information (S303). Then, the application process 35 of the server 30-2 operates from the process of step S304 described later when the corresponding session is detected, and ends the session disconnection process when the corresponding session cannot be detected.

  The method in which the application process 35 of the server 30-2 confirms whether there is an ongoing session related to the old ACT application process 35 from the collected session information is not limited. The application process 35 of the server 30-2 acquires, for example, the core dump output from the application process 35 from the old ACT server 30-1 via the system configuration management 34, and the old ACT application process is obtained from the core dump. 35 sessions (session IDs) may be identified and collated with collected session information.

  When a corresponding session is detected, the application process 35 of the server 30-2 designates the session and requests the database server 10 (server process 111) to disconnect. Then, the session processing requested on the database server 10 (server process 111) side is executed. Then, the database server 10 (server process 111) notifies the application process 35 of the server 30-2 of the session disconnection result (success or failure of disconnection) (S304).

  In this case, it is necessary for the application process 35 of the server 30-2 to be given an authority capable of disconnecting even a session from another application process 35 on the database server 10. At this time, the application process 35 of the server 30-2 may request disconnection of a plurality of sessions. For example, when the application process 35 has a multi-thread configuration and each thread generates a connection session, there are a plurality of connection sessions from the old ACT system.

  FIG. 6 is an explanatory diagram showing the state of the database system 1 when the new ACT application process 35 at the time of system switching forcibly disconnects the connection session from the old ACT system at the time of restart activation processing.

  As shown in FIG. 6, the database server 10 (server process 111) determines that the transaction related to the disconnected session is no longer completed due to the session disconnection based on the request, and the transaction is retained. Rollback processing is performed on the record being held (the record that maintains the contention lock state and is waiting for commit). In other words, the database server 10 selects the rollback instead of the commit when the communication with the database client 20 is abandoned with respect to the record being updated that has not been committed. Write back to Note that when an abnormal state is detected in the old ACT system (when an abnormality is detected in the application process 35), which is a condition for starting the session disconnection process, the application process 35 cannot complete the transaction being executed. The connection session disconnection and rollback processing is an expected operation.

  Then, the application process 35 of the server 30-2 confirms whether or not all sessions requested to be disconnected (including forced rollback processing) have succeeded (S305). The application process 35 of the server 30-2 ends the forced rollback process when all the sessions requested to be disconnected are successful, and returns to the above-described step S302 if not (incomplete). Execute the session disconnection process again.

  FIG. 7 is an explanatory diagram showing the state of the database system 1 after the restart process of the new ACT application process 35 at the time of system switching (after system switching).

  In the state of FIG. 7, the old ACT connection session disconnection process (including the forced rollback process) is completed by the new ACT application process 35. Therefore, in the state of FIG. 7, the database server 10 has no obstruction factor for the data update request related to the record A and can perform data update.

(A-3) Effects of Embodiment According to this embodiment, the following effects can be achieved.

  In the database client 20, when the ACT application process 35 is in an abnormal state, the new ACT application process 35 performs session disconnection and forced rollback processing related to the old ACT system at the time of system switching. Request. As a result, in the new ACT system application process 35, even immediately after the system switching, the effect at the time of system switching due to the abnormal state of the ACT system application process 35 (for example, the influence of the failure of the old ACT system) And waiting for record competition, etc.) can be reduced.

(B) Other Embodiments The present invention is not limited to the above-described embodiments, and may include modified embodiments as exemplified below.

(B-1) In the above embodiment, the session disconnection process illustrated in FIG. 5 is described as being performed by the new ACT server 30-1, but the application process 35 is performed on the old ACT server 30-1. This failure (core dump output) may be executed when the failure management 33 detects it. Further, it may be executed in both the old ACT system and the new ACT system, and the session disconnection process (including the forced rollback process) may be executed more reliably.

  In the process shown in FIG. 5, the database server 10 (server process 111) side rolls back the corresponding record when the session is disconnected, but the database client 20 side (server 30-2 application process 35 side). ) To request a rollback by designating a session or record.

(B-2) While the process of the flowchart of FIG. 5 described above is performed on the new ACT server 30-1, the old ACT application process 35 continues the abnormal state (core dump output state). Alternatively, the system configuration management 34 may forcibly terminate the process. This is because even if the application process 35 disappears due to the end of the core dump prior to the connection session disconnection request from the new ACT system, the rollback should be performed as expected.

  DESCRIPTION OF SYMBOLS 1 ... Database system, 10 ... Database server, 11 ... Server apparatus, 111 ... Server process, 12 ... File system, 121 ... Data storage area, 122 ... Log storage area, 20 ... Database client, 30, 30-1, 30- 2 ... server, 31 ... hardware, 32 ... kernel, 33 ... failure management, 34 ... system configuration management, 35 ... application process.

Claims (3)

In a database system comprising a database server and a database client in which a redundant configuration is formed by a plurality of data processing units,
An abnormality detection means for detecting that an abnormality has occurred in the first data processing unit when the first data processing unit is operating as an active system;
A system switching unit for performing processing for switching the second data processing unit operating as a standby system to the active system when the abnormality detection unit detects an abnormality in the first data processing unit;
Session information collection means for collecting session information relating to a connection state session that is in a connection state with the database server when the first data processing unit is abnormal;
Based on the session information collected by the session information collection means, the connection state session related to the first data processing unit on the database server is specified, and the connection related to the first data processing unit is connected to the database server Session disconnection means for executing state session disconnection;
When the session disconnection unit disconnects the connection state session, the database server has a rollback unit that rolls back the data updated in the connection state session to the state before the update by the connection state session. A database system characterized by that.   The database system according to claim 1, wherein the processes of the session disconnecting unit and the rollback unit are executed before the system switching process by the system switching unit. In a database system control method comprising a database server and a database client in which a redundant configuration is formed by a plurality of data processing units,
Anomaly detection means, system switching means, session information collection means, session disconnection means and rollback means,
The abnormality detection means detects that an abnormality has occurred in the first data processing unit when the first data processing unit is operating as an active system,
The system switching unit performs a process of switching the second data processing unit operating as a standby system to the active system when the abnormality detection unit detects an abnormality of the first data processing unit,
The session information collection means collects session information related to a connection state session that is in a connection state with the database server when the first data processing unit is abnormal,
The session disconnecting unit specifies a connection state session related to the first data processing unit on the database server based on the session information collected by the session information collecting unit, and sends the first session to the database server. Disconnect the connection status session related to the data processing unit,
When the session disconnection unit disconnects the connection state session, the rollback unit rolls back the data updated in the connection state session to the state before the update by the connection state session on the database server. And a database system control method.

Images