JP2006139696A - Data restoration method and data restoration program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To suppress access performance deterioration from a data base application at a check point, and to perform recovery time reduction at the time of failure recovery by reducing an amount of disk output data at the check point. <P>SOLUTION: In an on-memory data base system, a data base update processing function part 100 for managing a data area on a memory of the on-memory data base by splitting it into a plurality of segments outputs journal information to be history information on data update; a check point data acquisition area extraction function part 500 extracts a segment where the data have been updated and outputs it to a disk 1300 at the time of check point; and a recovery function part 1100 performs the pursuing processing of the check point data update stored in the disk 1300 on the basis of journal data after the immediate check point stored in the disk at the time of system activation. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to backup processing in an on-memory database that requires high-speed search / update performance.

  In a conventional on-memory database system, all database data is secured on a volatile memory, and high-speed data access is provided by referring to and updating the data. However, since data is secured on a volatile memory, it is necessary to guarantee data in some form against sudden power outages.

  Generally, in order to guarantee data in the event of a failure, any timing can be secured by ensuring timing called a checkpoint that permanently guarantees the data state at a certain moment, and journal information that is data update history between multiple checkpoints. The database status is guaranteed. Especially at the time of checkpoint, the entire data in the memory is continuously placed on the disk from the beginning in order to secure the contents close to the instantaneous state of the data on the memory that is frequently updated in the hard disk as a permanent area. By outputting, data is secured in a short time, and update information generated during data securing is secured in the form of journal information.

In the event of a failure, the memory space is secured upon restart, the checkpoint data secured at the latest checkpoint is expanded on the memory, and the data update by the journal information after the checkpoint is further followed. By implementing this, it is possible to restore a database state in which transaction processing that has been determined at an arbitrary time from immediately after securing a checkpoint file to immediately before the occurrence of a failure is performed (for example, see Non-Patent Document 1). ).
Kenneth Salem, Hector Garcia-Molina, “System M: A Transaction Processing Tested for Memory Resident Data”, IEEE Transactions on Knowledge. Vol. 2, No. 1. March 1990

  However, in the conventional data guarantee method, because the snapshot of the entire database memory is secured on the disk, disk I / O for several minutes is generated at the checkpoint timing, and the CPU resource is correspondingly large accordingly. Consume. This is a factor that degrades the database access performance from the application.

  In addition, the amount of journal information during the checkpoint file acquisition is considerable, so when recovering the database after a failure occurs, the journal information required for restoration up to the point where the minimum data consistency is guaranteed. Time is required for follow-up processing, which is particularly a problem in a system in which the service stop time in the event of a failure such as an exchange or finance must be shortened as much as possible.

  To shorten the follow-up process, the checkpoint interval can be shortened. However, as mentioned above, the checkpoint file output time is long and the load is high, so increasing the checkpoint frequency increases the overall system performance. Will not be a realistic countermeasure.

  In the present invention, the amount of disk output data is suppressed by changing the data structure that must be secured at the checkpoint, thereby suppressing deterioration in access performance from the database application at the checkpoint, and the recovery time at the time of failure recovery Shorten. In addition, a data storage technique at a checkpoint for realizing this and a data recovery technique based thereon are provided.

  In order to solve the above problem, the invention according to claim 1 is a data restoration method in an on-memory database system, wherein the on-memory database system divides a data area on the memory of the on-memory database into a plurality of segments. The database update processing function unit that manages and updates the data separately sends the journal information that is the data update history information to the disk, and the checkpoint data acquisition area extraction function unit updates the data. Extracting segment information and transmitting it to the disk at the time of a checkpoint; and a recovery function unit storing the information on the disk based on journal data after the previous checkpoint stored on the disk at the time of system startup Updated checkpoint data Performing a post-chase process, characterized by having a.

  The invention according to claim 2 is a data restoration method in an on-memory database system, wherein the on-memory database system manages a data area on a memory of the on-memory database by dividing it into a plurality of segments, and An update processing function unit receives an update request for data, determines a data segment on the memory to be updated, and rewrites the data; and a journal information output function unit includes: The journal information output step of receiving journal information as the data update history information and outputting it to the disk, and the update sub-segment management function unit receives the update target data segment from the database update processing function unit The corresponding subsegment update The subsegment update step that sets the lag to the updated state, and the checkpoint data acquisition area extraction function unit extracts the segment information corresponding to the updated subsegment update flag and determines the checkpoint data An extraction step, a checkpoint data output unit that outputs the checkpoint data to the disc at the time of the checkpoint, and a checkpoint data recorded on the disc by the recovery function unit and the previous checkpoint And a restoration step of restoring subsequent journal data to the memory area at the time of system startup and performing a follow-up process for updating checkpoint data based on the journal data.

  According to a third aspect of the present invention, the data restoration method according to the first or second aspect of the present invention is described by a computer program so as to be executable.

  According to the first to third aspects of the present invention, the entire data area of the database data in the memory is not output on the disk, but the data area in the memory is divided into segments each having a plurality of blocks, and the previous checkpoint Update information is managed in units of blocks for segments that have been updated thereafter.

  At checkpoint, only the relevant segment is output to the disk. At the time of recovery after a failure occurs, data is expanded from the latest generation segment data file group on the disk to memory, and the transaction is followed by journal information after the checkpoint is acquired. Execute.

  Such data restoration eliminates the need for disk output of the entire data in the conventional memory, reduces the required machine resources associated with shortening the checkpoint processing time, and the checkpoint interval can be set accordingly. By reducing the amount of journal information to be performed and reducing the amount of post-journal processing at the time of recovery, the system stop time can be shortened.

  According to the first to third aspects of the present invention, in the conventional on-memory database system, by reducing the disk output amount at the checkpoint, CPU resources and I / O resources necessary for the background of normal operation are suppressed, Data guarantee processing in case of failure can be implemented while maintaining normal access performance.

  In addition, since the checkpoint processing time is shortened, the checkpoint frequency can be increased, and the data amount of journal information, which is update history information between checkpoints, can be reduced. As a result, it is possible to shorten the follow-up processing time of the update transaction required at the time of recovery, and it is possible to shorten the service stop time until the system recovery.

  As a result, stable access performance of the database application can be provided to general users, and a short service stop can be required for a failure.

  FIG. 1 shows an on-memory database system configuration diagram of the present invention, and FIG. 2 shows a checkpoint data acquisition flow in the system.

  As shown in FIG. 1, the on-memory database system of the present invention includes a “database update processing function unit 100”, an “update segment management function unit 200”, a “system area / database area 300”, a “subsegment update flag 400”, “Checkpoint data acquisition area extraction function unit 500”, “Differential subsegment data group 600”, “Checkpoint data output function unit 700”, “JNL output function unit 800”, “Checkpoint data 900”, “JNL data 1000” ”,“ Recovery function unit 1100 ”,“ in-memory data area 1200 ”, and“ magnetic disk device 1300 ”.

  The database update processing function unit 100 controls a data update request for database data.

  The update sub-segment management function unit 200 manages the status of the data segment to which changes are made when data is updated.

  The checkpoint data acquisition area extraction function unit 500 extracts a data segment group necessary for backup from the update segment information from the in-memory data area 1200.

  The checkpoint data output function unit 700 outputs the extracted data segment group to the magnetic disk device 1300.

  The JNL output function unit 800 outputs the update history generated during the checkpoint acquisition process to the magnetic disk device 1300.

  The recovery function unit 1100 restores the data from the checkpoint data 900 and JNL data 1000 stored in the magnetic disk device 1300 to the memory during system operation.

  As shown in FIG. 2, in this on-memory database system, the system area / database area 300 in the memory is configured based on a unit of a fixed size called a segment.

  In the on-memory database system, when the database update processing function unit 100 receives external data update processing, the data segment to be updated is determined by index information or the like, the data update processing is performed, and the update sub-segment management function The unit 200 changes the flag of the corresponding segment on the segment update flag 400 to the “updated” state.

  In parallel, the database update processing function unit 100 transmits the JNL information, which is update history information, to the JNL acquisition function unit 800 so as to output it as JNL data 1000 on the magnetic disk device 1300.

  When a certain amount of time or when the JNL data 1000 exceeds a certain amount, the checkpoint data acquisition area extraction function unit 500 extracts the segment in the “updated” flag state by the subsegment update flag, The sub-segment data group 600 is determined, and the checkpoint data output function unit 700 outputs the checkpoint data 900 to the magnetic disk device 1300 at the checkpoint.

  When the system is started, the recovery function unit 1100 is called, and the checkpoint data 900 is expanded in the in-memory data area 1200, and the transaction follow-up process is performed on the system area / database area 300 using the JNL data 1000. Then, reproduce the transaction that was confirmed immediately before the failure occurred.

  In this way, the processing of the on-memory database system is roughly divided into backup processing and recovery processing. Hereinafter, the backup process and the recovery process will be described in detail with reference to FIGS.

(Backup process)
The database update processing function unit 100 receives an external database update processing request (S31).

  The database update processing function unit 100 analyzes the data position to be rewritten by the database update request based on the information in the in-memory data area 1200, and rewrites the system area / database area 300 in the in-memory data area 1200. In parallel, the position of the corresponding in-memory data area 1200 is sent to the update sub-segment management function unit 200, and JNL information as update history information is sent to the JNL acquisition function unit 800 (S32).

  The updated sub-segment management function unit 200 sets the sub-segment update flag 400, which is a management flag at the position of the given in-memory data area 1200, to the “updated” state (S33).

  The JNL acquisition function unit 800 outputs the given update history information as the JNL data 1000 to the magnetic disk device 1300 (S34).

  When the amount of JNL accumulated from a certain time interval or the timing of the previous checkpoint exceeds a certain amount, the checkpoint data acquisition area extraction function unit 500 refers to the subsegment update flag 400 to check the previous checkpoint. The update difference information from the point time is acquired and sent to the checkpoint data output function unit 700, and the subsegment update flag 400 is returned to the initial state at that timing (S35).

  The checkpoint data output function unit 700 outputs the received update difference information as checkpoint data 900 to the magnetic disk device 1300 at the time of the checkpoint (S36). When outputting the same sub-segment information as the update difference information secured at the previous checkpoint, the previously acquired data is overwritten. Also, a new file is created for JNL data at the timing when the above processing is completed, and the JNL output associated with the subsequent update processing is written.

  Steps S31 to S36 are repeated from the system startup to the end. Note that the system cannot be stopped during the processing from S35 to S36.

(Recovery process)
The recovery function unit 1100 is called into the memory area secured when the system is started, and the checkpoint data recorded in the magnetic disk device 1300 is loaded into the in-memory data area 1200 (S41).

  After S41 ends, the recovery function unit 1100 performs update follow-up processing on the memory using the JNL data recorded in the magnetic disk device 1300 (S42).

  As soon as S42 is completed, the database update processing function unit 100 is released to the outside and the service is started (S43).

  Examples of application of the on-memory database system of the present invention include the following.

  In services such as telephone exchange processing, one database process (transaction) is extremely small, and instead, the processing series per unit time is extremely large. In addition, due to the nature of the service, the system cannot be stopped for maintenance operations such as backup, and all maintenance operations must be performed online, so backup must also be performed in an online background. .

  When the on-memory database system of the present invention is applied to such a system, it is an on-memory database system in response to a service request such as a short transaction and a high throughput. By providing the online backup processing according to, it becomes possible to provide maintenance operation without service interruption.

  Moreover, the following examples can be considered as application examples using the on-memory database system of the present invention.

  In the exchange process, various types of data are managed, from telephone user registration information to status information such as busy management. Especially, status information is frequently updated data, and JNL information accompanying the update is also considerable. The amount of output. In recovery processing, after loading the latest generation checkpoint data into memory, the transaction processing is followed by JNL information. By applying this server system, the checkpoint interval is shortened by reducing the checkpoint processing amount. The amount of JNL information used in the recovery process is suppressed, and recovery in a short time is possible.

  The on-memory database system of the present invention is realized by a CPU included in a computer device constituting the on-memory database system, for example, and requires a required database update processing function, update subsegment management function, checkpoint data acquisition area extraction function In addition, a checkpoint data output function, a JNL output function, a recovery function, and the like can be installed as application programs.

  The present invention also provides a recording medium in which a program code of software for realizing the functions of the above-described embodiments is recorded to a system or apparatus, and a program in which the CPU (MPU) of the system or apparatus is stored in the recording medium. This can also be realized by reading and executing the code. In that case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and a storage medium storing the program code, for example, a CD-ROM, a DVD-ROM, a CD-R, CD-RW, MO, HDD, etc. constitute the present invention.

The on-memory database system block diagram of this invention. The figure which shows the checkpoint data flow in the on-memory database system of this invention. Backup processing flowchart. Recovery processing flowchart.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Database update process function part 200 ... Update subsegment management function part 300 ... System area | region / database area 400 ... Subsegment update flag 500 ... Checkpoint data acquisition area extraction function part 600 ... Difference subsegment data group 700 ... Checkpoint data Output function unit 800 ... JNL output function unit 900 ... Check point data 1000 ... JNL data 1100 ... Recovery function unit 1200 ... Data area in memory 1300 ... Magnetic disk device

Claims (3)

A data restoration method in an on-memory database system,
The on-memory database system includes:
Manage the on-memory database data area by dividing it into multiple segments,
A database update processing function unit for updating data transmits journal information which is data update history information to the disk;
A checkpoint data acquisition area extraction function unit extracts segment information whose data has been updated, and transmits the segment information to the disk at the time of a checkpoint;
The recovery function unit has a step of performing a follow-up process for updating the checkpoint data stored in the disk based on journal data after the immediately preceding checkpoint stored in the disk at the time of system startup. To restore data. A data restoration method in an on-memory database system,
The on-memory database system includes:
Manage the on-memory database data area by dividing it into multiple segments,
A database update processing function unit receives a data update request, determines a data segment on the memory to be updated, and rewrites the data; and
A journal information output function unit that receives journal information that is history information of the data update from the database update processing function unit and outputs the journal information to a disk; and
An update subsegment management function unit receives the update target data segment information from the database update processing function unit, and sets a corresponding subsegment update flag to an updated state, a subsegment update step;
An extraction step in which the checkpoint data acquisition region extraction function unit extracts the segment information corresponding to the subsegment update flag in the updated state, and determines the checkpoint data;
A checkpoint data output unit that outputs the checkpoint data to the disc at the time of a checkpoint;
The recovery function unit restores the checkpoint data recorded on the disk and journal data after the previous checkpoint to the memory area at the time of system startup, and performs a follow-up process for updating the checkpoint data based on the journal data. A data restoration method comprising: performing a restoration step.   A program characterized in that the data restoration method according to claim 1 or 2 is described as a computer program and can be executed.

Images