KR20020030223A - Logging and recovery method for supporting high performance transaction in main memory resident database system - Google Patents

Abstract

The present invention relates to a logging and recovery method for supporting high performance transactions in a main memory database system. The present invention provides a transaction completion step of storing a tuple log record in a log file 34 of a disk 39 by generating only a log record for a tuple without generating a log record for an index 35 when performing a transaction; When a transaction is canceled, the tuple recovers using the log record of the tuple, and the index 35 restores the index value to the state before the start of the transaction by using the tuple log record according to the operation that generated the tuple log record. Withdrawing the transaction to maintain the consistency of the index 35; And a system restart step of restoring the data 33 of the database 37 as a log record for the tuple at the system restart to maintain the consistency of the tuple, and then rebuilding the index to maintain the consistency of the entire database. It is done. Therefore, according to the present invention, it is possible to perform fast transaction processing by reducing the disk 39 input and output in the main memory database system, it is possible to maintain the consistency of the database 37 when restarting the system due to the transaction withdrawal and malfunction of the database system have.

Description

Logging and recovery method for supporting high performance transaction in main memory resident database system}

The present invention relates to a method for logging and recovering a database system. In particular, a main memory device capable of performing a fast transaction processing by reducing disk I / O, and maintaining a database consistency when a system is restarted due to a transaction withdrawal and a malfunction of the database system. A logging and recovery method for supporting high performance transactions in a database system.

In general, a disk-based general-purpose database system records changes in a table on disk during transaction processing of insert / delete / update operations of a tuple, which is a record. In order to efficiently search for each tuple, index data is kept on disk. When the system is restarted due to transaction withdrawal and system malfunction, such a disk-based general-purpose database system stores log records of changes to tables and indexes in a safe storage space in order to maintain database consistency.

However, since the main memory resident database system resides all the tables in the main memory to record the changes of the table due to transaction processing in the main memory, it is necessary to store log records on disk to maintain the consistency of the database. There is a big overhead in the resident database system, which causes slow transaction processing.

SUMMARY OF THE INVENTION The present invention has been proposed to solve the problems of the prior art, and an object of the present invention is to reduce the amount of disk input and output, which takes up the largest portion of the processing time, for fast transaction processing in the main memory database system, and at the same time reduce transaction and database systems. This is to provide logging and recovery method to support high performance transaction in main memory database system that can maintain database consistency when restarting the system due to malfunction.

1 is a transaction processing flow diagram in accordance with the present invention.

2 is a flow chart upon system restart in accordance with the present invention.

3 is a transaction completion block diagram according to the present invention.

4 is a transaction withdrawal block diagram in accordance with the present invention.

5 is a system restart block diagram according to the present invention.

Explanation of symbols on the main parts of the drawings

30: Tuple to be inserted 31: Tuple to be deleted

32: tuple to be changed 33: data

34: log file 35: index

36: database backup file 37: database

38: main memory 39: disk

In order to achieve the above object, the present invention provides a logging and recovery for performing transaction processing in a database system having a main memory device for storing a database containing data and indexes, and a disk for storing log files and database backup files A method, comprising: a transaction completion step of storing a tuple log record in a log file on a disk by generating only a log record for a tuple without generating a log record for an index when a transaction is performed; When a transaction is canceled, the tuple recovers using the log record of the tuple, and the index restores the index value to the state before the start of the transaction by using the tuple log record according to the operation that generated the tuple log record. Withdrawing the transaction to maintain; And a system restart step of restoring the data of the database as log records for the tuple at system restart to maintain the consistency of the tuple, and then rebuilding the index to maintain the consistency of the entire database.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flow diagram of a transaction processing in accordance with the present invention.

At the start of the transaction, a database tuple of the main memory 38 is changed (step S2) by inserting / delete / update a tuple of the table of the database 37 (step S1). Create a pre- and post-log record (step S3) and change the index 35 (step S4). The transaction end is checked (step S5), if not, the process returns to step S1, and if it is terminated, it is checked whether the transaction termination situation is complete or abort (step S6).

If the transaction end situation is complete, a log record is recorded in the log file 34 of the disk 39 (step S14).

Main Memory Resident Database Management System (DBMS) maintains all data in main memory (memory) so that transaction processing can be processed only by memory operation, so it supports fast real-time transaction processing. However, in order to maintain the consistency of the database 37, disk 39 I / O, which stores log records on the disk 39, is generated, and the more such disk 39 I / O is, the lower the performance of the entire database system is. Will be. Therefore, in order to support high performance transaction processing, the amount of disk 39 input / output should be reduced by reducing the amount of log records.

3 is a transaction completion block diagram according to the present invention. As shown in Fig. 3, when the transaction is completed, only the pre and post log records for the tuples are stored in the log file 34 of the disk 39, and the log records for the index 35 are the log file 34. Do not save on. Therefore, it is possible to reduce the amount of disk 39 input and output by reducing the total amount of log records to support high performance transaction processing.

In the step S6, if the transaction termination situation is withdrawn, it is checked whether the log record to be applied is a transaction start log record (step S7), and if the transaction start log record, the log records are written to the log file 34 of the disk 39 (step S7). S14), or restore the tuple of the database 37 to a state before the start of the transaction by using the log record (step S8). Then, the type of the operation that generated the tuple log record is checked (step S9). If the delete operation is performed, the index insert operation is performed using the pre-log record (step S10). If the insert operation is performed, the index is deleted using the post log record. The index of the database 37 by performing an index delete operation using a post log record (step S12) and performing an index insert operation using a pre log record (step S13). Restore (35) to the state before the start of the transaction.

In step S8 to step S13, the tuples and indexes 35 of the database 37 are restored to the state before the start of the transaction, and the process returns to the step S7.

4 is a transaction withdrawal block diagram according to the present invention. As shown in FIG. 4, both the tuple and the index can be recovered using the log record for the tuple. Therefore, it is possible to reduce the amount of disk 39 input and output by reducing the total amount of log records to support high performance transaction processing.

2 is a flow chart upon system restart in accordance with the present invention.

When the system is restarted due to a system malfunction, the database 37 is loaded into the main memory 38 from the database backup file 36 stored in the disk 39 (step S17), and the log file of the disk 39 is stored. The data 33 of the database 37 is restored to a log record for the tuple stored in 34 (step S18). At this time, the log record for the completed transaction is applied to the database 37 of the main memory 38, but the log record for the transaction that is not withdrawn and the operation is not applied.

It is checked whether there is an index 35 to be rebuilt (step S19), and if not, restarting is finished. If so, the index 35 for the table is rebuilt (step S20).

5 is a system restart block diagram according to the present invention.

As shown in FIG. 5, in the conventional database, the database is restored by loading the index and the tuple together into the main memory, but in the present invention, only the log records for the tuple stored in the log file 34 of the disk 39 are used. The data 33 of the database 37 can be recovered. Therefore, the amount of disk 39 input / output can be reduced, so that the database system restart can be performed quickly.

In this way, tuple log records can be used for transaction completion, transaction withdrawal, and system restart to maintain consistency for tuples and indexes, and reduce the amount of disk 39 I / O that stores log records in the main memory database system. It can improve the performance of the entire database system to provide high performance real time transaction processing services.

As described above, according to the logging and recovery method for supporting a high-performance transaction in the main memory database system of the present invention, it is possible to perform fast transaction processing by reducing the disk 39 input and output in the main memory database system, transaction withdrawal and When the system is restarted due to a malfunction of the database system, there is an effect of maintaining the consistency of the database 37.

Claims (5)

A database system having a main memory 38 for storing a database 37 including data 33 and an index 35 and a disk 39 for storing a log file 34 and a database backup file 36. In the logging and recovery method for performing transaction processing in A transaction completion step of generating a log record for the tuple without storing a log record for the index 35 when the transaction is performed, and storing the tuple log record in the log file 34 of the disk 39; When a transaction is canceled, the tuple recovers using the log record of the tuple, and the index 35 restores the index value to the state before the start of the transaction by using the tuple log record according to the operation that generated the tuple log record. Withdrawing the transaction to maintain the consistency of the index 35; And And a system restart step of restoring the data 33 of the database 37 as a log record for the tuple upon system restart to maintain the consistency of the tuple, and then rebuilding the index to maintain the consistency of the entire database. Logging and recovery methods to support high performance transactions in main memory database systems. The method of claim 1, wherein the transaction completion step, Inserting / deleting / modifying a tuple of the database 37 table at the start of a transaction (S1); Changing the database 37 of the main memory 38 (S2); Generating a pre- and post-log record for the tuple (S3); Changing the index 35 (S4); Checking a transaction termination (S5); If the transaction has not ended, returning to step S1 and checking if the transaction ended is complete or withdrawn (S6); And Logging and recovery to support high performance transactions in the main memory database system, characterized in that the step of recording the tuple log record in the log file 34 of the disk 39 when the transaction termination situation is complete (S14). Way. The method of claim 2, wherein the transaction withdrawal step, Checking whether the log record to be applied is a transaction start log record (S7) when the transaction termination situation is withdrawn in step S6; If it is a transaction start log record, log records are written to the log file 34 of disk 39 (step S14), and if it is not a transaction start log record, the tuples of the database 37 are restored to the state before the transaction start using the log record. Step S8; By checking the type of the operation that generated the tuple log record (S9), if the type of the operation that generated the tuple log record is a delete operation, the index insert operation (S10), if the insert operation (S11), In the case of a change operation, after performing an index delete operation (S12) and then performing an index insert operation (S13), the main memory database is configured to recover the index 35 of the database 37 to a state before the start of a transaction. Logging and recovery methods to support high performance transactions on the system. The method of claim 1, wherein the system restarting step, Loading the database 37 into the main memory 38 from the database backup file 36 stored in the disk 39 (S17); Recovering the data 33 of the database 37 to log records for the tuples stored in the log file 34 of the disk 39 (S18); Checking whether there is an index to be rebuilt (S19); And If there is no index to be rebuilt, the restart is terminated, and if there is a logging and recovery method for supporting high performance transactions in the main memory database system comprising the step (S20) of rebuilding the index (35) for the table. A database system having a main memory 38 for storing a database 37 including data 33 and an index 35 and a disk 39 for storing a log file 34 and a database backup file 36. To A computer-readable record that records a program on the computer that can reduce the amount of disk I / O that the log record is stored on while maintaining consistency with the database by using tuple log records at transaction completion, transaction withdrawal, and system restart. media.