KR20050054380A - Database management system and the method of controlling consistency of system table in dbms - Google Patents

Abstract

The present invention relates to a database management system for concurrency control of system tables at the query processor level, and a method of concurrency control for system tables in the system.

The database management system of the present invention includes a disk on which system tables and user tables are physically stored; A storage manager including a buffer management module configured to load a system table and a user table into main memory to perform buffer management, and a first lock management module to support consistency and concurrency of the user table; And a system table management module that converts and manages system table information in a disk structure form from a storage manager into a memory structure and provides corresponding system table information according to the requirements of each application transaction, and the concurrency and consistency of system table information. It consists of; a query processor including a second lock management module to support.

Description

Database management system and the method of controlling consistency of system table in DBMS}

The present invention relates to a method for controlling concurrency of system tables in a database system. More specifically, the present invention relates to a database management system for concurrency control of a system table with a high access frequency and transaction sharing and a constant access type among data types at the query processor level. It relates to a method of controlling concurrency on system tables in the system.

In general, a database management system stores a buffer management for data sharing and a lock management for concurrency control at the storage manager level when multiple users (transactions) access the same data (table or record) at the same time. Support.

1 is a structural example of a conventional database management system in which concurrency control for all tables is processed in a storage manager while multiple transactions are performed.

Referring to FIG. 1, the database management system includes a query processor 110 for generating a transaction, a storage manager 120 including a buffer management module 122 and a lock management module 124, and a system table 132. It consists of a disk 130 including a user table 134 to process the requests of the application programs (101-1 ~ 101-n).

The applications 101-1 to 101-n generate the transactions 112-1 to 112-n for the query processing requested by the user, and the transactions 112-1 to 112-n are the users requested by the user. To access the table 134, first access the system table 132 to obtain information for accessing the user table. The procedure of accessing the system table first obtains the locking authority from the lock management module 124 of the storage manager, and then stores the system table 132 in a space managed by the buffer management module 122 (hereinafter referred to as a buffer space). Find out. If the system table 132 is in buffer space, the contents of the disk structure are changed to the memory structure format and stored in the memory structure format so that the contents can be used by the query processor 110, and the user table information that is ultimately desired to be read and read. Get If the system table 132 to be found is not in the buffer space, the system table 132 is read from the disk 130 into the buffer space, and then the above-described execution order is followed.

Therefore, system tables have a higher frequency of access than user tables, and most of the properties and read operations are shared by most transactions. However, since the buffer management module 122 applies the buffer replacement algorithm equally to the system table and the user table, the buffer management module 122 may not be present in the buffer space even when searching the system table, and thus a relatively large portion of the time spent on disk I / O. There is a problem that is very high.

In addition, even though the system table sharing of the transactions (112-1 ~ 112-n) is very high, each transaction (112-1) as the system table information is not shared at the query processor (110) level managed for each transaction 112-n) has a problem in that even when accessing the same system table, the disk format structure is converted into the memory structure format at the same time as the transaction processor is repeatedly imported to the query processor 110 for each transaction.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to apply a concurrency control scheme in a query processor, that is, in a memory structure format itself, to system tables so that transactions can generate system table information at the query processor level. The present invention provides a database management system that improves concurrency performance by sharing, and reduces the number of operations for converting a system table into a memory structure format and a method of controlling concurrency on system tables in the system.

In addition, another object of the present invention is to ensure that the system table information once read is always stored until the database management system is down, so that disk I / O occurs only once, so that the database management system and system tables in the system can be improved. It is to provide a concurrency control method for.

In order to achieve the above object, a database management system of the present invention includes a disk on which system tables and user tables are physically stored; A storage manager including a buffer management module configured to load a system table and a user table from the disk into main memory to perform buffer management, and a first lock management module to support consistency and concurrency of the user table; And a system table management module which converts and manages system table information in a disk structure form from the storage manager into a memory structure and provides corresponding system table information according to a request of each application transaction, and concurrency and consistency of system table information. And a query processor that responds to a query of each application program, including a second lock management module supporting the application.

In addition, in order to achieve the above object, the concurrency control method for a system table in the database system of the present invention determines a hash entry through a user table identifier value and obtains a latch of the entry to simultaneously access other transactions. Preventing; As the record is retrieved from the entry, acquiring read lock permission and then obtaining system table information from the record; And reading the necessary user table information from the disk according to the obtained system table information, and returning the entry latch. The system table information of the query processor level for a transaction request is searched.

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

2 is a block diagram of a database management system to which a concurrency control method for a system table in a query processor according to the present invention is applied.

Referring to FIG. 2, the database management system includes a lock management module 214 and a system table management module 216, a query processor 210 for generating transactions 212-1 to 212-n, and a buffer management module. 224 and a storage manager 220 including a lock management module 222, and a disk 230 for storing the system table 232 and the user table 234 application programs (201-2 ~ 101) -n) process the request (query).

That is, there are various applications 201-1 to 201-n using the database management system, and query executions requested from them, that is, transactions 212-1 to 212-n are separately determined according to each application. exist. The multiple transactions 212-1 to 212-n obtain system table information necessary from the system table management module 216 in the process of executing a query. The system table management module 216 converts system table information in a disk structure format obtained through the storage management buffer management module 224 into a memory structure, which is shared by multiple transactions 212-1 through 212-n. Consistency and concurrency are supported through the lock management module 214 of the query handler. The buffer manager 224 of the storage manager loads a system table and general data (user table) in a disk structure into main memory to perform buffer management. The lock management module 222 of the storage manager part stores information of a general data file. Only responsible for maintaining consistency and concurrency. The disk 230 physically stores a system table 232 and a general data file 234. The information once loaded into the system table management module 216 is maintained in the system table management module 216 unless the corresponding user file is deleted.

3 is an example of a data structure managed by a system table management module according to the present invention. The system table management module 216 manages all the system tables 232 by type. The table type illustrated in FIG. 3 is only for the system table 310 indicating the table information and the system tables 320 indicating the index information. For example. Each entry in the system table list has latches 311 and 312, and the system table nodes (records) currently in use are linked.

Referring to FIG. 3, the system table 310 representing table information includes system table records linked to a list with the same hash values as the system table list managed by the hash table. Reference numeral 313 is a system table record that stores information about the user table named "employee". Records 314 having the same hash value operate on a linked list.

When the database management system is first started, it creates only system table lists for system table management, the number of entries in the system table list by the number of records (m) in each system table, and the hash function adds m to the user table identifier value. The remaining divided value was used as a hash value to prevent collisions as much as possible. Reference numeral 320 denotes an initialization state at system startup when the number of records in the index information table is m.

When multiple system searches, deletes, and inserts a system table record into an entry in the system table list at the same time; In order to obtain or change the actual record information itself, a lock function such as 315 is used to support consistency and concurrency.

4 is a system table search procedure for finding a record of a corresponding system table when information about a user table is in a database according to the present invention.

First, when a request for information about a system table comes from a transaction, a hash entry is determined using a hash function as a user table identifier value (401, 402). If a hash entry is determined, a latch (311, 312) of the entry is obtained, thereby preventing other transactions from accessing this entry at the same time (403). After obtaining the latch for the entry, the system table record associated with the entry is searched (404).

If a corresponding system table record is found as a result of the search, a read lock right is acquired to acquire table information (405). This is described in detail in FIG. 7. After the acquisition is completed, the information to be read is read and the corresponding entry latches 311 and 312 are returned (409 and 410).

If a record does not exist in the corresponding hash entry, a new node for storing the record content is created and added to the entry (406). After acquiring a read lock to acquire system table information, the system manager receives a system table record from the storage manager 220 (step 407). In this case, the record information of the disk structure format is converted into the memory structure format used by the query processor 210 and reflected to the new node (408). After reading all the necessary information, the entry latches 311 and 312 are returned (409 and 410).

5, when deleting a user table, system table information related to the user table should also be changed, which is an operation procedure of the system table manager.

Referring to FIG. 5, in order to know if a system table record related to a user table is in the system table management list 310, a hash entry is first selected and a latch is obtained (501, 502).

After the system table record is searched in the corresponding entry, if there is no system table record, it means that there is no transaction being accessed. Therefore, the storage manager 220 immediately deletes the corresponding record portion (503, 506).

If there is a record in the entry, another transaction may be currently in use. Thus, after obtaining the write lock right (504), the record node is deleted (505). After deleting the record node, the storage manager 220 deletes the record (506) and returns the latch of the entry (507). The record node deletion procedure will be described in detail with reference to FIG. 7.

6 illustrates an operation procedure of the system table manager when a user table is newly created according to the present invention. To know if a record associated with the user table is in the system table management list 310, first select a hash entry and obtain a latch (601, 602). It then searches for the system table record in that entry and, if there is a record, returns the acquired latch, returns it as an error, and ends because the same table already exists (603-605).

If there is no record by searching the system table record in the corresponding entry, the storage manager 220 also changes the user table creation and related system table information, and then places the node in the system table list 310 managed by the query processor 210. Add and store relevant system table information (606,607). The lock is then acquired and the latch of the entry is returned (608.609). The acquired lock authority is returned at the end of the transaction. Changing the user table structure is a similar concept.

7 is a flowchart illustrating a processing procedure for obtaining a lock according to the present invention.

Referring to FIG. 7, when a table is found through a system table list search while a latch of a hash entry is acquired, the table can be shared by comparing a read / write lock mode currently held in a table node with a lock mode that the current transaction is to perform. It is determined whether or not (701 ~ 703).

The lock mode is divided into 'read mode' and 'write mode', and the read modes can be shared, but the write mode cannot be shared with any other mode. Even if sharing is possible, check for pending transactions in the node's wait list.

If the share is available and there are no waiting transactions, the lock is obtained by recording the requested lock mode in the node and increasing the value of the currently accessed transaction by one (705).

If a transaction exists in the wait list or sharing is not possible, the transaction identifier value is linked to the wait list (703), and the latch of the entry is returned (704). The transaction is blocked.

When the transaction that has acquired the lock is finished, if the first transaction in the waiting list waits in the write mode, only the first transaction is awakened, and if it waits in the read mode, it wakes up all of the read mode among the waiting transactions and acquires the lock again. Follow the procedure.

As described above, the present invention has the following effects by allowing multiple transactions to access the system table at the query processor level of the database management system.

First, since system tables are accessed in a constant format, performance can be improved by managing only the record level to manage both the file level and the record level.

Second, in general, since system table sharing is supported in a buffer space, table information to be always found may not be in memory. However, in the present invention, table information read once is always managed in memory, thereby improving search performance.

Third, in the conventional method, the system table information that is first accessed per transaction must be converted every time from the disk structure format to the memory structure format. However, in the present invention, the system table information is converted only once in the database management system. This can speed up system table information access.

1 is a diagram illustrating a structure of a conventional database management system in which concurrency control for all tables is processed in a storage manager while multiple transactions are performed.

2 is a diagram illustrating a structure of a database management system managed by a query processor only for system tables in a process of performing multiple transactions according to the present invention.

3 is an example of a data structure required for managing system table information provided by the present invention.

4 is a flowchart illustrating a process of searching for a specific record in the system table management module provided by the present invention.

5 is a flowchart illustrating a process of deleting one record of a system table due to deletion of a user file in a system table management data structure according to the present invention.

6 is a flowchart illustrating a process of inserting a record of a system table due to user file creation in a system table management data structure according to the present invention.

7 is a flow chart illustrating a process of obtaining a lock on a selected record in accordance with the present invention.

<Explanation of symbols for the main parts of the drawings>

201-1 through 201-n: Application 210: Query Handler

212-1 ~ 212-n: Transaction 214: Lock management module

216: system table management module 220: storage manager

222: lock management module 224: buffer management module

230: Disk 232: System Table

234: user table

Claims (17)

A disk on which system tables and user tables are physically stored; A storage manager including a buffer management module configured to load a system table and a user table from the disk into main memory to perform buffer management, and a first lock management module to support consistency and concurrency of the user table; And A system table management module for converting and managing system table information in a disk structure form from the storage manager into a memory structure and providing corresponding system table information according to a request of each application transaction, and concurrency and consistency of system table information. And a query processor responding to a query of each application program, including a second lock management module supporting the database. The method of claim 1, wherein the system table management module, Once the system table information has been loaded into main memory, it is maintained in main memory unless the system is down or the corresponding user table is deleted. The method of claim 1, wherein the system table management module, A system structure for managing system table information, comprising: a system table list for each type of system table and a plurality of entries having latches in each list, and then managing the system table information by linking records having the same hash value to each list. Database management system. The system table management module of claim 3, wherein The number of entries in the system table list is made by the number of records belonging to each system table, and the hash function uses the remaining value obtained by dividing the number of records by the user table identifier value as a hash value. The system table management module of claim 3 or 4, wherein when a plurality of transactions simultaneously perform a search, delete, and insert operation on an entry of a system table list, A database management system that uses a latch of an entry to retrieve a corresponding record or to change link information, and uses a locking function to acquire or change the actual record information itself. The system table management module of claim 3 or 4, wherein the system table management module is configured to search for a record of a system table required by an application program transaction. (a) determining a hash entry with a user table identifier value and obtaining a latch of the entry to prevent entry of another transaction; (b) acquiring system table information after acquiring a read lock right through the second lock management module as the corresponding record is found in the entry; And (c) reading necessary user table information through the storage manager according to the acquired system table information, and returning the entry latch. The method of claim 6, wherein the system table management module, if the record does not exist in the determined hash entry in the step (b), Creating a new record node and adding it to the hash entry; And And receiving system table information from the storage manager and converting the system table information into a memory structure format to reflect the read record right to the new record node. The system table management module of claim 3 or 4, wherein the system table management module requests deletion of the user table in the disk from a transaction. Determining a hash entry through the user table identifier value and obtaining a latch of the entry; Searching for the relevant system table record in the entry; If a related system table record is found, acquiring a write lock right through the second lock management module to delete the corresponding record node; And And deleting the corresponding record at the buffer level and the corresponding user table in the disk through the storage manager, and then returning the entry latch. The method of claim 3 or 4, wherein the system table management module creates a new user table on the disk. Determining a hash entry through a user table identifier value and obtaining a latch of the entry; Searching for whether a corresponding system table record exists in the hash entry; If the corresponding system table record exists, returning the entry latch and treating it as an error; And If there is no corresponding system table record, create a user table through the storage manager, change the relevant system table information, and create a new record node in the hash entry to add and store the relevant system table information, and acquire lock permission. And returning the entry latch after performing the corresponding transaction. The method of claim 3 or 4, wherein the second lock management module, Comparing a lock mode held in the target table record node with a lock mode that the current transaction is to perform, and determining whether or not it can be shared, and checking whether a pending transaction exists; If there are no sharable and waiting transactions, writing the requested lock mode to the node, incrementing the number of access transactions by one and acquiring a lock; And If there is a non-sharing or waiting transaction, or if sharing is not possible, include the transaction identifier value in the waiting list, and return the entry latch to block the transaction; system. The method of claim 3 or 4, wherein the second lock management module, When the transaction that has acquired the lock is finished, if the first transaction in the waiting list waits in the write mode, it wakes up only the first transaction, and if it waits in the read mode, it wakes up the read mode among the waiting transactions and resumes the lock acquisition procedure. A database management system, characterized in that to perform. A method of controlling concurrency for a system table requested by application transactions in a database system including a query processor, a storage manager, and a disk, the method comprising: (a) determining a hash entry through a user table identifier value and acquiring a latch of the entry to prevent simultaneous access of other transactions; (b) acquiring system table information from the record after acquiring read lock permission as the corresponding record is found in the entry; And (c) reading user table information required from the disk according to the obtained system table information, and returning the entry latch; and searching system table information of a query processor level for a transaction request by performing To control concurrency for system tables in the database system. The method of claim 12, wherein the concurrency control method further comprises: when the record does not exist in the hash entry in step (b), Creating a new record node and adding it to the hash entry; And Acquiring a read lock right, receiving system table information from the storage manager, converting the system table information into a memory structure format, and reflecting the system table information to the new record node; How to control concurrency on system tables in the system. The method of claim 12, wherein the concurrency control method comprises: when there is a delete request for a user table from a transaction, Determining a hash entry through the user table identifier value and obtaining a latch of the entry; Searching for the relevant system table record in the entry; If a related system table record is found, acquiring write lock permission to delete the corresponding record node; And And deleting the corresponding record at the buffer level and the corresponding user table in the disk through the storage manager, and then returning the entry latch. 2. The method of claim 12, wherein the concurrency control method is to create a new user table on disk. Determining a hash entry through a user table identifier value and obtaining a latch of the entry; Searching for whether a corresponding system table record exists in the hash entry; If the corresponding system table record exists, returning the entry latch and treating it as an error; And If there is no corresponding system table record, create a user table through the storage manager, change the relevant system table information, and create a new record node in the hash entry to add and store the relevant system table information, and acquire lock permission. And returning the entry latch after performing the corresponding transaction. The method of claim 12, wherein the processing for obtaining the lock comprises: Comparing a lock mode held in the target table record node with a lock mode that the current transaction is to perform, and determining whether or not it can be shared, and checking whether a pending transaction exists; If there are no sharable and waiting transactions, writing the requested lock mode to the node, incrementing the number of access transactions by one and acquiring a lock; And If there is a non-sharing or waiting transaction, or if sharing is not possible, including the transaction identifier value in the waiting list, and returning the entry latch to block the corresponding transaction; To control concurrency for system tables in the database. The method of claim 12, wherein the processing for obtaining the lock comprises: When the transaction that has acquired the lock is finished, if the first transaction in the waiting list waits in the write mode, it wakes up only the first transaction, and if it waits in the read mode, it wakes up the read mode among the waiting transactions and resumes the lock acquisition procedure. A concurrency control method for system tables in a database system, characterized in that for performing.