JP2002049519A - Device and method for information processing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To actualize processing wherein the consistency of data is secured without being aware of data properties when many kinds of small-amount data are controlled by using a common memory. SOLUTION: When the resource management of the common memory is performed, properties of a table 5 and initial values of records are defined with user indexes as to resources (table 5 and its records) that an application 11 uses and held as a library (user index table 4) which can specify a user index by the specification of the table number of a resource and a record number. At a request to use the resource whose table number and record number are specified, the resource is accessed according to the user index held in the library.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an information processing apparatus and method, and more particularly to an information processing apparatus and method for managing resources of a shared memory used for transaction processing in a computer system.

[0002]

2. Description of the Related Art In information processing using a computer system, client-server systems have become popular, and the performance of UNIX (registered trademark) has been improved.
A transaction processing system using enOLTP (Open / On-Line Transaction Processing) has been developed.

[0003] A database management system is usually used as a resource management system for managing and controlling a large amount of user data using a device on which UNIX operates as a server. Since the data processing in this case involves a disk access, there is a problem that the processing efficiency is poor in a transaction processing system that handles a small amount of various data.

[0004] As existing techniques for solving the above problems,
Memory file creation using a shared memory is known. This is a technique for managing resources on a semiconductor memory whose data access speed is much higher than that of a disk system. However, since processing is performed on semiconductor memory, it is difficult to manage simultaneous access compared to a database management system. To enable access from a transaction processing program that performs parallel processing, data consistency must be ensured. There must be.

[0005]

SUMMARY OF THE INVENTION It is an object of the present invention to realize a process in which data consistency is ensured without being conscious of data attributes in data control of a small quantity and many kinds using a shared memory. .

Further, in order to realize high-speed access,
Another object of the present invention is to efficiently manage and recover resources on a shared memory.

[0007]

The present invention has the following configuration as one means for achieving the above object.

An information processing apparatus according to the present invention is an information processing apparatus for managing resources of a shared memory, and defines an attribute of a table and an initial value of a record by a parameter for a table and a record which are resources used by an application. Means for holding the defined parameters as a library that can be specified by specifying the table number and record number of the resource; and Access means for accessing resources according to the parameters held in the library, wherein the library is provided for each application sharing a predetermined resource.

In an information processing apparatus for simultaneously controlling resources having different attributes, for resource shared by a plurality of applications, resource information including attribute information indicating whether the resource is a non-recoverable type or a recoverable type is provided. A memory to be recorded, a resource management table that enables identification of resource information recorded in the memory by a pointer, a transaction for the resource, and a transaction for the resource by the application and a specification of the record lock information by the transaction. , Management means enabled by identification information,
In response to the resource use request of the application, the resource information is specified by using the pointer and the resource management table, and a check unit for checking the attribute of the requested resource, based on a check result of the check unit,
Access means for accessing the requested resource, wherein the access means reads or updates the resource when the inspection result indicates the non-recoverable attribute, and indicates the recoverable attribute. Is characterized by retrieving update data for the record from the record lock information specified by the identification information, and reading or updating the retrieved update data.

An information processing apparatus for backing up update data in the shared memory during online processing, a sampling means for periodically collecting a dump of the shared memory and outputting the collected dump to a recording device; Recording means for recording information on the updated record in the resource, and information indicating the dump collection time and the resource update time in a resource management table, wherein the collection means is recorded in the resource management table. Compare the dump collection time and the resource update time
When the update time of the resource is late, sampling is performed on the updated record based on the information on the updated record.

An information processing method according to the present invention is an information processing method for managing resources of a shared memory, wherein a table attribute and an initial value of a record are defined by parameters for a table and a record which are resources used by an application. Then, the defined parameters are held as a library that can be specified by specifying the table number and record number of the resource, and the parameter stored in the library is used in response to a use request of the resource with the specified table number and record number. , And wherein the library is provided for each application sharing a predetermined resource.

[0012] Further, for a resource shared by a plurality of applications, a resource in which resource information including attribute information indicating whether the resource is a non-recoverable type or a recoverable type is recorded. A resource management table that allows identification of, and manages transactions for the resources,
A management unit that enables identification of the transaction for the resource by the application and the record lock information by the transaction,
An information processing method for simultaneously controlling resources having different attributes, wherein resource information is specified using the pointer and the resource management table in response to a resource use request of an application, and the attribute of the resource requested to be used is specified. Inspecting, and based on the inspection result, each step of accessing the resource requested to be used, and when the inspection result indicates the non-recoverable attribute, the resource is read or updated. When the recovery type attribute is indicated, the update data for the record is searched from the record lock information specified by the identification information, and the searched update data is read or updated.

An information processing method for backing up update data in a shared memory during online processing, wherein a dump of the shared memory is periodically collected, the collected dump is output to a recording device, and the Information relating to the updated record, and each step of recording information indicating a dump collection time and a resource update time in a resource management table, wherein the dump collection step includes a step of recording the dump recorded in the resource management table. The collection time is compared with the update time of the resource, and when the update time of the resource is late, the collection of the updated record is performed based on the information on the updated record.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, information processing according to an embodiment of the present invention will be described in detail with reference to the drawings.

[Overview] FIG. 1 is a diagram for explaining a shared memory resource management system in information processing according to the present embodiment.
2 is a diagram illustrating an example of transaction management, table management, and data table management in a shared memory resource management system, and FIG. 3 is a diagram illustrating backup of updated data.

In the present embodiment, in order to solve the above-mentioned problem, a memory resource manager that creates a small amount of frequently accessed user data in the form of a user table in a shared memory, and maintains and manages the created user data. (MRM: Me
mory Resource Manager).

The data in the user table 5 need to be continuously maintained and managed, the data such as control information has a fixed data content that is not changed from an initial value, and the data needs to be continued even if changed. There are some that do not exist (in other words, the ones that can be initialized each time). So user table 5
A type (recovery type) that enables data recovery in the event of a failure and a type that does not require recovery (non-recovery type) are specified for the data, and the information (parameters) is stored in the user index table 4 for the user table 5. In addition, a connection management table 2 and a transaction management table 3 are provided to make the MRM play a role as a resource manager in the open environment, and the RM (ResourceManager) defined in the X / Open DTP model (distributed transaction processing model) ) Function.

[MRM] MRM needs to be able to handle various kinds of data in transaction processing by simple operations. This variety of data includes immediate update data, delayed update data, two-phase commit (2PC: Two Phase Commi
t) Corresponding data and one-phase commit (1PC: One P
hase Commit) data. That is, the immediate update and the delayed update data can be handled simultaneously in the same process, and the process for handling 2PC data and the process for handling 1PC data can access the same MRM.

The MRM has an MRM system management table 1 at the center of the control table to realize the above functions.
Then, the process ID connected to the MRM is managed as an entry of the connection management table 2. If it is a transaction type, the transaction ID is managed as an entry in the transaction management table 3, and the update data to the recovery type table to be used is managed.

To access the MRM, a simple API (Applicati
on Program Interface) 11 is provided. The API 11 accesses the shared memory table controlled by the MRM concurrently and simultaneously. The recovery type update data is transmitted to the AUDIT update log by the MRM server process 13 after its looks (update data of the record update information table 7 shown in FIG. 2).
Output to file as 15.

Also provided is an MRM cyclic dump process 14 for periodically dumping table data on the shared memory to a file. By this process, the recovery process performed by the dump file 8 and the AUDIT update log 15 is made more efficient. Note that this process can dump the table data to a file without interrupting the online action.

It should be noted that the above processes 13 and 14 are based on the MRM
Monitored and controlled by monitor process 12. Therefore, the user can easily handle the delayed update type and the immediate update type data by calling the provided MRM API 11.

Some problems and problems arising in the development of the above MRM are listed.

[1] Simultaneous control of recovery-type and non-recovery-type data tables A normal database management system is compatible with RM.
It supports the Xa interface, but does not handle non-recoverable data that does not require recovery, and manages all as recoverable data. The MRM to be developed this time needs to control the shared memory in consideration of the fact that recoverable and non-recoverable data are accessed simultaneously and concurrently from one program.

[2] Effective Use of Memory It is necessary to efficiently arrange data tables to be developed on the shared memory. Since the maximum value of the memory size of the shared memory is limited by the kernel parameter of the UNIX OS, the data table must be expanded on the shared memory according to the maximum value.

[3] Lock Control Data is referred to and updated from a plurality of programs simultaneously and concurrently by transaction processing. this is,
This indicates that efficient exclusion control and lock release at the time of abnormal termination of a program are essential. Further, in order to improve the processing efficiency of the MRM and perform multi-threading, it is necessary to perform exclusive control corresponding to multi-threading.

[4] Simple and Efficient Processing From the viewpoint of a user-arbitrary table, it is necessary for a user to be able to independently and easily define data, and to be able to refer to and update data through a simple interface. There is. Furthermore, in order to realize high-speed access, management and recovery only on the shared memory are required.

[Mechanism of Memory Management Table] The entire MRM is controlled by the system management table 1 shown in FIG. 2, and all MRM accesses from the user are performed via the system management table 1.

[1] Management Information of System Management Table 1 The following information is managed by the system management table 1 for each process connected to the MRM. Pointer to connection management table 2 Pointer to transaction management table 3 Pointer to user index table 4

[2] Management Information of Connection Management Table 2 The following information is managed by the connection management table 2 for each process connected to the MRM. The process ID (PID) of the program that called the MRM The thread ID (TID) of the program that called the MRM The ID (RMID) of the RM to be connected, here the ID of the MRM Xa_Infor String (however, only the process of the Xa interface)

The Xa_Infor character string is used to manage a global transaction, perform MRM, and perform commit control for resource managers other than MRM.
Information used when an external management program opens a resource. Generally, a database name and an access mode are specified by character strings. In the present embodiment, only the database name is specified.

[3] Management Information of Transaction Management Table 3 The following information is managed by the transaction management table 3 for each transaction. PID TID Transaction ID (XID) ID of shared memory that stores A / L (After Looks) (ALID) (that is, shared memory ID of record update information table 7) Processing status 2PC (Xa) or 1PC identification information (lock Holds the recovery type table number and record number that were performed) Pointer to lock information table 6

[4] Management Information of User Index Table 4 The following information is managed by the user index table 4 for each table number. Number of records to be stored Record length Table type (recoverable / non-recoverable) ID of shared memory for storing data Offset of lock area Offset of data area

[5] Management Information of Lock Information Table 6 Information of the locked / updated recovery type table is managed by the lock table 6. Table number Record number A / L write pointer

[7] Management Information of Record Update Information Table 7 Update data of the updated recovery type table is managed. Update data length Update data

[Processing Mechanism] Since the MRM is accessed from the transaction processing program, an efficient and robust lock control mechanism is required.

[1] Lock Control Method As a lock control method, Mute is used from the viewpoints of effective use of memory, portability to other systems, multithread support, and the like.
It employs the x Lock mechanism, which is an advantage compared to semaphores.

The Mutex Lock is a mutual exclusion lock, and is a mechanism for preventing the risky area of the code from being simultaneously executed by a plurality of processes or threads when serializing the execution of a process or a thread.
If the lock has already been obtained, the process waits until the lock is released, and if the lock is acquired, processing such as updating is performed to release the lock. Lock resource management is only possible with shared memory, and the mechanism is simple. It can be easily used on multi-threads. Lock processing is faster POSIX (Portable Operating System environment-Intera
ctive eXecutive) compliant and has the same function in WIN32,
Excellent portability

[2] Lock Control Mechanism The minimum unit of lock is a record, and the lock information is managed for each data table. For controlling the update data (A / L), the lock information table 6 linked from the transaction management table 3 is accessed.
The process of record lock and release is performed based on the lock information for each data table.

When a user issues a read request with a lock to the n-th record of a table,
The corresponding record of the user index table 4 is searched through the MRM system management table 1, and the expanded data table (included in the user table 5) is searched by the shared memory ID stored in the table number. Then, the lock state of the designated record number is checked from the lock management area set in the data table.

The Mutex controlled by the OS is in the lock management area.
In addition to the lock area information and Mutex lock information, information for controlling the MRM system includes a locked PID and TID, and a lock flag (LK-F) indicating lock setting / release.

As a check of the lock state, the MRM searches the lock management area and checks whether LK-F is set. If LK-F is set, it is determined to be locked. If not, it is determined to be unlocked.

The reason why the determination is not made based on the Mutex area information and the Mutex lock information of the OS is that if the program (process) setting the lock terminates abnormally without releasing the lock set by itself, the lock information remains. It is because. In other words, in order to avoid the problem that the lock information remains, an LK-F for the MRM to check the lock state is provided.

Upon receiving the lock request, the MRM confirms that the record is not set to the locked state,
Performs a series of processing from tex Lock → LK-F setting → Mutex Unlock. If the lock information remains in this state, the lock is released by using the PI of the connection management table 2 where the internal thread of the MRM is linked from the MRM system management table 1 at regular intervals.
Search D and PID of lock management area. If the process corresponding to the searched PID has been completed, it is determined that the process has ended abnormally, and the LK-F of the record is released. As a result, the record is unlocked and can be accessed from another process.

The lock control is the same for the recovery type and non-recovery type tables.

[3] Recoverable / Non-Recovery Lock Control The MRM of the present embodiment aims to manage control information required for transaction processing. Therefore, delayed update type (recovery type) data and immediate update type (non-recovery type)
Data can be mixedly managed.

The delayed update type data is declared to be updated at the start of a transaction (TRBegin) and is actually updated at the end of a transaction (Commit or Rollback), or the updated data is canceled. Therefore, the updated data will not be updated until the end of the transaction.
Retained as Looks. On the other hand, the immediate update type data is immediately updated at the same time as the data update request, and is treated as a non-recovery type.

Recoverable and non-recoverable data can be handled from the same process. Further, table attributes indicating the recovery type or the non-recovery type are managed in the user index table 4, and the following processing is performed according to the attribute.

Reading Process with Lock The reading process with lock is basically the same for both the recovery type and the non-recovery type.

The table / record to be read is Mutex
Lock and set LK-F, then release MutexLock. The lock management method has the following differences between the recovery type and the non-recovery type.

In the case of the recovery type, the information (PID, table number, record number, etc.) is further entered in the lock information table 6. As a result, in the recovery type, the lock is managed as an update lock, and the lock is held until the transaction is confirmed (Commit or Rollback). If the record has already been updated in the same transaction, the record is read from the record update information table 7 (details will be described later). That is, the updated record is read.

In the case of the non-recovery type, the corresponding record is locked, and the lock is released by a lock release request.

Further, the lock set in the read processing with the lock is automatically released in the update processing of the record.

Update Processing The update processing differs greatly between the recovery type and the non-recovery type.

The non-recovery type is immediately updated, the lock is released, and the user returns. That is, the lock is released by the update API at the same time as the update.

On the other hand, the recovery type is a delayed update, and the update content is stored in the record update information table 7 as update data (After Looks). Record update information table 7
Are managed by the transaction management table 3 in transaction units. The actual resilient update is
Performed when the transaction is finalized. That is,
Since the update is delayed, the update API uses the update data (After Lo
oks), and when the user calls the commit API later, the lock is released at the same time as the actual update.

Release of Lock In the non-recovery type, as described above, the lock is released by a lock release request or a request for updating the same record. That is, the lock information in which the LK-F is set in the user table 5 is searched, and the LK-F is released.

The recovery type lock is released by confirming the transaction. Transaction management table
The lock information table 6 is searched from 3 and the set lock information is released. At this time, the held update data (After Looks) is updated or canceled (discarded) by the transaction confirmation method (Commit or Rollback).

[4] 2PC and 1PC Control The MRM of this embodiment supports the Xa interface. That is, it corresponds to 2PC. It also has a 1PC function that enables control independent of other databases. These functions can be used properly by declaring transaction processing conforming to the Xa interface or original local transaction processing for each process. Since a use declaration is required, these functions must be used separately for each process, and mixed use is not allowed.

2PC The 2PC supported by the MRM of the present embodiment conforms to the Xa interface specification defined by the X / OPEN DTP model, and Xa_o
pen / Xa_close / Xa_start / Xa_end / Xa_rollback / Xa_prepar
Supports each interface of e / Xa_commit / Xa_recover / Xa_forget.

According to the transaction management table 3,
It manages the transaction ID (XID) and thereafter controls the Xa interface from the transaction manager (TM). Write the update data to the AUDIT update log file with Xa_prepare, and then prepare for Xa_commit. Xa_commit
In response to the request, write the confirmation information of the AUDIT update log and perform the actual update processing to the user table 5 in the shared memory. When Xa_rollback is requested, the Rollback information is written to the AUDIT update log, and the cancellation of the update process is recorded in the log. On the shared memory, the update data is discarded, and the state before the update is guaranteed.

● 1PC Has a local transaction function, writes update data to the AUDIT update log in response to a Xa_commit request,
In addition, the actual update on the shared memory is performed at the same time. In response to the Xa_rollback request, the held update data is discarded, and the contents of the shared memory remain unchanged.

[5] Multi-Thread Control The MRM of this embodiment stores a shared memory in a database (RM:
Resource manager), and one of its purposes is to improve processing efficiency. Therefore, the MRM server is processed in multi-threads for each command (each API),
It is considered that a plurality of processes operate simultaneously.

[6] Efficient Dump Processing The MRM of the present embodiment periodically dumps data in the shared memory to a file, and provides a backup function in preparation for a failure. FIG. 4 is a diagram for explaining the processing of the MRM cyclic dump process 14.

This dump function resides in the memory while the MRM is operating, and collects a dump of the shared memory at a designated time regardless of whether the online processing is being performed. The dumped data is AUDIT update log file 1 collected at the time of update
Becomes the recovery source data while being merged with the updated contents of 5.

In the dump processing, since the contents of the shared memory are copied to a buffer and output, the larger the shared memory is, the longer the processing time of the CPU is required, which may hinder the on-in processing. To prevent this, the MRM of the present embodiment has a mechanism that has update record information in the user table 5 of the shared memory and dumps only updated records. Since only the updated records are targeted, the processing load can be significantly reduced as compared with the process of dumping the entire table.

In order to make it possible to dump updated records, the dump file 8 takes over the previous dump data as it is, and the updated data is overwritten with the inherited data.

As described above, according to the present embodiment, by using the MRM, it becomes possible to process data on the shared memory as a user's library via a simple API. Furthermore, since data can be read and written only on the shared memory, effective use of the memory and improvement in processing efficiency are realized.

The MRM manages recoverable and non-recoverable data (records) in a single management table, and makes a determination of recoverable or non-recoverable when those data (records) are accessed. Effective utilization of resources and simplification of the control mechanism can be achieved. The user can process the data without being aware of the recovery type or the non-recovery type.

The online dump of the MRM, which is used as a means for recovery in the event of a failure, can be performed by the above method.
The processing load on the CPU for dump processing can be reduced, and the effect on online processing can be minimized. That is, the updated data (record) is managed, and the range of the updated data (record) is checked at the time of dumping, thereby narrowing the range of data (record) to be dumped. If such narrowing is performed, a great effect can be expected for accumulation type data (records) that are sequentially accumulated. That is, usually, the time required for the dump processing (dump time) increases with the accumulation of data (records), but the dump processing of the MRM according to the present embodiment dumps only updated data (records). Almost constant dump time.

[0071]

As described above, according to the present invention,
In data control of small quantity and many kinds using shared memory,
Processing that ensures data consistency can be realized without being conscious of data attributes.

Further, in order to realize high-speed access,
Resource management and recovery on the shared memory can be performed efficiently.

[Brief description of the drawings]

FIG. 1 is an exemplary view for explaining a shared memory resource management system in information processing according to an embodiment;

FIG. 2 is a diagram showing an example of transaction management, table management, and data table management in a shared memory resource management system;

FIG. 3 is a diagram illustrating backup of update data;

Claims (11)

[Claims] An information processing apparatus that manages resources of a shared memory, comprising: a definition unit that defines, by parameters, table attributes and record initial values for tables and records that are resources used by an application. Means for holding the parameters set as a library that can be specified by specifying the table number and record number of the resource, and responding to a use request of the resource with the specified table number and record number in accordance with the parameters held in the library. And an access unit for accessing a resource, wherein the library is provided for each application sharing a predetermined resource. 2. An information processing apparatus for simultaneously controlling resources having different attributes, wherein resource information including attribute information indicating whether the resource is a non-recoverable type or a recoverable type is provided for a resource shared by a plurality of applications. A memory to be recorded; a resource management table that enables identification of resource information recorded in the memory by using a pointer; and a transaction for the resource that manages transactions for the resource. Management means enabling the identification information, and checking means for specifying resource information using the pointer and the resource management table in response to a resource use request of an application, and checking attributes of the requested resource. , Previous Based on the inspection result of the inspection means, and a access means for accessing the available requested resource, said access means, if the inspection result indicates an attribute of the non-restoring performs read or update the resource,
An information processing apparatus that, when indicating the recovery type attribute, searches for update data for the record from the record lock information specified by the identification information and reads or updates the searched update data. . 3. The information processing apparatus according to claim 2, wherein the access unit updates or determines the resource with the update data when an instruction to determine the resource is issued from an application. 4. An information processing apparatus for backing up update data in a shared memory during online processing, a collecting means for periodically collecting a dump of the shared memory and outputting the collected dump to a recording device; Information about updated records in the resource,
Recording means for recording information indicating a dump collection time and a resource update time in a resource management table, wherein the collection means includes a dump collection time and a resource update time recorded in the resource management table. An information processing apparatus characterized in that, when the update time of a resource is late, sampling is performed on the updated record based on information on the updated record. 5. The information processing apparatus according to claim 4, wherein the collection for the updated record collectively dumps a predetermined range including the updated record. 6. An information processing method for managing resources of a shared memory, wherein a table attribute and a record initial value are defined by parameters for a table and a record that are resources used by an application, and the defined parameter is defined by the parameters. A library which can be specified by specifying a table number and a record number of a resource, and which accesses a resource in accordance with the parameters held in the library in response to a use request of the resource whose table number and the record number are specified. An information processing method, comprising the steps of: providing a library for each application sharing a predetermined resource. 7. A memory in which, for a resource shared by a plurality of applications, resource information including attribute information indicating whether the resource is a non-recoverable type or a recoverable type is recorded.
A resource management table that enables identification of resource information recorded in the memory by a pointer, and a transaction for the resource that is managed by the application, and a transaction for the resource by the application and a record lock information by the transaction are identified by the identification information. An information processing method for simultaneously controlling resources having different attributes, comprising specifying resource information using the pointer and the resource management table in response to a resource use request of an application. Examining the attribute of the requested resource, and accessing the requested resource based on the result of the examination. When the access result indicates the non-recoverable attribute, Reading resources If update or update is performed and indicates the recovery type attribute, update data for the record is searched from the record lock information specified by the identification information, and the searched update data is read or updated. An information processing method characterized by the following. 8. The information processing method according to claim 7, wherein when an application instructs resource determination, the resource is updated or determined by the update data. 9. An information processing method for backing up update data in a shared memory during online processing, comprising: periodically collecting a dump of the shared memory; outputting the collected dump to a recording device; Information about the updated record,
And a step of recording information indicating a dump collection time and a resource update time in a resource management table, wherein the dump collection step includes a dump collection time and a resource update time recorded in the resource management table. An information processing method, wherein if the update time of the resource is late, sampling is performed on the updated record based on the information on the updated record. 10. The information processing method according to claim 9, wherein the collection for the updated record is performed by dumping a predetermined range including the updated record at once. 11. A storage medium storing a program for realizing the information processing method according to claim 6. Description: