JPH0324645A - Information processor - Google Patents

Abstract

PURPOSE:To improve the updating processing efficiency of a data base file by collectively updating data records as necessary without executing addition, change and deletion processing of data records to/from a data base file at the time of updating each data record. CONSTITUTION:In the case of sharing the data base file by plural processes, the addition, change and deletion processing of data records are not executed at the time of updating each data record and its index updating information is stored in an index updating information storage device 11 in each process. At the time of setting up a rest point in each process, batch updating is executed by a whole index batch updating device 13. When an index is updated at the time of referring to the index, the start of the process is waited by an index waiting device 16 until the batch updating is ended by the device 13, and after receiving updating end information from an index updating end informing device 18, the process is started by a process starter 19.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention is utilized for data processing in an information processing device. In particular, the process of adding, changing, and deleting index records accompanying the process of adding, changing, and deleting data records in database files is not performed when updating data records, and the shared index is then updated in bulk as necessary. It relates to an information processing device.

〔overview〕

The present invention provides an information processing apparatus equipped with a data pace file having an index for a plurality of items of a data record, with item values as key values and addresses of data records as pointer values. Database file update processing is improved by not performing the addition, modification, and deletion processing of index records when updating data records, and then updating them all at once as necessary. The aim is to improve efficiency and usage rate.

[Conventional technology]

The index in the database file is additional information for improving the database reference processing speed. In particular, when field values of data records are stored redundantly as key values of an index, the index does not require consistency with the data section until reference processing using the index occurs. On the other hand, if an index record is updated using the address of the data record as a pointer value when updating a data record, if the index key value is different from the update order of the data records, there will be variations in access to the database. Therefore, it is inefficient. Therefore, a method has been adopted in which when updating a data record, update information for the corresponding index is stored, and the index is updated all at once later.

FIG. 4 shows a database file that stores a plurality of data records and index records as one block and performs input/output for each block. Considering the case where a data record 41 is added, block 42 is first read and the data record is stored in that block. Next, an index record 43 is created, and block 44 is read and the created index record is stored. Similarly, an index record 45 is created and stored in block 46.

Now, consider the case where there are two input/output buffers 52 and 53 available for human output of this database file. Since buffers 52 and 53 are full by the time block 46 is read, block 4
You need to write out 2 or 44.

The block to be written is assumed to be block 42, meaning the block that was accessed first. Next, data record 47
, it is necessary to read block 42 again, but at that time it is necessary to write block 46. If there are 10 data records to be added, 10 x 3 x 2 = 60 input/outputs will occur to the database file.

When updating the index all at once, the number of input/outputs that occur when adding 10 data records is block 4.
2 is read and written once each, and blocks 44 and 469 are read and written four times in the index record addition process. A total of six inputs and outputs to the database will be required, which is one-tenth of the number of times compared to the case where the batch update is not performed. This method of updating indexes all at once is effective for buffers that update a large number of indexes at once compared to input/output buffers.

[Problem that the invention seeks to solve]

By the way, when performing a batch update of indexes, in order to maintain consistency between the data section and the index section, it is necessary to match the update order of the data section and the index update order. Considering the case where a database file is shared by multiple processes, a means is needed to guarantee the order in which the index section is updated as the data section is updated by each process. In addition, it is necessary to guarantee consistency between the data part and the index part during failure recovery when updating the database file. Furthermore, when referring to an index that is being updated at once, the process is to refer to an index that reflects the update information of the index up to that point. If it is a single process, it is enough to reflect the index update information that occurred in that process, but if multiple processes are updating the index, in order to process the index reference request that occurred in one process, It is necessary to reflect all index update information in all processes at that time.

The conventional index batch update method does not solve the problem of update order and failure recovery when database files are shared by multiple processes, and the problem of update information reflection method when referencing the index. There was a problem in that it only allowed access from one process, making it impossible to improve the update processing efficiency and usage rate of database files.

The purpose of the present invention is to solve the above-mentioned problems.
Adapts to database files that are shared, updated, and referenced by multiple processes. The index section of the database file can be updated all at once after updating the data section, improving the update processing efficiency and usage rate of the database file. An object of the present invention is to provide an information processing device equipped with an index batch update method.

[Means for solving problems]

The present invention provides an information processing apparatus including a database file having an index for a plurality of items of a data record, with the value of the item as a key value and the address of the data record as a pointer value. An index update information storage means for storing index update information in units of processes and in units of indexes without actually updating the index when an index is updated when an index part is shared and used by a plurality of processes; an index update start notifying means for notifying the start of storage of the index update information; and an index update start notification means for notifying the start of storage of the index update information in that process when a quiescent point is established for each process, and indexes for which storage of the index update information has been notified by the index update start means in that process are collectively updated based on the index update information. an index update information erasing means that invalidates the index update information in a recovery process in a process in which a failure occurs during a database file update; and an index update information eraser that disables the index update information by the index update notification means when the index is referenced. an index reference notification means for notifying that the index will be referenced when there is a notification that the index has been stored; an index update waiting means for waiting for use of the index until the index can be referenced; Index batch update means that actually updates the index based on the index update information when the corresponding index update information exists in all processes including the process that attempts to refer to the index that has been notified of index reference by the notification means. and an index update end notification means for notifying the end of index update when the index batch update is completed, and notification of the end of index update from all the processes that have been notified to store the index update information by the index update start notification means. If the index update waiting means waits for the index update, the index update waiting means starts the process waiting for the index update waiting means.

[Effect]

When a database file is shared and used by a plurality of processes, the index update information is stored in the index update information storage means for each process without updating the index every time it is updated. Then, when a quiescent point is established for each process, a batch update is performed by the batch update means for all indexes. Furthermore, if a failure occurs while updating the database file, the index information of the process is invalidated by the index update information erasing means. Furthermore, if an index is updated when the index is referenced, the index waiting means waits for the startup of the relevant process until the batch update is completed by the index batch updating means, and after the index update notifying means notifies the completion of the update, the process starting means Activate it.

In other words, update processing for the same data record is
Taking advantage of the property of waiting until the quiescent point of the first process to update, index update information is stored for each index in each process, and the index information is restricted during recovery processing, allowing multiple processes to share the database. Even when updated, consistency between the data section and index section is guaranteed. This makes it possible to improve the update efficiency and usage rate of database files.

〔Example〕

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.

This embodiment is an information processing system that includes a database file that has an index for a plurality of items in a data record, with the value 48 of the item shown in FIG. 4 as a key value 49, and the address 50 of the data record as a pointer value 51. In the device, when the data part and index part of the database file are shared and used by multiple processes, an index that stores index update information in each process and each index without actually updating the index when updating the index. Update information storage means 11;
an index update start notification means 12 for notifying the start of storage of the index update information; and an index update start means 1 for a process when a quiescent point is established for each process.
All index batch update means 13 updates all indexes for which storage of index update information has been notified by 2 at once based on the index update information, and the index update information is updated in a recovery process in a process in which a failure occurs during database file update. index update information erasing means 14 for invalidating the index update information, and an index reference notification for notifying that the index will be referenced if there is a notification that the index update information has been stored by the index update notification means 12 when the index is referenced. means 15, index update waiting means 16 that waits for index use until index reference becomes possible, and index reference notification means 15.
An index batch update means 17 that actually updates the index based on the index update information if the corresponding index update information exists in all the processes including the process that attempts to refer to the index and is notified of index reference by. When the index batch update is completed, the index update end notification means 18 notifies the end of the index update, and the index update start notification means 12 sends an index update end notification from all the processes that have been notified to store the index update information. The process starting means 19 is provided for starting the process being waited for by the index update waiting means 16 when the index update waiting means 16 is notified. In addition,
In FIG. 1, E is the update start message queue and F is the reference start message queue.

The features of the present invention are as shown in FIG. 1, index update information storage means 11, index update start notification means 12,
All index batch update means 13, index update information deletion means 14, index reference notification means 15, index update waiting means 16, index batch update means 17, index update completion notification means 18, and process starting means 19 are provided. .

Next, regarding the operation of this embodiment, FIGS. 2 to 4,
This will be explained with reference to Tables 1 to 5.

FIG. 2 shows an example of update processing performed by a plurality of processes on the database file in the state shown in FIG. 4.

First, in process 1, data record 21 is added to address 3. At this time, as shown in Table 1, the index update information of the key value C1 pointer value 3 and the index update information of the key value c1 pointer value 3 are stored in the process 1 by the index update information storage means 11 (Step 61) .In addition, index update start notification means 1
2, the update start message queue E is notified of the start of index update regarding index 1 and index 2 of process 1.

Next, in process 2, data record 22 is added to address 4. At this time, as shown in Table 1, the index update information storage means 11 stores the index update information of key value D, pointer value 4, and index update information of key value d1 and pointer value 4 in process 2 (procedure 62〉.

Furthermore, the index update start notifying means 12 notifies the update start message queue E of the start of index update regarding index l and index 2 of process 2.

Next, in process 1, the data record at address l is replaced with data record 23. At this time, as shown in Table 1, index update information is created by replacing the change process with deleting the index record before change (step 63) and adding the index record after update (step 64). Also, since process 1 has already been registered in the update start message queue E, the index update start notification means l
2 does not work.

Thereafter, process 2 attempts to delete the data record at address l, but since this data record has already been updated by process l, process 2 enters a wait state. Table 1 shows the contents of the index update information at the time when process 1 establishes a quiescent point.

(Hereinafter referred to as the margin of this page) Once the static point is established, all indexes are updated simultaneously 1
3 takes out the update notification regarding process 1 from the update start message queue E, and sends the index batch update means 17
Index 1 is updated all at once, and then index 2 is updated all at once.

Index update information that has been updated is deleted. At the quiescent point of process l, process 2 is released from waiting, and the data record at address 1 is deleted. At this time, the contents of the data record at address 1 have been replaced by data record 23, so the index update information for key value X1 pointer value 1 and key value As shown in the table, the index update information is added to process 2 (step 65).Table 2 shows the contents of the index update information at the time when process 2 reaches the quiescent point.

Next, a case where the database file shown in FIG. 4 is updated by multiple processes shown in FIG. 3 will be described. Figure 3 shows the process l of the process in Figure 2.
shows the case where a failure occurs before the quiescent point is established. The contents of the index update information at the time the failure occurs are shown in Table 1.

When a failure occurs, the database system automatically performs a database file recovery process, and the contents of the database file are returned to the state shown in FIG. 4. Along with this, the index update information erasing means 14 takes out the update notification regarding process 1 from the update start message queue E, and deletes the update information of index 1 and index 2. This operation maintains consistency between the data section and the index section. After the recovery processing of process 1 is completed, the waiting state of process 2 is released, and the data record at address 1 is deleted. At this time, the contents of the data record at address 1 remain as shown in Figure 4, so as shown in Table 3, the index update information of key value A1 pointer value 1 and key value a1 pointer value 1 is the index update information. added to process 2 by storage means l (step 6
6).

Table 3 shows the contents of the index update information when process 2 reaches the quiescent point.

(The following is the margin of this page) As mentioned above, by taking advantage of the property that update processing for the same data record waits until the quiescent point of the process that first updated it, index update information is stored in each process, and recovery By deleting index update information during processing, consistency between the data section and index section can be guaranteed even when the database is shared and updated by multiple processes.

Next, the operation when the reference process and the update process share an index will be described. Table 4 shows the case where one index is simultaneously updated and referenced by three processes.

Process 1 stores index update information by index update information storage means 11 as data records are updated. At this time, the index update start notification means 12 registers the process 1 in the update start message queue E (step 71).Next, the process 2 receives the index update information by the index update information storage means 1 as the data record is updated. Process 2
Store in . At this time, the index update start notification means 12 registers the process 2 in the update start message queue E (step 72).Furthermore, as the process 3 updates the data record, the index update information storage means 11 receives the index update information. It is stored in process 3. At this time, the index update start notification means 12 registers process 3 in the update start message queue E (step 73).

When the process 2 establishes a quiescent point, the index of the process 2 is updated at once by the all index batch update means 13, and the process 2 is deleted from the update start message queue E by the index update end notification means 18 (step 74). Even if the process 2 attempts to refer to the index at this point, there is an index that requires updating that has been notified to the update start message queue E, so the index waiting means 16 waits for the completion of the index update.

Prior to the meeting, the index reference start notification means 15 registers the process 2 in the reference start message queue F (step 75). After that, when the process 3 attempts to update the index, the index reference start notification means 15 registers the process 2 in the reference start message queue F.
Since the start of reference to the index has been notified, the index batch update means 17 is operated to collectively update the index.

At this time, the index update end notification means 18 deletes process 3 from the update start message E (step 76). Even when process l performs the update process,
Since the reference start message queue F is notified of the index reference start, the index batch update means 17
and update the index in bulk. At this time,
The index update end notification means 18 deletes the process 1 from the update start message queue E (step 77). When the contents of the update start message queue E become empty, the process activation means 19 deletes the process registered in the reference start message queue F. This allows process 2 to refer to the index that reflects all updates of process 1, process 2, and process 3.

Finally, the case where there are multiple reference processes that share an index will be explained using Table 5.

The process 1 stores index update information by the index update information storage means 11 as the data records are updated. At this time, the index update start notification means 12 registers the process 1 in the update start message queue E (step 81>. Next, the process 2
The index update information is stored in process 2 by the index update information storage means 11 as the data record is updated. At this time, the index update start notification means 12 sends the update start message I. -Register process 2 in E (step 82). Furthermore, the process 3 updates the index update information storage means 1 as the data records are updated.
1, index update information is stored in process 3. At this time, the index update start notification stage 12 registers the process 3 in the update start message queue E (
Step 83).

When the process 2 establishes a quiescent point, the index of the process 2 is updated at once by the all index batch update means 13, and the process 2 is deleted from the update start message queue E by the index update notification means 18 (
Step 84). Even if process 2 attempts to refer to the index at this point, there is an index that requires updating that has been notified to the update start message queue E, so
The index waiting means 16 waits for the completion of index updating. Prior to the meeting, the index reference start notification means 12 registers process 2 in the reference start message queue F (step 85).

When the process 3 establishes a quiescent point, the index of the process 3 is updated at once by the all index batch update means 13, and the process 3 is deleted from the update start message queue E by the index update completion notification means 18 (step 86>. Then) , even if process 3 attempts to refer to the index, there is still an index that requires updating that has been notified to the update start message queue E, so the index waiting means 16 waits for the index update to be completed.

Prior to the meeting, the index reference start notification means 15 registers process 3 in the reference start message queue F (step 87). At the time when the process 1 performs the update process, the reference start message queue F is notified of the start of index reference, so the index batch update means 17 is operated to update the indexes in batch. At this time, the index update end notification step 18 deletes the process 1 from the update start message queue E (step 88
〉.

When the contents of the update start message queue E become empty, the process activation means 19 sequentially activates the processes registered in the reference start message queue F. This allows Process 2 and Process 3 to refer to the index that reflects all updates of Process 1, Process 2, and Process 3.

(Hereinafter, the margin of this page) As described above, according to the procedure of the present invention, even when a database is shared and updated and referenced by multiple processes, indexes can be updated all at once while maintaining the consistency of data and indexes. Update processing becomes possible.

〔Effect of the invention〕

As explained above, the present invention has the effect of improving the update processing efficiency of database files by collectively updating indexes, and the effect of improving the usage rate of database files by being able to share the indexes that are being updated in batches. I have both.

Below are the effects of sharing indexes that are updated in bulk.

l) If processing to update dozens of data at once occurs irregularly, it can be processed immediately at the location where it occurs.

If shared use is not possible, it is necessary to accumulate the generated data and process it in one place.

2> Database files that are being batch processed can be referenced on other terminals. For example, you can check intermediate results and processing progress. If shared use is not possible, you will have to wait until the process is finished.

Next, we will discuss the effects of batch updating when indexes are shared.

3) Processing efficiency is improved by updating all at once. This is especially effective when the number of buffers that can be used for input/output with database files is small.

4) By uniquely determining the index update order and key update order for each index for batch updates,
The order in which resources are secured by updating the index is uniquely determined, and deadlock conditions caused by lock control can be reduced.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention. FIG. 2 is an explanatory diagram showing an example (1) of update processing in a plurality of processes according to the embodiment. FIG. 3 is an explanatory diagram showing an example (2) of update processing in a plurality of processes according to the embodiment. FIG. 4 is an explanatory diagram showing data and index input/output in the embodiment. 1l... Index update information storage means, 12...
Index update start notification means, 13... All index batch update means, 14... Index update information deletion means, 15... Index reference notification means, l6.
... Index update waiting means, 17... Index batch update means, l8... Index update completion notification means, 19... Process starting means, 21, 22
, 23, 41, 47...data record, 42, 44
, 46... Block, 43, 45, 49... Index record, 48... Item value, 50... Address, 51... Pointer value, 52, 53... Buffer, E...・Update start message queue, F...Reference start message queue El Nail Closure Me・Sozejiki 11 ``: Bandit lfl@Me・Nsezokyu- Okinuti no Hashi Yako 1 Figure

Claims (1)

[Scope of Claims] 1. An information processing device comprising a database file having an index for a plurality of items of a data record, with the value of the item as a key value and the address of the data record as a pointer value, comprising: an index update information storage means for storing index update information in each process and each index without actually updating the index when updating the index when the data part and the index part are shared and used by multiple processes; an index update start notification means for notifying the start of storing the index update information; and an index update information for an index for which storage of the index update information has been notified by the index update start means in that process when a process-based quiescent point is established. an index update information erasing means for invalidating index update information in a recovery process in a process in which a failure occurs during a database file update; An index reference notification means for notifying that the index will be referenced when there is a notification that index update information has been stored, and an index update waiting means for waiting to use the index until the index can be referenced. , If the corresponding index update information exists, all processes including the process that attempts to refer to the index that has been notified of index reference by the index reference notification means actually update the index based on the index update information. an index batch update means for notifying the end of index update when the index batch update is completed; 1. An information processing apparatus comprising: a process starting means for starting a process being waited for by the index update waiting means when the completion of the update of the index update waiting means is notified.