JPH02266444A - Fault recovery system for distributed data base - Google Patents

Abstract

PURPOSE:To shorten the recovery time of a fault site by substituting a saved site for the data processing transaction of the fault site, and when the fault is recovered, transferring the saved data to the fault site to restore the data of the fault site. CONSTITUTION:When a fault is generated, a data projecting means 18 in the fault site 11 protects the data of a main storage 16 for a fixed period, and simultaneously with the protection, a fault informing means 19 in the fault site 11 informs the generation of the fault to the other site 12 and a data save determining means 21 determines which data in the main storage 16 are to be saved to which site. Based upon the determination, both the data communication means of respective sites transmit/receive data and the data are saved to the determined site. On the other hand, the site 12 receiving the data from the fault site 11 substitutes the transaction processed buy the fault site at the time of generating the fault. When the fault of the fault site 11 is recovered, the internal data communication means 22 transmits the saved data to the fault site 11. Consequently, the recovery of the fault site is rapidly executed.

Description

Detailed Description of the Invention [Object of the Invention] (Industrial Application Field) The present invention relates to a failure recovery method in a distributed database management system.

(Prior Art) Conventional disaster recovery methods for distributed databases include, for example, the one described in "Principles of Database Systems" by J. Ulman, translated by Kunii and Taiho (May 25, 1985). , published by Japan Computer Association, p.54
4).

As shown in FIG. 3, each site (computer) 2A, 2B connected by a communication path 1 is equipped with a CPU 3, a main memory 4, a database 5, and a log 6, and a failure occurs at one site 2A. In this case, the faulty site 2A notifies all other sites 2B that the fault has occurred, and the site 2B that received the notification records which site the fault has occurred in in the log 6, and when the fault is recovered. , an example is described in which the log 6 is examined going back to the time when a failure occurred, and all data common to the data of the failure site 2A is sent to the failure site 2A as the latest data.

However, in the above-described failure recovery method, when a failure occurs, all data on the main memory 4 of the failure site 2A must be discarded and all transactions that were processing that data must be aborted.

In this case, the storage capacity of storage devices has become extremely large due to the development of VLSI technology, so at the time a failure occurs, a large amount of unprocessed data exists on the main memory 4 in the failure site 2A. As mentioned above, the amount of data to be discarded is enormous.

Therefore, after the failure at the failure site 2A is recovered, in order to return the data on the main memory 4 to the state before the failure, the necessary data is read from the database 5 and the transaction that was aborted at the time of failure is restarted. must be executed from

Furthermore, while the fault at the faulty site 2A is not recovered, other sites 2B cannot execute transactions that process data existing only at the faulty site 2A.

(Problems to be Solved by the Invention) As described above, in the conventional failure recovery method, there is a problem that not only does it take time to recover from a failure, but also that the number of transactions that are aborted while the failure is not recovered increases.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a distributed database failure recovery method that can shorten the recovery time of a failed site and prevent an increase in transactions due to aborts.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention solves a problem in a distributed database that handles data groups scattered at multiple computer sites interconnected by communication paths. In the recovery method, each site has a data protection means that protects the data on main memory for a certain period of time when a failure occurs, a failure notification means that notifies other sites of the occurrence and recovery of a failure, and a method that evacuates the data on main memory. Equipped with a data evacuation decision means for determining a site and a data communication means for transmitting and receiving data to be evacuated, when a failure occurs in one of the sites, the data necessary for the failed site to perform processing on behalf of the other site is provided. By evacuating the data to the failed site, the evacuated site handles data processing transactions for the failed site, and when the failure is recovered, the data at the failed site is transferred to the failed site by transferring the evacuated data to the failed site. It is characterized by recovery.

(Operation) In the present invention, with the above configuration, when a failure occurs at one site, the data protection means of the failed site protects the data on the main memory for a certain period of time. At the same time, the failure notification means of the failed site notifies other sites of the occurrence of the failure, and the data evacuation determining means determines which data in the main memory is to be evacuated to which site. Based on this determination, the data communication means of each site sends and receives data, and the data is saved to the determined site. This data is necessary for another site to handle data processing for the failed site. The data on the main memory in the failed site is protected until the above processing is completed.

On the other hand, the site that accepts data from the failed site will act on behalf of the transactions that were being processed by the failed site at the time of the failure. Further, when the fault at the faulty site is recovered, the fault notification means of the faulty site notifies other sites that the fault has been recovered. Then, at the site that receives this notification, its internal data communication means sends the data that was evacuated as described above to the failed site, and as a result, the main memory of the failed site has the same data as at the time of the failure. is stored as data.

(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 distributed database failure recovery method of the present invention. The figure shows an example in which three computer sites 11, 12, and 13 are connected to each other by a communication path 14.

Each site 11, 12, 13 is a CP that controls the overall operation
The system includes U 15, main memory 16, database 17, data protection means 18, fault notification means 19, load status notification means 20, data evacuation determination means 21, and data communication means 22.

The data stored in the main memory 16 is protected for a certain period of time by the data protection means 18 based on a command from the CPU 15 when a failure occurs.

For example, when a failure occurs within the site 11, the failure notification means 19 notifies other sites 12.
This is a means for notifying the site 12 and 13 that a failure has occurred, and when the failure has been recovered, it also notifies the other sites 12 and 13 of the failure recovery.

The load status notification means 20 is a means for checking the load status of the CPU 15, main memory 16, etc. in the own site 12, 13, and notifying the failed site 11 of the contents when receiving a notification of the occurrence of a failure.

The data evacuation determining means 21 is a means for determining which data is to be evacuated to which site based on the contents of the load status notification means 20 of the normal site 12.13, and is adapted to reduce the load on the normal site 12.13. This is to distribute the data to be saved.

The data communication means 22 transfers the data determined by the data evacuation determination means 21 as described above to the normal site 12.1.
:3 is a means for transmitting to the data communication means 22.

Next, a failure recovery method will be explained using the flowchart shown in FIG.

First, in step 201, a certain site, for example site 1
Assume that a failure occurs in 1.

Then, in step 202, the data protection means 18 in the failed site 11 protects the data stored in the main memory 16 for a certain period of time. This protection operation is performed in step 206 when the data in the main memory 16 of the failed site 11 is transferred to the other site 12.
This will continue until it is completely evacuated to 13.

In step 203, the other sites 12 and 13 are notified of the failure at the same time as the failure occurs.

Next, in step 204, the load status notification means 20 of the sites 12 and 13 that have received notification of the occurrence of the failure check the load status of the CPU 15 main memory 16, etc. of each of their own sites 12 and 13, and notify the failure site 11 of the results. do.

In step 205, the data evacuation determining means 21 of the failed site 11 determines which data in the main memory 16 of the failed site 11 is to be saved to which site based on the load status of the other site 12.13.
Decide whether to evacuate. In this case, other sites 12.
The data saved to 13 will be processed by another site 12.1.
3 is the data necessary for acting on behalf of the user.

In step 206, the data communication means 22 of the faulty site 11 transmits data to the data communication means 22 of the other sites 12.13 based on the data evacuation decision in step 205, so that the data in the main memory 16 of the faulty site 11 is It will be evacuated to another site 12.13. These saved data are stored in the main memory 16 of the other site 12.13. During this time, the data in the main memory 16 of the failed site 11 is protected by the data protection means 18.

Next, in step 207, the other site 12.13 whose data has been evacuated performs the transaction of the failed site 11 on behalf of the other site 12.13. That is, the data in the main memory 16 of the faulty site 11 is saved to the main memory 20 of the other site 12.13, so transactions that were being processed at the faulty site 11 will continue to be processed at the other site 12.13. . Also,
While the failure at the failed site 11 is not recovered, the other sites 12 and 13 execute not only transactions that process data in each database 17, but also transactions that process data that was in the main memory 16 of the failed site 11 at the time of the failure. Also execute.

Next, when the fault at the faulty site 11 is recovered in step 208, one fault notification means of the faulty site 11 notifies the site 12.13 that the fault has been recovered in step 209.

In step 210, in response to this recovery notification, the sites 12 and 13 that received the notification transmit the evacuated data from their internal data communication means 22 to the data communication means 22 of the faulty site 11. As a result, the same data as when the failure occurred is stored as the latest data on the main memory 16 of the failure site 11.

Finally, in step 11, when a failure occurs, the failure site 1
If data that was not present in the main memory 16 of the site 12 is updated while the failure has not been recovered, the updated data is sent from the site 12.13 to the failed site 11. This results in
The failed site 11 completely recovers from the failure.

In the following explanation, it is assumed that a failure has occurred at site 11, but the same applies if a failure occurs at other sites (12 and 13).

Also, when saving data, it takes time to communicate the data, but the time required to communicate this data is compared to the time required to read the same amount of data from a database stored on a magnetic disk etc. to main memory. It is very short. For example, if you send 109 bytes of data at a communication speed of 1010
If the data is transmitted via a bit/second LAN, the time required for the communication is approximately 109×8÷1010=0.8 (seconds). on the other hand,
Access time for the same amount of data is 3X10-2 seconds (10
When reading data of 7 bytes or less) from a database stored on a magnetic disk, etc., to the main memory, the time required is approximately 3 x 10-2 x (109 ÷ 107) = 3 C seconds)
It is. Therefore, the time to save data is very short.
There is no practical problem.

As explained above, according to this example, the following effects are achieved.

<1) If a failure occurs at one of the sites, the data that is being processed in the main memory of the failed site is evacuated to another site, so the data from the failed site can be processed at the site where the data was evacuated. Transactions can be performed on your behalf. Therefore, even if a failure occurs, the system not only functions effectively, but also discards large amounts of data as in the past, reads the data necessary to recover the failed site from the database, and then aborts when a failure occurs. It is possible to effectively solve the problem that a transaction has to be executed from the beginning and it takes time to recover, and the problem that the number of transactions that are aborted while the failure is not recovered increases.

(2) Transactions that process saved data can be executed while the failure is not recovered, so the number of transactions that are aborted or made to wait can be reduced compared to the past.

Furthermore, since the saved data is always updated to the latest data, there is no need to newly update the data when the failure is recovered, and the data can be processed as is.

Note that the present invention is not limited to the above embodiments,
It can be implemented with appropriate modifications without departing from the gist of the invention.

[Effects of the Invention] As described above, since the present invention is a distributed database failure recovery processing method as described in the claims, it is possible to shorten the recovery time of a failed site and prevent an increase in transactions due to aborts.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the distributed database failure recovery method of the present invention, FIG. 2 is a flowchart showing its operation, and FIG. 3 is an explanatory diagram showing a conventional failure recovery method. 11.12.13...Computer site 14...Communication path 15...CPU 16...Main memory 17...Database 18...One data protection means...Failure notification means 20...・Load status notification means 21...Data evacuation decision means 22...Data communication means fee

Claims (1)

[Claims] In a disaster recovery method for a distributed database that handles data groups scattered across a plurality of computer sites interconnected by communication paths, data protection means protects data in main memory for a certain period of time when a failure occurs at each site; A failure notification means for notifying other sites of the occurrence and recovery of a failure, a data evacuation determining means for determining a site to evacuation of data on the main memory, and a data communication means for transmitting and receiving the data to be evacuated. When a failure occurs on one site, the data necessary for processing on behalf of the failed site is evacuated to another site, so that the evacuated site can perform data processing transactions on behalf of the failed site. . A disaster recovery method for a distributed database, characterized in that when a fault is recovered, the data at the faulty site is restored by transferring the evacuated data to the faulty site.