JP2621777B2 - Deadlock avoidance method in distributed system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a deadlock avoiding method in a distributed information processing system.

[0002]

2. Description of the Related Art A deadlock in a distributed processing system is also called a global deadlock. The detection of this deadlock can be roughly classified into a method using a wait-for graph and a method using a timed lock. These can be referred to the publication "Distributed Operating System-Next to UNIX" published by Kyoritsu Shuppan.

(1) Method using wait-for graph A graph based on valid branches is created depending on the waiting state between transactions, and when a closed circuit is established, deadlock has occurred. In the case of a distributed system, deadlock is detected by communicating locking information indicating a relationship between a transaction and a resource between the systems.

A technique using this method is disclosed in
No. 208139, JP-A-62-203259 and JP-A-63-52263.

[0005] (2) Time-based locking method [0005] If the resource cannot be locked, it waits for a predetermined time, and if the time has elapsed, it is determined that a deadlock has occurred.

As a means for avoiding deadlock, a directed graph can be used to always check the directed graph before a transaction waits for a lock to determine the possibility of deadlock.

[0007]

The deadlock detection method in the above-mentioned distributed processing system has the following problems.

[0008] That is, in the wait-for graph system, it is indispensable that the locking information is communicated between the systems, which imposes a load on the communication line.

In file input / output, there are cases where the entire file to be accessed is locked and a case where the lock is applied to a part of a directory or a file block. Become.

Therefore, in the present invention, in order to reduce the amount of communication,
The relationship between deadlocked transactions is stored in advance, and the possibility of falling into a deadlock is determined before calling a transaction from a transaction.
The purpose is to avoid deadlocks in advance.

[0011]

A first method of the present invention is as follows.
A registration unit (hereinafter referred to as “registration unit”) for inputting transaction information to be suppressed and a condition at the time of the suppression from an input device and registering the deadlock among a plurality of transactions that may be deadlocked when the parallel execution is performed. Deadlock information storage means (hereinafter referred to as a deadlock information storage unit) for inputting relevant transaction from the registration means and storing information and status of a target transaction in a memory or a file; When starting, refer to the deadlock information storage means, determine the start of the transaction,
A transaction activation determination unit (hereinafter, transaction activation determination unit) for suppressing transaction activation when the condition is satisfied, and a transaction control unit (hereinafter, transaction control unit) for changing the state of the deadlock information storage unit when the transaction is activated or terminated. It is characterized by having.

According to a second method of the present invention, in the deadlock avoiding method in the distributed processing system adopting the first method, the transaction activation determining means suppresses transaction activation and does not return control to the transaction. Further, the transaction which has issued the transaction start request is abnormally terminated once, the resources occupied by the transaction are released, and then the transaction is restarted.

According to a third method of the present invention, in the deadlock avoiding method in the distributed processing system employing the first method, the registration unit inputs a transaction to be abnormally terminated when a deadlock occurs, in addition to the transaction information. The deadlock information storage unit stores a transaction to be abnormally terminated when a deadlock occurs, and the transaction activation determining unit determines, when the condition is satisfied, the transaction that is abnormally terminated and input by the registration unit as deadlock information. The transaction is obtained by obtaining from the storage means.

According to a fourth method of the present invention, in the deadlock avoiding method in the distributed processing system employing the first method, the registration unit, together with the transaction information, occupies the transaction as a condition for suppressing the transaction. The deadlock information storage unit stores the relationship with the resource together with the transaction information, and the transaction start determination unit uses the relationship between the resource and the transaction as a determination condition. I do.

[0015]

Next, an embodiment of the present invention will be described in detail with reference to the drawings.

Referring to FIG. 1 showing one embodiment of the present invention, the registration section 1 will be described first. The registration unit 1 inputs, from an input device 5 such as a keyboard, transaction information that may cause a deadlock when executed in parallel. The input information is a transaction relationship, a device name on which each transaction operates, and a transaction identifier assigned to the transaction.
For example, if a deadlock occurs as shown in FIG. 2, S1 and S2 are input as the names of the devices operating with T1, T2, T3 and T4 as the transaction identifiers, and the waiting relationship of each transaction is input. In the deadlock between the two devices shown in FIG. 2, as a precondition for occurrence of the deadlock, a relation between a transaction on another device and a transaction of the own device started from the transaction is input by the input device 5.
Enter from. In the case of the deadlock between the devices shown in FIG. 2, if the device with the device name S1 is its own device, the fact that the transaction T1 is waiting for the completion of the execution of the transaction T2 is a precondition for the occurrence of the deadlock. input.

Next, when the preconditions are satisfied, the relationship between the transaction on its own device and the transaction on another device whose startup is to be inhibited from the transaction is input from the input device 5 as the transaction information to be inhibited from being activated. . In the case of a deadlock between the two devices shown in FIG. 2, if the device having the device name S1 is its own device,
When the precondition is satisfied, the transaction T3 inputs that the activation of the transaction T4 is to be suppressed.

The control of the registration unit 1 is shifted when the input command is activated. The registration unit 1 sets, in the condition determination table 31 of the deadlock information storage unit 2, information relating to a precondition for the occurrence of a deadlock input from the input device 5 and transaction information whose activation is to be suppressed.

Next, the deadlock information storage unit 2 will be described.

The deadlock information storage unit 2 is provided in a memory or a file, and forms a control table. In this embodiment, a condition determination table 31 as shown in FIG. 3 is mounted on a memory. The condition determination table 31 is used to set information for detecting a precondition for generating a deadlock and transaction information for suppressing startup.

FIG. 3 shows an example in which the condition shown in FIG. 2 is set in the condition determination table 31. The deadlock state between the two devices shown in FIG.
Is set in the entry. Own device condition transaction identifier 3 in the second entry of condition judgment table 31
3 is set to T2 which is an identifier of the transaction T2. Similarly, in the second entry of the condition determination table 31, the condition S2 is assigned to the other device name 34, T1 is assigned to the other device condition transaction identifier 35, T3 is assigned to the own device suppression transaction identifier 36, and T2 is assigned to the other device suppression transaction identifier 37. S2 is set in T4 and the other device name 38 of inhibition. The condition counter 39 is set to “0” as an initial value.

Next, the transaction control unit 3 will be described.

The transaction control unit 3 determines the condition of the own device condition transaction identifier 33 from the first entry to the last entry of the condition determination table 31 of the deadlock information storage unit 2 for receiving a transaction start request from another device. A check is made to see if there is a transaction identifier that matches the transaction identifier on the own device requested to be started by the device. If the corresponding entry 32 exists, the transaction identifier of the activation source and the device name are set to the corresponding entry 3.
It is determined whether or not the other device condition transaction identifier 35 in 2 matches the condition other device name 34, and if they match, the value of the condition counter 39 is incremented by "1". If the condition counter is 1 or more, the precondition for deadlock occurrence is satisfied. Also, condition counter 3
When the transaction to which 9 has been added ends, control is transferred to the transaction control unit 3, and “1” is subtracted from the value of the condition counter 39.

In the case of FIG. 2, when the transaction T1 activates the transaction T2, the value of the condition counter 39 of the second entry of the condition judgment table 31 of FIG. .

Next, the transaction start determination unit 4 will be described.

The transaction start determination unit 4 is controlled when a transaction start on another device is requested from the transaction 6 on the own device. When a request to start a transaction on another device is made, a transaction start request for another device is issued from the own device-suppressed transaction identifier 36 from the first entry to the last entry of the condition determination table 31 of the deadlock information storage unit 2. Check whether there is a transaction identifier that matches the transaction identifier on the own device. If there is a corresponding entry 32, the other device-suppressed transaction identifier 37 in the entry 32
If the transaction identifier on the other device requesting activation matches the value of the condition counter 39 of the corresponding entry 32 is "1" or more, there is a possibility of deadlock, and the transaction of the other device is activated. Control is returned as an error to the transaction that requested the transaction start of the other device. In the transaction whose control has been transferred, the transaction is terminated after performing necessary post-processing.

In the deadlock between the devices shown in FIG. 2, when the transactions T3 and T4 are activated, the transaction activation determining unit 4 determines that the value of the condition counter 33 of the second entry of the condition determination table 31 is "1". Therefore, control is returned to the transaction T3 without activating the transaction T4. In transaction T3, the transaction ends. When the transaction T3 ends, the occupied resources are released, and the resource occupation of the transaction T2 is started. As a result, the transaction T2 is released from the WAIT state without falling into a deadlock state, and can continue processing. FIG. 4 shows this operation.

In the case of a deadlock between two devices shown in FIG. 2, there are two devices, but the present invention can be applied even when there are three devices as shown in FIG. . In the case of FIG. 5, if T5 is set as the other device-suppressed transaction identifier 37 of the condition determination table 31 and S3 is set as the suppressed other device name 38, a transaction on the other device S3 is started from the own device transaction T3. At this time, if the value of the condition counter 39 is “1” or more, the activation of the transaction T5 is suppressed, and deadlock can be prevented.

Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

Referring to FIG. 1, when the value of the condition counter 39 of the second entry of the condition judgment table 31 is "1", that is, when the transaction start is inhibited, the transaction start judging unit 4 sends the request to another device. The transaction on the own device is abnormally terminated once without returning control to the transaction on the own device that issued the transaction start request. By ending the above, the transaction is restarted after releasing the resources occupied by the transaction. By this operation, it is possible to prevent a deadlock and automatically restart a suppressed transaction. FIG. 6 illustrates this operation. FIG. 6 shows a state after the transaction T3 is restarted for the first time from the transaction T3 and is suppressed by the transaction startup determining unit 4, and the transaction T3 is restarted.

After the restart, if the transaction T2 ends and the resources are released, the transaction T3 can occupy the resources and activate the transaction T4. At the end of the transaction T2, the value of the condition counter 39 of the second entry of the condition determination table 31 in FIG. Therefore, when the transaction T4 is activated, the transaction activation determination unit 4 activates the transaction T4 because the transaction is not suppressed.

Next, a third embodiment of the present invention will be described in detail with reference to the drawings.

Referring to FIG. 1, the registration unit 1 inputs a parameter for abnormally terminating a prerequisite transaction as a transaction to be abnormally terminated when a deadlock occurs, together with the transaction information described above. When this parameter is input, the abnormal end flag 40 in the entry 32 of the condition determination table 31 shown in FIG. 7 is turned on.

When the transaction suppression condition is satisfied, the transaction activation determination unit 4 determines whether the abnormal termination flag 4 in the corresponding entry 32 of the condition determination table 31 is satisfied.
If 0 is on, the transaction having the own device condition transaction identifier 33 in the entry 32 is forcibly abnormally terminated. In the case of the deadlock between the two devices shown in FIG. 2, when the transaction T4 is started from the transaction T3, the transaction T2 ends abnormally, and the abnormal end result is notified to the transaction T1. Since the transaction T1 terminates and releases the occupied resources, the transaction T4 does not wait for resource occupation and can prevent a deadlock state.

Next, a fourth embodiment of the present invention will be described in detail with reference to the drawings. In the case of the deadlock between the two devices shown in FIG. 2, if the transaction T3 does not occupy resources, the deadlock does not occur. However, in the above-described embodiment, transaction activation is also suppressed for a transaction that does not occupy resources. If the target resource is known in advance, it is better to make a transaction start determination in consideration of whether or not the resource is occupied, which leads to an improvement in system throughput. An embodiment in which the presence / absence of resource occupancy is thus considered will be described below. The registration unit 1 shown in FIG. 1 inputs a relationship with a resource together with a transaction relationship.

In order to suppress the activation of the transaction of another device from the transaction on the own device, the name of the resource already occupied by the transaction on the own device is input.

The deadlock information storage unit stores a transaction relationship and a resource relationship. FIG. 8 shows an example of the condition determination table 31 in this case. Entry 32
Is set to the resource name 41.

In the case where the value of the condition counter 39 in the corresponding entry 32 of the condition judgment table 31 is "1" or more in FIG.
If the resource name 41 is already occupied by a transaction having the own device suppression transaction identifier 36, the activation of the transaction of another device is suppressed.

In the case of a deadlock between the two devices shown in FIG. 2, the transaction start determination unit 4 determines whether or not the transaction T3 has already occupied the resource R1. For this purpose, the resource R1 is registered by the registration unit 1 and set in the resource name 41 of the second entry of the condition determination table 31 in FIG.

[0040]

According to the present invention, in a distributed processing system,
By suppressing transaction activation, it is possible to prevent deadlock. In addition, the present invention can automatically suppress a transaction and, after releasing resources once, automatically restart the transaction, and can also select a transaction to be suppressed.

According to the present invention, it is possible to control the suppression of transaction startup depending on whether or not resources are occupied. These effects result in an improvement in system throughput.

[Brief description of the drawings]

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

FIG. 2 is a diagram illustrating an example of a deadlock between two devices in a distributed system.

FIG. 3 is a diagram illustrating an example of a deadlock information storage unit.

FIG. 4 is a diagram for explaining an example of a deadlock suppression operation.

FIG. 5 is a diagram for explaining an example of deadlock between three devices in a distributed system.

FIG. 6 is a diagram illustrating an example of a restart operation after a transaction is suppressed.

FIG. 7 is a diagram illustrating an example of a deadlock information storage unit to which a suppression transaction selection function is added.

FIG. 8 is a diagram illustrating an example of a deadlock information storage unit on which suppression control based on whether or not resources are occupied is loaded.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 registration unit 2 deadlock information storage unit 3 transaction control unit 4 transaction start determination unit 5 input device 6 transaction 7 own device 8 other device

Claims (4)

(57) [Claims] 1. A registration unit for inputting and registering, from an input device, transaction information whose activation is to be suppressed and conditions for suppressing the transaction, among a plurality of transactions which may be deadlocked when the parallel execution is performed, and a transaction related to the deadlock. A deadlock information storage means for inputting a segment from the registration means and storing information and a state of a target transaction; and referring to the deadlock information storage means when starting another transaction from one transaction, and And a transaction control unit that changes the state of the deadlock information storage unit when the transaction is started or terminated when the condition is satisfied. Distributed processing system Deadlock avoidance scheme that. 2. The method according to claim 2, wherein
In addition to suppressing transaction startup, without returning control to the transaction, the transaction that made the transaction startup request is terminated abnormally once, the resources occupied by the transaction are released, and then the transaction is restarted. 2. A method for avoiding deadlock in a distributed processing system according to claim 1. 3. The registration means inputs, in addition to the transaction information, a transaction to be abnormally terminated when a deadlock occurs. The deadlock information storage means stores a transaction to be abnormally terminated when a deadlock occurs. 2. The distributed processing according to claim 1, wherein, if the condition is satisfied, the activation determining unit obtains, from the deadlock information storage unit, the transaction which is input abnormally by the registration unit and terminates the transaction. Deadlock avoidance method in the system. 4. The registration unit inputs a relationship with a resource occupied by the transaction as a condition for suppressing the transaction together with the transaction information, and the deadlock information storage unit inputs the relationship with the resource together with the transaction information. 2. The deadlock avoidance method in a distributed processing system according to claim 1, wherein a relationship is stored, and said transaction activation determination means uses a relationship between a resource and a transaction as a determination condition.