JPS63307538A - Control system for simultaneous access of files - Google Patents

Abstract

PURPOSE:To effectively avoid a deadlock state without deteriorating the performance of a computer system by advancing the processing after production of an unlock request which unlocks only one of shared blocks. CONSTITUTION:A shared block deadlock avoiding means 21 is provided on each process and produces an unlock request only to the block designated by the block identifying information included in the shared block dead lock information received from a shared block deadlock detecting means 31. Then a shared block deadlock after-the-fact means 22 is provided on each process to cancel the replacement of files done by its own process and to produce an unlock request for all blocks locked by its own process when the need is detected for the second access given by its own process to the block to which an unlock is requested by the means 21. Thus it is possible to minimize the production of unlock requests so that a deadlock state is avoided.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a simultaneous file block-by-block access control method that allows a file to be accessed simultaneously by multiple processes. Concerning methods for avoiding deadlock between processes.

[Conventional technology]

Conventionally, in this type of concurrent file access control method, techniques for deadlock avoidance include: ■ When a deadlock occurs, one of the deadlock-related processes
A technology that cancels all of the file access processing performed by one process after that process's latest checkpoint, thereby breaking the deadlock state.

■ To reduce the probability of deadlock occurrence, among the shared locks of blocks locked for file access after the process's latest checkpoint, locks of blocks that are logically deemed unnecessary are sequentially unlocked. Techniques are known that indirectly freeze the probability of deadlock occurrence by reducing the probability of waiting for a lock.

An example of the latter deadlock avoidance technology is DML (Data Manipulator) in database access.
One technique is to reduce the probability of deadlock occurrence by unlocking shared locked blocks that are no longer subject to the latest currency of a database operating language (DML) in units of DML.

[Problem that the invention seeks to solve]

In the case of the former (2) deadlock avoidance technology, the conventional file simultaneous access control method described above requires the process that performed the cancellation in addition to the cancellation process itself as a cancellation process after the latest checkpoint when a deadlock occurs. Since it is necessary to restart the process from the latest checkpoint point, it has the disadvantage that it causes a decrease in the performance of the computer system.

In addition, in the case of the latter deadlock avoidance technique, it is necessary to issue frequent unlock requests to sequentially unlock shared locks, so it can be used as a management table for lock management when unlofter performance is poor. In computer systems that use media other than memory, or in computer systems that control simultaneous access to files shared between hosts in a loosely coupled multiprocessor environment, this disadvantage can cause a decline in the performance of the computer system. There is.

In view of the above-mentioned points, an object of the present invention is to minimize the issuance of unlock requests to avoid deadlock by determining when there is no need to cancel processing at the time of occurrence of deadlock. The object of the present invention is to provide a simultaneous file access control method that allows processing to proceed in the forward direction while guaranteeing the consistency of file contents.

[Means for solving problems]

The simultaneous file access control method of the present invention includes a file, a plurality of processes accessing this file, a shared lock request to a block within the file from these plural processes, and 1Jl-receives other lock requests and unlock requests. It monitors the lock status of blocks, and when a lock wait occurs in response to a lock request, it causes the lock request process to wait, and if the lock wait process is in a deadlock state as a result of multiple waits, it is related to a deadlock. In a computer system equipped with a lock management device that returns a deadlock occurrence notification to any one of the processes, another process issues an exclusive lock request to a block that is provided with the lock management device and has a shared lock. A shared block that returns, as a type of deadlock occurrence notification, a shared block deadlock notification containing deadlock occurrence information and block identification information of the block that has held the shared lock to the process that has acquired the shared lock when a deadlock occurs. and a shared block deadlock detection means provided in each process, when receiving the shared block deadlock notification from the enclosing block deadlock detection means, sends an unlock request only to the block specified by the program identification information. The issued shared block deadlock avoidance means and the shared block deadlock avoidance means provided in each process monitor whether or not the own process will need to access the block again, and whether re-access is necessary. The shared block deadlock post-deadlock means includes a shared block deadlock post-deadlock means for canceling a file update by the own process and issuing an unlock request for all blocks locked by the own process when a deadlock is detected.

[Effect]

In the file simultaneous access control method of the present invention, a shared block deadlock detection means is provided in the lock management device, and when a deadlock occurs due to another process issuing an exclusive lock request to a block that has a shared lock, the shared block deadlock detection means is installed in the lock management device. A shared block deadlock notification that includes deadlock occurrence information and block identification information of the block that has acquired a shared lock is returned to the process that applied the shared lock as a type of deadlock occurrence notification. When a shared block deadlock notification is received from the block deadlock detection means, an unlock request is issued only for the block specified by the block identification information, and shared block deadlock post-measures are provided in each process to avoid shared block deadlock. The process monitors whether the block requested to be unlocked by the process needs to be accessed again by the process, and when the need for re-access is detected, the process cancels the file update by the process and locks the block. Performs unlock request issuing processing for all blocks.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a simultaneous file access control system according to an embodiment of the present invention.

The simultaneous file access control method (2) of this embodiment consists of a file group 1 consisting of a plurality of files 10 and a process group 2 consisting of a plurality of processes 20 including a shared block deadlock avoidance means 21 and a shared block deadlock post-deadlock means 22. , and a lock management device 3 including a shared block deadlock detection means 31.

As a rule, the process 20 issues a lock request R to the lock management device 3 in advance when executing file access A to the file 10 to lock the block to be accessed in the file 10. . In other words, file 2o is read by file access A (
If it is a READ) access, at least a shared lock request is issued to the lock management device 3, and a write (WRITE) request is issued to the lock management device 3.
) access, an exclusive lock request is issued. Furthermore, when the series of file accesses A ends, the process 20 issues an unlock request R for canceling the lock request R to the lock management device 3 to release the locked state of the corresponding block.

When the lock management device 3 returns a shared block deadlock notification in response to a lock request R from the self-process 20 to the lock management device 3, the shared block deadlock avoidance unit 21 prevents the shared block deadlock from being notified. If the block specified by the block identification information can be unlocked at that time, it is a means for issuing an unlock request only for this specified l block.

The shared block deadlock post-execution means 22 monitors whether the unlocked block will not be accessed again in subsequent processing using the shared block deadlock avoidance means 21 of the own process 20, and if the need for re-access has arisen. In this case, at this point, the current deadlock avoidance process is to cancel the file update by the own process 20 and issue a request to unlock all the blocks locked by the own process 2o.

When the lock management device 1!3 receives a lock request R or an unlock request R from a process 20, it manages which process 20 is locking which block, and manages the block that is already locked by another process. When a lock request R from 20 is received, except when a shared lock request for a block in a shared lock state is received, the process 20 requesting the lock is placed in a waiting state, and the process 20 in such a waiting state is placed in a waiting state. If a deadlock state occurs due to a plurality of deadlocks, a deadlock occurrence notification indicating that a deadlock has occurred is returned to any one of the processes 20 related to the deadlock.

The shared block lock detection means 31 detects when there is a deadlock-related process 20 that is waiting for another process 20 to issue an exclusive lock request to a block that has a shared lock. includes deadlock occurrence information indicating that a deadlock has occurred for any one of the processes 20 holding the shared lock, and block identification information of the block in the shared lock state causing the deadlock. This is a means of returning a special deadlock occurrence notification called a shared block deadlock notification.

Next, the operation of the simultaneous file access control system of this embodiment configured as described above will be explained with reference to the timing charts shown in FIG. 2 fat to +Il.

In FIG. 2 (al to (L)), a process 20 and another process different from this process 20 (denoted with the reference numeral 20'' to distinguish it from the process 20) 20'' is a file 1.
A two-dimensional block consisting of block resources assuming that a deadlock occurs due to a lock request R for block N 0 and another block N' different from this block N, and a shared block deadlock notification is returned to process 20. This is a timing chart showing the order of lock requests on a plane, and illustrates how the process 20 proceeds while avoiding a deadlock state after a deadlock occurs. Here, FIG. 2(a) to (T1 in 11 to
T9 indicates time, and a larger number indicates a later time. Also, a white circle indicates that the block is in a shared lock state, and a black circle indicates that the block is in an exclusive lock state.

As shown in FIG. 2 fat, at time T1, when the process 20 issues a lock request R to the lock management device 3 in order to lock the block N° currently in the exclusive lock state by the process 20°, the process 20 is placed in a waiting state by the lock management device 3.

Next, as shown in FIG. 2 (bl), at time T2,
This time, if process 20" issues an exclusive lock request to lock management device 3 in an attempt to take exclusive access to block N, which is currently in a shared lock state, a deadlock state will occur in lock management JIt3. Therefore, the lock management value W3 is determined by applying the shared block deadlock detection means 31.
A shared block deadlock notification containing deadlock occurrence information and program identification information of block N in a shared lock state is returned to the process 20 that holds a shared lock on block N.

Subsequently, as shown in FIG. 2 fcl, at time T3, the process 20 that received the shared block deadlock notification
uses the shared block deadlock avoidance means 21 to issue an unlock request to the lock management device 3 to unlock only the block N that is in the shared lock state. Therefore, the lock management device 3 unlocks the block N, and the deadlock state between the processes 20 and 20'' is avoided. However, the process 20 remains in the waiting state.

Next, as shown in FIG. 2 1d), at time T4,
Since block N has been unlocked, process 20'' is no longer waiting for an exclusive lock request to block N, and process 20'' proceeds to the next process.

Subsequently, as shown in FIG. 2 (el), at time T5, when the process 20'' is completed, the process 20''
All locked blocks are unlocked, and this also releases the process 20 from waiting for a lock on block N''.

Next, as shown in FIG. 2 ([1], at time T6,
When the process 20 is released from waiting for the lock on block N'', processing proceeds using the shared block deadlock post-deadlock means 22.If the process 20 does not access block N again in the subsequent processing, This eliminates the need for the self-process 20 to cancel a file update in order to avoid deadlock and to issue an unlock request for all blocks locked by the self-process 20.

If the process 20 attempts to re-access the block N unlocked by the shared block deadlock avoidance means 21 in the process 20 after the state shown in FIG. Then, this situation is detected by the shared block deadlock post-deadlock means 22, and the process of canceling the file update by the own process 20, which was conventionally done to avoid deadlock, and the own process 20
As a result, an unlock request is issued for all locked blocks.

As a result, as shown in FIG. 2+11, at time T8, the process 20 files l
The update of O is canceled, and in the lock management device 3, all blocks locked by the own process 20 are unlocked, and the process 20 returns to the processing start state.

The process 20 normally involves reprocessing.

Therefore, as shown in FIG. 2+11, at time T9, as a result of reprocessing, the process 20 is
This time, you will be able to go through the places where you had to go back.

〔Effect of the invention〕

As explained above, in order to avoid deadlocks in shared blocks, the present invention issues an unlock request to unlock only one block of a shared block and proceeds with the process. The probability that update cancellation processing, unlock request issuing processing, and process re-execution processing will be performed when a deadlock occurs is reduced, and performance degradation of the computer system will not occur as with the deadlock avoidance technology of the conventional file simultaneous access control method. This has the effect of effectively avoiding a deadlock situation without causing any problems.

Another advantage of the present invention is that it can be easily constructed based on deadlock avoidance techniques in conventional file simultaneous access control systems.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, and FIG. 2 (11 is a timing chart for explaining the operation of the simultaneous file access control method of this embodiment. ... File group, 2... Process group, 3... Lock management device, 10... File, 20... Process, 21... Shared block deadlock avoidance means, 22.
. . . Shared block deadlock post-measurement means; 31 . . . Shared block deadlock detection means.

Claims (1)

[Claims] A file, multiple processes that access this file, and receiving shared lock requests, exclusive lock requests, and unlock requests for blocks within the file from these multiple processes, and monitoring the lock state of the blocks. However, when a lock wait occurs in response to a lock request, the lock request process is made to wait, and if the lock wait process becomes deadlocked as a result of multiple lock waits, one of the deadlock-related processes is In a computer system equipped with a lock management device that returns a deadlock occurrence notification to a user, when a deadlock occurs due to another process issuing an exclusive lock request to a block that is provided with a shared lock in the lock management device. shared block deadlock detection means for returning, as a type of deadlock occurrence notification, a shared block deadlock notification including deadlock occurrence information and block identification information of the block that has acquired the shared lock to the process that has acquired the shared lock; shared block deadlock avoidance means provided in the shared block deadlock detection means for issuing an unlock request only for the block specified by the block identification information when the shared block deadlock notification is received from the shared block deadlock detection means; The shared block deadlock avoidance means monitors whether the block requested to be unlocked by the shared block deadlock avoidance means will need to be accessed again by the own process, and cancels the file update by the own process when the need for re-access is detected. and shared block deadlock post-deadlock means for issuing unlock requests for all blocks locked by the own process.