JPH0361209B2 - - Google Patents

Description

[Detailed description of the invention]

[Technical Field of the Invention] The present invention provides a method for determining whether a task has a deadlock exit or not, whether interactive processing is being performed, or whether batch processing is being performed when a task enters a deadlock state due to conflicting resource occupation requests. This invention relates to a deadlock/cancellation scheduling method in which tasks to be deadlocked/cancelled are determined by taking into consideration whether the current level of the deadlock/cancellation is present or not. [Prior art and problems] Figure 1 is a diagram for explaining deadlock. In Figure 1, A to E are tasks, and DA ₁
to DA ₅ indicate the data sets, respectively. Currently, task A occupies dataset DA ₁ , task B occupies dataset DA ₂ , task C occupies dataset DA ₃ , task D occupies dataset DA ₄ , and task E occupies dataset DA ₅ . shall be taken as a thing. Under this state, when task B issues an exclusive request specifying data set _DA1 , task B enters a wait state until task A releases data set _DA1 . Similarly, under this state, when task C specifies dataset DA ₂ and issues an exclusive request, task C enters the waiting state, and when task D specifies dataset DA ₃ and issues an exclusive request, task C enters the waiting state. D enters the waiting state, and when task E issues an exclusive request specifying data set DA ₄ , task E enters the waiting state. If task A issues an exclusive request specifying data set DA ₅ while tasks B to E are in the waiting state, all of the tasks A to E become in the waiting state and a deadlock occurs. In the prior art, when a deadlock occurs, a deadlock is canceled for the task that pulled the deadlock trigger (task A in the above example). If a task that has been canceled due to a deadlock does not have a deadlock cancel exit, this task will terminate abnormally.
The task must be restarted by an operator or the like. [Object of the invention] The present invention is based on the above considerations, and includes:
When performing deadlock cancellation, consider whether or not there is a deadlock exit, whether or not interactive processing is being performed, etc. when performing deadlock cancellation, and reduce the system utilization rate due to deadlock as much as possible. The purpose of this invention is to provide a scheduling method for deadlock cancellation that can reduce the number of deadlock cancellations. [Structure of the Invention] Therefore, the deadlock cancellation scheduling method of the present invention gives a deadlock cancellation priority to tasks, and when a deadlock occurs, the deadlock cancellation of multiple tasks in an inoperable state is performed. - One task to be deadlocked and canceled is determined from among the plurality of tasks in the inoperable state based on the cancel priority. [Embodiments of the Invention] The present invention will be described below with reference to the drawings. FIG. 2 is a diagram for explaining the meanings of blocks and arrows used in FIG. 3, and FIG. 3 is a block diagram of one embodiment of the present invention. As shown in FIG. 2, the simply blank blocks in FIG. 3 indicate information, the blocks with double lines on both sides indicate processors (processing means or control means), and the thin solid lines with arrows indicate references. A thick white line with an arrow indicates an update, and a thick diagonal line with an arrow indicates a control. In FIG. 3, 1 is resource occupancy request information, 2
3 is an occupancy determination unit, 3 is an occupancy registration unit, 4 is an occupancy request result notification unit, 5 is an occupancy request result information, 6 is a deadlock detection & cancel task determination unit, 7 is an occupancy wait control unit, and 8 is a deadlock cancellation unit. Reference numeral 9 indicates an occupancy status information storage section, numeral 10 indicates an occupancy status information storage section, and numeral 11 indicates a task attribute information storage section. When the resource occupancy request information 1 is sent, the occupancy determination unit 2 refers to the contents of the occupancy status information storage unit 9 and determines whether occupancy is possible or not. If the possession is possible, the possession registration section 3
After this occupancy registration is performed, the occupancy result notifying unit 4 writes the name of the task that issued the occupancy request and the name of the resource that is the object of the occupancy request in association with the occupancy status information storage unit 9. Then, an occupancy request result status indicating that occupancy is possible is sent to a predetermined management section. When the occupancy determination section 2 determines that the occupancy is not possible, the deadlock detection and cancel task determination section 6 operates next. The deadlock detection & cancel task determination unit 6 stores the name of the task that issued the occupation request, the name of the resource that is the subject of the occupation request, the name of the task that occupies the resource that is the subject of the occupation request, and information on waiting for occupation. The contents of the section 10 and the task attribute information storage section 11 are referred to, and if it is determined that no deadlock has occurred, the task of the source of the occupation request is to be placed in a waiting state for occupation, and if a deadlock has occurred. If it is determined, the
Decide which tasks to cancel. If it is determined that the task that issued the occupancy request should be placed in the occupancy wait state, the occupancy control section 7 updates the contents of the occupancy wait information storage section 10, and then the occupancy result notification section 4 updates the contents of the occupancy wait information storage section 10. Occupancy request result information is sent to a predetermined management unit, instructing that the task that issued the occupation request should wait for exclusive use. When a deadlock occurs and a task to be deadlocked or canceled is determined, the deadlock/cancellation control section 8 updates the occupancy status information storage section 9 and the occupancy waiting information storage section 11, and then the occupancy result notification section 4 The predetermined management unit is notified of occupancy request result information instructing that the task that issued the occupancy request should be deadlocked and canceled. The exclusive waiting information storage unit 10 stores exclusive waiting information consisting of the name of the task that issued the exclusive request, the name of the resource that was the subject of the exclusive request, and the name of the task that is currently occupying the resource. There is.
The task attribute information storage unit 11 also contains the name of the task, deadlock/cancellation of the task, etc.
Task attribute information consisting of priorities is stored. For example, the deadlock cancel priority is "1" for interactive processing with a deadlock exit;
"2" for batch processing with a deadlock exit; "3" for interactive processing without a deadlock exit;
It is set to "4" for those that perform batch processing without a deadlock exit. Note that the larger the value, the higher the priority. Next, the deadlock detection logic will be explained. Waiting for exclusive use is a relationship between two tasks, and if one task is waiting for the release of a resource occupied by the other task, then "one task is waiting for exclusive use of the other task." Expressed as.
The deadlock detection & cancel task determination unit 6 in FIG. starts by storing the names of the tasks in the exclusive waiting relationship and their deadlock/cancel/priorities one after another in the register, creating a task exclusive waiting relationship table as shown below.

【table】

Claims (1)

[Claims] 1 Deadlock, cancel, and
A priority is assigned, and when a deadlock occurs, one task should be canceled from among the multiple tasks in an inoperable state based on the deadlock cancel priority of the multiple tasks in an inoperable state. A deadlock/cancel scheduling method characterized by determining tasks.