WO2024001851A1 - Resource scheduling method, apparatus and system - Google Patents

Abstract

Provided are a resource scheduling method, apparatus and system, which relate to the technical field of communications, and can increase the resource utilization rate. The resource scheduling method comprises: acquiring a first job of a resource to be scheduled; screening for a job set of borrowable resources in a cluster computing system, wherein the job set comprises at least one job, and the at least one job is a job of an allocated resource and is a borrowable resource matching the first job; and determining a resource scheduling policy for the first job according to the borrowable resources for the job set, wherein the resource scheduling policy is used for indicating a processing mode in which the first job is executed using the borrowable resources for the job set at a resource-borrowable stage.

Description

A resource scheduling method, device and system

This application is required to be submitted to the State Intellectual Property Office on June 27, 2022, application number 202210741724.6, and the Chinese patent application titled "A method of resource scheduling" and submitted to the State Intellectual Property Office on November 15, 2022, application number 202211427018.0, the priority of the Chinese patent application titled "A Resource Scheduling Method, Device and System", the entire content of which is incorporated into this application by reference.

Technical field

The present application relates to the field of communication technology, and in particular, to a resource scheduling method, device and system.

Background technique

In a cluster computing scenario, a user can apply for computing resources from the cluster computing system to run a job submitted by the user, and the management node in the cluster computing system schedules computing resources for the job.

Usually, the computing resources of the cluster computing system are limited, and the management node uses reserved resources to allocate resources to jobs. However, jobs are not continuously running, that is, the resources reserved for the job are often not in use all the time, which results in low resource utilization.

Contents of the invention

This application provides a resource scheduling method, device and system that can effectively improve resource utilization.

In order to achieve the above purpose, this application adopts the following technical solutions:

In a first aspect, this application provides a resource scheduling method, which includes: a management node of a cluster computing system obtains a first job of a resource to be scheduled; and selects a job set of loanable resources in the cluster computing system, where the job set includes at least one Jobs, the at least one job is a job that has allocated resources and can be borrowed resources that match the first job; and then determines the resource scheduling policy of the first job based on the loanable resources of the job set, and the resource scheduling policy is used to indicate that in In the resource borrowing stage, the loanable resource is used to execute the first operation.

In this application, for the first job of resources to be scheduled, the management node can determine the resource scheduling policy of the first job based on the job set of loanable resources matching the first job in the cluster computing system. In this way, the resource scheduling policy of the first job can be flexibly allocated to the first job. The loanable resources of the job set that allocates resources are seconded to the first job, making full use of the resources of the cluster computing system and improving the resource utilization of the cluster computing system.

In a possible implementation, the above-mentioned screening of job sets that can be loaned resources in the cluster computing system includes: filtering from one or more jobs according to the resource loan information of one or more jobs in the cluster computing system. A first job set whose job type matches the first job is selected, and a second job set that can borrow resources to meet the resource requirements of the first job is selected from the first job set. The resource loan information is used to indicate the job types supported by the loanable resources of the job when the job's loanable resources are seconded to other jobs. The job types include: at least one of breakpoint resume type, termination type or continuous type; resource The requirements include requirements of resource types and configuration requirements of the resource types, and the resource types include at least one of computing resources, storage resources, or network resources.

In another possible implementation, the above-mentioned determining the resource scheduling policy of the first job based on the loanable resources of the job set includes: determining the resource scheduling policy of the first job based on the loanable resources of the second job set.

In this application, the job set selected based on job type and resource requirements (i.e., the above-mentioned second job set) matches the first job, that is, the resources of the jobs in the job set are qualified to be seconded to the first job. On this basis on, seconding the loanable resources of the job set to the first job, so that the resources of the jobs in the job set are fully utilized.

In another possible implementation manner, the loan duration of the loanable resource is less than or equal to the duration of the resource loanable phase, so that the management node schedules the resource for the first job based on the loan duration. Before the secondment period expires, the loanable resources can be seconded to the first operation; when the secondment period expires, the resources can no longer be seconded to the first operation.

In another possible implementation, the second job set includes a second job, and when the job type of the first job is a resumable transfer type, the second job set determines the first job based on the secondable resources of the second job set. The resource scheduling strategy includes: seconding the loanable resources of the second job to the first job during the resource borrowing stage of the second job, and suspending the loaning of the second job when the secondment period of the second job expires. Resources are seconded to the first job; when the second job ends, the seconded resources will be transferred to the first job to continue running the first job. The resource borrowing period of the second job can be seconded to the first job, which improves the utilization of the resources allocated to the second job, and the borrowing time expires. Pause the first job, so that it does not affect the operation of the second job.

In another possible implementation, the second job set includes a second job, and when the job type of the first job is terminated, the resource scheduling policy of the first job is determined based on the loanable resources of the second job set. , including: seconding the loanable resources of the second operation to the first operation during the resource borrowing stage of the second operation, and terminating the secondment of the loanable resources of the second operation to the first operation when the secondment period of the loanable resources expires. The first operation can second the resources of the second operation to the first operation during the resource borrowing period, which improves the utilization of the resources allocated for the second operation, and terminates the first operation when the borrowing time expires. In this way, it does not affect Run the second job.

In another possible implementation, the second job set includes a second job. When the job type of the first job is continuous, the resource scheduling policy of the first job is determined based on the loanable resources of the second job set. , including: during the resource borrowing stage of the second operation, seconding the second operation's borrowable resources to the first operation, and when the secondment period of the loanable resources expires, continuing to secondment the second operation's resources to the first operation operation, the resource can be seconded to the first operation during the resource borrowing stage of the second operation, thereby improving the utilization of the resources allocated to the second operation.

In a second aspect, this application provides a management node, including: an acquisition module, a first determination module, and a second determination module, wherein the acquisition module is used to acquire the first job of resources to be scheduled; and the first determination module is used in the cluster. Screening a job set of loanable resources in the computing system, the job set includes at least one job, the at least one job is a job that has allocated resources and can be loaned to the resource that matches the first job; the second determination module is used to select the job set based on the job set. The loanable resources determine the resource scheduling policy of the first job, and the resource scheduling policy is used to indicate the processing method of using the loanable resources to execute the first job during the resource borrowing stage.

In a possible implementation manner, the above-mentioned first determination module is specifically configured to filter the first job that matches the job type of the first job from one or more jobs according to the resource secondment information of one or more jobs in the cluster computing system. A job set is provided, and a second job set that can borrow resources to meet the resource requirements of the first job is selected from the first job set. The resource loan information is used to indicate the job types supported by the loanable resources of the job when the job's loanable resources are seconded to other jobs. The job types include: at least one of breakpoint resume type, termination type or continuous type; resource The requirements include requirements of resource types and configuration requirements of the resource types, and the resource types include at least one of computing resources, storage resources, or network resources.

In another possible implementation, the above-mentioned second determination module is specifically configured to determine the resource scheduling policy of the first job according to the loanable resources of the second job set.

In another possible implementation, the loan duration of the loanable resource is less than or equal to the duration of the resource's loanable period.

In another possible implementation, the above-mentioned second job set includes a second job. When the job type of the first job is a breakpoint resumable transfer type, the second determination module is specifically configured to determine when the resources of the second job can be borrowed. In the stage, the second job's loanable resources are seconded to the first job, and when the secondment period of the second job expires, the second job's loanable resources are suspended to the first job; when the second job ends , schedule the loanable resources to the first job to continue running the first job.

In another possible implementation, the above-mentioned second job set includes a second job. When the job type of the first job is a termination type, the second determination module is specifically configured to determine the second job during the resource borrowing phase of the second job. The loanable resources of the second operation are seconded to the first operation, and when the secondment period of the loanable resources expires, the secondment of the loanable resources of the second operation to the first operation is terminated.

In another possible implementation, the above-mentioned second job set includes a second job. When the job type of the first job is continuous, the second determination module is specifically configured to determine the second job during the resource borrowing stage of the second job. The loanable resources of the second operation are seconded to the first operation, and when the secondment period of the loanable resources expires, the resources of the second operation are seconded to the first operation.

In a third aspect, the present application provides a management node, including a memory and at least one processor connected to the memory. The memory is used to store computer program code. The computer program code includes computer instructions. When the computer instructions are executed by at least one processor , causing the management node to execute the method described in any one of the first aspect and its possible implementations.

In a fourth aspect, the present application provides a computer-readable storage medium on which computer instructions are stored. The computer instructions execute the method described in any one of the first aspect and its possible implementations when the computer is running.

In a fifth aspect, the present application provides a computer program product. The computer program product contains computer instructions. When the computer instructions are run on a computer, the method described in any one of the first aspect and its possible implementations is executed.

In a sixth aspect, this application provides a chip, including a memory and a processor. The memory is used to store computer instructions. The processor is used to call and run the computer instructions from the memory to execute any of the first aspect and its possible implementations. one of the method described.

In a seventh aspect, this application provides a cluster computing system, including a management node and at least one computing node. The management node uses the method described in any one of the first aspects and its possible implementations.

It should be understood that the beneficial effects achieved by the technical solutions of the second to seventh aspects of the present application and the corresponding possible implementations can be referred to the above-mentioned technical effects of the first aspect and the corresponding possible implementations, here No longer.

Description of drawings

Figure 1 is a schematic architectural diagram of a cluster computing system provided by this application;

Figure 2 is a schematic diagram of the stage division of an operation provided by this application;

Figure 3 is a schematic diagram of the hardware structure of a management node provided by this application;

Figure 4 is one of the schematic diagrams of a resource scheduling method provided by this application;

Figure 5 is one of the schematic diagrams of a resource scheduling strategy provided by this application;

Figure 6 is a second schematic diagram of a resource scheduling strategy provided by this application;

Figure 7 is the third schematic diagram of a resource scheduling strategy provided by this application;

Figure 8 is the fourth schematic diagram of a resource scheduling strategy provided by this application;

Figure 9 is one of the structural schematic diagrams of a management node provided by this application;

Figure 10 is the second structural schematic diagram of a management node provided by this application.

Detailed ways

In order to solve the problem of low resource utilization in the prior art, this application provides a resource scheduling method that screens a set of jobs that can borrow resources in a cluster system, and then determines the resource scheduling strategy for the first job. In this way, the loanable resources of the job can be used to perform the processing of other jobs to achieve the purpose of making full use of resources.

The resource scheduling method provided by this application will be introduced in detail below with reference to the accompanying drawings.

The method provided by this application is applied to the cluster computing system. The user can submit a job to the cluster computing system. The management node in the cluster computing system allocates resources to the user's job and runs the job based on the allocated resources. The application scenarios of cluster computing systems can include high performance computing (HPC), artificial intelligence (artificial intelligence, AI), and big data fusion.

Figure 1 is a schematic architectural diagram of a cluster computing system. Referring to Figure 1, the cluster computing system includes a management node 101 and one or more computing nodes 102. Among them, the management node 101 is mainly responsible for the management of the cluster computing system (including configuration management, resource management, etc.). The management node 101 includes a scheduler. The scheduler is used to allocate computing resources to jobs submitted by users and schedule the jobs (or dispatched) to the corresponding computing node 102, so that the computing node 102 executes the processing process of the job based on the allocated resources. The process in which the above scheduler allocates resources to jobs and schedules jobs to corresponding computing nodes is resource scheduling.

The resource borrowing stage of a job refers to the stage when the computing resources of the job are allowed to be borrowed. That is, in the resource borrowing phase of the job, the resources of the job (the resources of the job refer to the resources allocated to the job) are allowed to be loaned to other jobs. .

The resource unborrowable stage of a job refers to the stage when the computing resources of the job are not allowed to be borrowed. That is, during the resource unborrowable stage of the job, the resources of the job are not allowed to be loaned to other jobs.

The resource borrowing stage and the resource non-borrowing stage of a job can be defined based on the resource utilization of the computing node during the operation of the job and/or the time period during which the job is run. Of course, the resource borrowing stage can also be defined based on other factors. This article The application examples are not limiting.

In one implementation, during the process of running a user-submitted job on a computing node, the resource utilization of the computing node cannot reach a fully utilized state. In some stages, the resource utilization of the computing node is high, and in some stages, the resource utilization of the computing node is high. Resource utilization is low. Therefore, the job running stage in which the resource utilization rate is lower than the preset utilization rate can be regarded as the resource borrowable stage, and the stage in which the resource utilization rate is higher than the preset utilization rate can be regarded as the resource unborrowable stage.

During the job running process, some jobs not only have a real computing phase (hereinafter referred to as the run phase), but also have a job preparation phase (hereinafter referred to as the pre-job phase) and/or a job post-processing phase (hereinafter referred to as the post -job stage). Among them, the pre-job stage occurs before the run stage, and the post-job stage occurs after the run stage. In the pre-job stage, the environment deployment and inspection of the job, or job data transmission (also called large file transfer), etc. are performed. Perform environment cleanup or job data deletion in the post-job phase.

For example, referring to Figure 2, job 1 includes a pre-job phase, a run phase and a post-job phase, in which environment deployment and inspection are performed in the pre-job phase, and environment cleaning is performed in the post-job phase; job 2 also includes The pre-job stage, the run stage and the post-job stage, among which, the pre-job stage carries out job data transmission, and the post-job stage carries out job data deletion. Usually, the pre-job phase and the post-job phase usually only consume a small amount of resources, that is, the resource utilization rate is low. Therefore, in the embodiment of the present application, the pre-job phase or the post-job phase of the job can be used as a resource to borrow. Stages, for example, job 1 in Figure 2 includes two resource borrowing stages, and job 2 also includes two resource borrowing stages.

In addition, in the run phase of the job, there may be one or more phases with low resource utilization, and the phase with low resource utilization in the run phase of the job can also be used as the resource borrowing phase. In addition, other stages in the run stage are the stages where resources cannot be borrowed.

In another implementation, during the operation of the job, one or more time periods belong to the idle time period of the job. Therefore, the idle phase of the job can be regarded as the resource-borrowable phase, and the busy phase of the job can be regarded as the resource-unborrowable phase. .

In the embodiment of this application, the device that executes the resource scheduling method is a management node of the cluster computing system. The management node can be a desktop computer, a portable computer, a personal digital assistant (PDA), and other devices. Please refer to Figure 3 to introduce the hardware structure of the management node provided by this application. The various components shown in Figure 3 may be implemented in hardware, software, or a combination of hardware and software including one or more signal processing and/or application specific integrated circuits.

As shown in Figure 3, the management node may include: a processor 301, a memory 302, and a communication interface 303. Among them, the processor 301, the memory 302 and the communication interface 303 may be connected to each other through a bus 304, or in other ways.

Among them, the processor 301 is the control center of the communication device. The processor 301 can be a general central processing unit (CPU) or other general processor. The general processor can be a microprocessor or a CPU. Any regular processor etc. For example, the processor 301 may include an application processor (application processor, AP), a graphics processing unit (graphics processing unit, GPU), an image signal processor (image signal processor, ISP), a controller, etc.

The controller in the processor 301 is the nerve center and command center of the communication device. The controller can generate operation control signals based on the instruction operation code and timing signals to complete the control of fetching and executing instructions. Optionally, the processor 301 may also be provided with a memory for storing instructions and data.

Exemplarily, the processor 301 may include one or more CPUs, such as CPU 0 and CPU 1 shown in FIG. 3 .

Memory 302 includes, but is not limited to, random access memory (RAM), read only memory (ROM), erasable programmable read-only memory (EPROM), fast Flash memory, or optical memory, disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures that can be accessed by a computer. In this embodiment of the present application, the memory 302 may store computer instructions, which include instructions for executing the resource scheduling method provided by the present application.

In a possible implementation, the memory 302 may exist independently of the processor 301. The memory 302 can be connected to the processor 301 through the bus 304 for storing data, instructions or program codes. The processor 301 calls and executes instructions or program codes stored in the memory 302 to implement the function of a scheduler to schedule resources for jobs submitted by users.

In another possible implementation, the memory 302 can also be integrated with the processor 301 .

The communication interface 303 is used for the management node to communicate with other devices or communication networks, such as Ethernet, RAN, wireless local area networks (WLAN), etc.

The bus 304 may be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, CXL, UB, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of presentation, only one thick line is used in Figure 3, but it does not mean that there is only one bus or one type of bus.

It should be noted that the management node shown in Figure 3 is only an example of a management node. The management node may have more or less components than those shown in Figure 3, and may combine two or more components. parts, or can have different part configurations.

The resource scheduling method provided by the embodiment of the present application is described below with reference to the accompanying drawings. The execution subject of the resource scheduling method is the management node of the cluster computing system.

Based on the content described in the above embodiments, as shown in Figure 4, the resource scheduling method provided by the embodiment of the present application may include the following steps:

S401. Obtain the first job of the resource to be scheduled.

The first job is a job currently submitted by the user, that is, the first job is a job waiting for the management node to schedule resources for it.

In the embodiment of this application, when the user submits the first job, he needs to indicate the resource requirements of the first job, that is, he needs to indicate the resource requirements of the first job. The resource requirements include resource type requirements and resource type configuration requirements. In this way, After the management node obtains the resource requirements of the first job, it allocates appropriate resources to the first job according to the resource requirements of the first job when scheduling resources for the first job.

The resource type may include at least one of computing resources, storage resources or network resources. Computing resources include CPU or GPU, storage resources include memory and hard disk, and network resources include bandwidth. For example, if the resource type is computing resources, the corresponding resource configuration requirements may be the number of CPU and/or GPU cores; if the resource type is storage resources, the corresponding resource configuration requirements may be the size of memory and/or hard disk; If the resource type is a network resource, the corresponding resource configuration requirement can be bandwidth. For example, when the user submits the first job, he or she may indicate that when allocating resources for the first job, the GPU must be scheduled with two cores.

When the user submits the first job, he also needs to carry the job type of the first job. In this way, the management node in the cluster computing system can learn the job type of the first job. In the embodiment of the present application, the operation type includes: at least one of a resume type, a termination type, or a continuous type.

Breakpoint-resumable jobs refer to jobs that have checkpoints during the running process, that is, the intermediate calculation results need to be cached during the running process of the job. For some application jobs, such as AI training jobs, intermediate calculation results need to be cached regularly (which can be said to require regular checkpoints). Usually, during the running process of the job, the intermediate calculation results are not saved, and the final calculation results are saved when the calculation process ends. However, for some jobs with relatively long calculation processes (such as AI training jobs), if the intermediate calculations are not cached during the calculation process, As a result, once the intermediate calculation results are lost, the job needs to be restarted. Therefore, by establishing checkpoints, important intermediate calculation results can be stored in reliable storage space to ensure the smooth operation of the business.

A terminated job refers to a job that can be terminated periodically. For example, for some applications that save calculation results regularly, the application's job can be terminated after saving the results, that is, the job is killed periodically.

Continuous jobs refer to jobs that are not allowed to be paused during the running process, that is, it is necessary to ensure that the job runs continuously until the end of the operation.

Users submit jobs through the command line and carry the parameters of the job in the command line. The parameters of the job include the above job type and the resource requirements of the job. After the management node runs the command line, the resource requirements and job type of the job can be obtained.

After the above-mentioned user submits a job to borrow resources (for example, the above-mentioned first job), the management node places the job into a job queue corresponding to the job type according to the job type. The parameters of the job queue that borrows resources can be set in the text configuration file in the management node. For example, the queue parameters can include the queue name (QueueName) and the queue type (LoanType). The queue name and queue type are divided accordingly according to the job type. . For example, the queue name can be runlimitQueue or ckpntQueue or waitQueue, where runlimitQueue represents a terminated job queue, ckpntQueue represents a resume-type job queue, and waitQueue represents a continuous job queue; the queue type is runlimit or ckpnt or The wait type, among which, the runlimit type corresponds to the termination type operation, the ckpnt type corresponds to the breakpoint resume type operation, and the wait type corresponds to the continuous type operation.

S402. Screen the job set of loanable resources that match the first job in the cluster computing system.

The set of jobs that can be borrowed resources obtained through the above screening includes at least one job, and the at least one job is a job to which resources have been allocated.

When the user submits the first job, one or more jobs with allocated resources already exist in the cluster computing system. It should be understood that the method for the management node to allocate resources to the one or more jobs is to allocate resources to the one or more jobs according to the resource requirements of the one or more jobs.

In a possible implementation, as shown in Figure 4, the above S402 can be implemented through S4021-S4022.

S4021. Based on the resource loan information of one or more jobs that can loan resources in the cluster computing system, select the first job set that matches the job type of the first job from the one or more jobs.

It should be understood that for any one of the one or more jobs, the user also carries the resource loan information of the job when submitting the job, so that the management node can store the resource loan information of the job, and the resource loan information of the job is used to indicate that the job will be The job The types of operations supported by the loanable resources when the loanable resources are seconded to other operations. The job types supported by the loanable resources of at least one job can be understood as the types of jobs to which the loanable resources of at least one job can be loaned.

In addition, when users submit a job, they also need to carry information indicating the stages of the job and information indicating whether the resources at each stage can be borrowed. In this way, the management node can learn which stage of the job is the resource loanable stage and which stage is the resource. The unborrowable stage, for example, the resource borrowable stage of a job is a stage in which the resource utilization rate of the job is lower than the preset utilization rate. The preset resource utilization rate can be determined according to the actual situation, for example, it can be set to 50%. Regarding the resource loanable stage, For the description, reference is made to the relevant descriptions of the above embodiments, which will not be described again here.

Optionally, among one or more jobs that have allocated resources, some jobs may include multiple resource-borrowing phases. The loanable resources of the multiple resource-borrowing phases of the same job may support the same business type, or they may Differently, the embodiments of this application are not limited.

The following uses an example to illustrate the acquisition process of the first job set. Assume that the jobs with allocated resources in the cluster computing system include job1, job2, job3, job4, and job5. The first job is recorded as job6. The details of the job are shown in Table 1.

Table 1

If job6 is a resume-type job, by executing the above S4021, the jobs matching the job type of the first job (i.e., job6) are selected from job1, job2, job3, job4, and job5 to be job1 and job5, so the first The job set is represented as {job1, job5}.

S4022. Select a second set of jobs from the first set of jobs that can borrow resources to meet the resource requirements of the first job.

The resource type includes at least one of computing resources, storage resources, or network resources. For related content on computing resources, storage resources, and network resources, please refer to the description of the above embodiments, and will not be described again here.

The resource requirement of at least one job meeting the resource requirement of the first job means that the resource configuration of the resources corresponding to at least one job is higher than the resource configuration requirement of the first job. Continuing to take the first job set in S4021 as an example, the first job set includes two jobs, job1 and job5. Assume that the resources allocated for job1 are 4-core CPU and 4M memory; the resources allocated for job5 are 1-core CPU, 8M memory, the resource requirement of the first job is 5M memory; then the job selected from job1 and job5 that meets the resource requirement of the first job is job5, so the second job set is expressed as {job5} .

Optionally, in the embodiment of the present application, in the process of selecting the job set from one or more jobs with allocated resources in the cluster computing system, the process is not limited to filtering based on resource requirements and job types supported by seconded resources. In the order of S4021-S4022, you can first filter the jobs that can be seconded to support the job type of the first job, and then filter out the jobs that match the resource requirements from the jobs that match the job type to get the job set; or 6, you can also first Jobs with matching resource requirements are screened, and then jobs of a job type that can be seconded to support resources for the first job are selected from the jobs with matching resource needs to obtain a job set, which is not limited by the embodiments of this application.

The above S4021-S4022 is to filter at least one job that matches the resource requirements of the first job according to the job type. That is to say, the resource requirements of at least one job meet the resource requirements of the first job, and the secondable resources of at least one job support the second job. The job type of a job.

In another possible implementation, screening the job set of loanable resources that match the first job in the cluster computing system may include: filtering from one or more jobs that can loan resources in the cluster computing system that satisfy the first job resource requirements or filter the job set that matches the job type of the first job.

S403. Determine the resource scheduling policy of the first job according to the loanable resources of the job set.

The resource scheduling policy of the first job is used to indicate the processing method of executing the first job using loanable resources during the resource borrowing stage. It should be understood that when the job type of the first job is different, the processing method of executing the first job is different.

Combined with the contents of S402, the job set includes at least one job, and the resource scheduling policy of the first job is determined in S403. Including the following two situations:

Case 1: If the resource requirements of the at least one job and the job types supported by the loanable resources match the first job, the resource scheduling policy of the first job is determined based on all the resources in the loanable resources of the job set.

Case 2: If the resource requirements of the at least one job or the job type supported by the loanable resources match the first job, then a subset of the loanable resources based on the job set (the subset of loanable resources refers to the loanable resources) part of the resources) determines the resource scheduling strategy of the first job. For example, if the job type supported by the loanable resources of at least one job matches the first job, it is necessary to further filter out one or more jobs from the at least one job, and the resource requirements of the one or more jobs are consistent with the first job. The resource requirements of the job are matched, and then the resource scheduling strategy of the first job is determined based on the loanable resources of the one or more jobs, that is, the subset of loanable resources mentioned above refers to the loanable resources of one or more jobs in the job set. resource.

In one possible implementation, for one or more jobs that have allocated resources, the user can also specify the loan duration of the job's loanable resources. The loan duration of the job's loanable resources refers to the secondment of the job's loanable resources. When seconded to the first operation, the length of time the first operation can utilize the seconded resource. In this way, after the management node obtains the secondment period, it can schedule resources for the first job according to the secondment period. It should be noted that when the secondment period expires, its resources can no longer be seconded to the first job.

It should be noted that the loan duration of a job's loanable resources is less than or equal to the duration of the loanable phase of the job's resources. When the loan duration of the job's loanable resources is equal to the duration of the job's borrowable resources, the user also submits the job. The loan duration does not need to be specified. In this case, the default loan duration is the duration of the period when the resource can be borrowed.

Based on the above S4021-S4022 and related contents, continue to refer to Figure 4. Correspondingly, the above S403 can be implemented through S4031.

S4031. Determine the resource scheduling policy of the first job based on the loanable resources of the second job set.

In this embodiment of the present application, the second job set includes a second job. In this way, determining the resource scheduling policy of the first job based on the loanable resources of the second job set specifically determines the resources of the first job based on the loanable resources of the second job. Scheduling strategy.

For different types of jobs, resource scheduling strategies are different. The following takes the resumption type, termination type and continuous type in the above embodiment as examples to introduce the scheduling strategies of these three types of jobs respectively.

When the job type of the first job is breakpoint resume type and the second job set includes the second job, the above S4031 specifically includes S4031a:

S4031a. During the resource borrowing stage of the second operation, second the second operation's secondable resources to the first operation, and when the secondment period of the second operation expires, suspend secondment of the second operation's secondable resources to the first operation. One job; when the second job ends, the loanable resources are scheduled to the first job to continue running the first job.

For example, assume that the jobs to which resources have been allocated in the cluster computing system include Job 1 to Job 7. The details of the jobs are as shown in Table 2 below.

Table 2

The first job of the resource to be scheduled is recorded as job 8. It is assumed that the job type of job 8 is breakpoint resume type.

In one case, the above-mentioned S4021 is executed to filter jobs whose job types match the first job, and the first job set is obtained as {job 1, job 3, job 6}. Assume that S4022 is executed to filter jobs whose resource requirements match the first job. The second job set is obtained as {job 1}. The running conditions of job 1 and job 8 can be referred to (a) in Figure 5. The second job is job 1. The scheduler dispatches job 1 to a computing node. After job 1 enters the resource borrowing stage, the scheduler Job 8 submitted by the user is also dispatched to the computer node, and the computing node is instructed to use the resources of job 1 to run the job during the resource borrowing stage of job 1. Job 8; when the resource borrowing phase of job 1 expires, the management node notifies the computing node to stop running job 8 (that is, job 8 performs checkpoint), and notifies the computing node to start the computing phase of job 1; when the computing phase of job 1 runs At the end, the computing node notifies the scheduler that job 1 has ended. Afterwards, the scheduler notifies the computing node to continue running job 8, that is, restarting job 8.

In another case, the above-mentioned S4021 is executed to filter jobs whose job types match the first job, and the first job set is obtained as {job 1, job 3, job 6}. Assume that S4022 is executed to filter jobs whose resource requirements match the first job. , the second job set is obtained as {Job 6}, and the loaning time of the loanable resources of Job 6 is 30 minutes (the resource borrowing period of Job 6 is longer than 30 minutes). The running status of Job 6 and Job 8 can be referred to In (b) in Figure 5, the computing node uses the resources of job 6 to run job 8 during the resource borrowing stage of job 6; when the loan period of job 6 expires, it stops running job 8 (that is, job 8 performs checkpoint); When job 6 ends, the computing node continues to run job 8, that is, restarts job 8.

When the job type of the first job is termination type, the second job set includes the second job, and the above S4031 specifically includes S4031b:

S4031b. During the resource borrowing stage of the second operation, second the second operation's secondable resources to the first operation, and when the secondment period of the second operation expires, terminate the second operation's second operation's secondable resources to the first operation. One assignment.

Still taking jobs 1 to 7 in S4031a as an example, the first job of the resource to be scheduled is recorded as job 8. It is assumed that the job type of job 8 is termination type.

In one case, the above-mentioned S4021 is executed to filter jobs whose job types match the first job, and the first job set is obtained as {job 2, job 5, job 7}. Assume that S4022 is executed to filter jobs whose resource requirements match the first job. The second job set is obtained as {job 2}. The running status of Job 2 and Job 8 can be referred to (a) in Figure 6. The computing node uses the resources of Job 2 to run Job 8 during the resource borrowing phase of Job 2; when the resource borrowing phase of Job 2 expires, Terminate running job 8 (that is, job 8 is killed).

In another case, the above-mentioned S4021 is executed to filter jobs whose job type matches the first job, and the first job set is obtained as {job 2, job 5, job 7}. Assume that S4022 is executed to filter jobs whose resource requirements match the first job. , the second job set is obtained as {job 5}. The running status of Job 5 and Job 8 can be referred to (b) in Figure 6. In the resource borrowing phase of Job 5, the computing node uses the resources of Job 5 to run Job 8; when the resource borrowing phase of Job 5 expires, Terminate running job 8 (that is, job 8 is killed).

When the job type of the first job is continuous, the second job set includes the second job, and the above S4031 specifically includes S4031c:

S4031c. During the resource borrowing stage of the second operation, second the second operation's borrowable resources to the first operation, and when the secondment period of the second operation's borrowed resources expires, continue to second the second operation's resources to the first operation. .

Still taking jobs 1 to 7 in S4031a as an example, the first job of the resource to be scheduled is recorded as job 8. Assume that the job type of job 8 is continuous. In one case, the above S4021 is executed to filter the job type and the first job. A job matches the job, and the first job set is obtained as {Job 4}, and S4022 is executed to determine that Job 4 meets the resource requirements of the first job, so the second job set is obtained as {Job 4}. The running status of Job 4 and Job 8 can be seen in Figure 7. In the resource borrowing phase of Job 4, the computing node uses the resources of Job 4 to run Job 8; when the resource borrowing phase of Job 4 expires, Job 4 is suspended. Continue to run job 8; when job 8 ends, continue to run job 4, that is, job 4 needs to wait for job 8 to end before it can continue to run.

In a possible implementation, the job set matching the first job obtained by executing the above S4021-S4022 may include multiple jobs. The multiple jobs not only match the resource requirements of the first job, but also support the first job. The job type of the job. In this case, the scheduler can select one job from multiple jobs as the target job (ie, the above-mentioned second job), and then second the resources of the target job to the first job.

Optionally, the scheduler can select the job that first enters the job queue as the target job from multiple jobs based on the first-in, first-out principle; or the scheduler can also randomly select a job from multiple jobs as the target job, Second the resources of the target job to the first job; or, the scheduler can select the highest priority job as the target job based on the secondment priorities of multiple jobs.

In another possible implementation, the resources of the second job can be borrowed. In addition to being seconded to the first job, the resources of the second job can also be seconded to one or more other jobs. In this way, the resources of the second job are be fully utilized. For example, the first job is a short job. Since the short job consumes very few resources, there are still many remaining resources during the resource borrowing stage of the second job, and the remaining resources can also be seconded to other short jobs. For example, as shown in Figure 8, assuming that the second job is job 3 among the seven jobs mentioned above, and the first job is job 8, during the resource borrowing stage of job 3, job 8, job 9 and job can be run. 10.

In another possible implementation, the job set matching the first job obtained by executing the above S4021-S4022 includes multiple jobs, the multiple jobs support the job type of the first job, and each of the multiple jobs If the resources of a single job cannot meet the resource requirements of the first job, the sum of the resources of the multiple jobs can meet the resource requirements of the first job. In this case, the multiple jobs are all used as target jobs (that is, the second job includes multiple jobs), that is, the scheduler lends the loanable resources of the multiple jobs to the first job.

For example, still taking jobs 1 to 7 in S4031a as an example, the first job of the resource to be scheduled is recorded as job 8. If the job type of job 8 is a resumable transfer type, in one case, by The set of jobs matching the first job obtained by executing S4021-S4022 is {job 1, job 3}. In this way, during the resource borrowing stage of job 1 and job 3, job 8 is run using the resources of job 1 and job 3. When When the shortest loan duration among Job 1 and Job 3 expires, job 8 will be paused; when job 1 and job 3 both finish running, job 8 will continue to run.

If the job type of job 8 is terminated, in one case, the job set matching the first job obtained by executing S4021-S4022 is {job 2, job 7}. In this way, between job 2 and job 7 In the resource borrowing stage, use the resources of Job 2 and Job 7 to run Job 8. When the loan duration of Job 2 and Job 7 with the shortest loan duration expires, job 8 will be terminated.

In the above embodiments, only some examples are used to illustrate resource scheduling methods. During the specific implementation process, there may be more situations, and the embodiments of this application will not list them one by one.

In summary, in the resource scheduling method provided by the embodiments of the present application, for the first job of resources to be scheduled, the management node can filter out at least one available resource that has been allocated and matches the first job among multiple jobs in the cluster computing system. The job that borrows resources obtains a job set, and then determines the resource scheduling policy of the first job based on the loanable resources of the job set. In this way, the loanable resources of the job set that have allocated resources can be flexibly seconded to the first job, which fully Utilizing the resources of the cluster computing system improves the resource utilization of the cluster computing system.

Accordingly, the embodiment of the present application provides a management node. In the embodiment of the present application, the management node can be divided into functional modules according to the above method examples. For example, each functional module can be divided corresponding to each function, or two functional modules can be divided into two. Or two or more functions are integrated into one processing module. The above integrated modules can be implemented in the form of hardware or software function modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical function division. In actual implementation, there may be other division methods.

In the case where each functional module is divided corresponding to each function, FIG. 9 shows a possible structural diagram of the management node involved in the above embodiment. As shown in Figure 9, the management node includes an acquisition module 901, a first determination module 902 and a second determination module 903. The acquisition module 901 is used to acquire the first job of the resource to be scheduled, for example, executing S401 in the above method embodiment. The first determination module 902 is configured to screen a job set that can lend resources in the cluster computing system. The job set includes at least one job, and the at least one job is a job that has allocated resources and matches the first job and can lend resources, For example, S402 in the above method embodiment is executed. The second determination module 903 is configured to determine the resource scheduling policy of the first job based on the loanable resources of the job set. The resource scheduling policy is used to indicate the processing method of using the loanable resources to execute the first job during the resource borrowing stage, for example, executing the above S403 in the method embodiment.

It should be understood that the management node in this embodiment of the present invention can be implemented by a central processing unit (CPU), an application-specific integrated circuit (ASIC), or a programmable logic device. (programmable logic device, PLD) implementation, the above PLD can be a complex programmable logical device (CPLD), field-programmable gate array (field-programmable gate array, FPGA), general array logic (generic array logic, GAL ), data processing unit (DPU), system on chip (SoC), or any combination thereof. When the resource scheduling method shown in Figure 4 can also be implemented through software, the management node and its respective modules can also be software modules.

Optionally, the above-mentioned first determination module 902 is specifically configured to filter the first job that matches the job type of the first job from the one or more jobs according to the resource secondment information of one or more jobs in the cluster computing system. a job set, and select a second job set from the first job set that can lend resources to meet the resource requirements of the first job. The resource loan information is used to indicate the job types supported by the loanable resources of the job when the job's loanable resources are seconded to other jobs. The job types include: at least one of breakpoint resume type, termination type or continuous type; resource Requirements include resource type requirements and resource type configuration requirements. The resource type includes at least one of computing resources, storage resources or network resources, such as performing S4021-S4022 in the above method embodiment.

Optionally, the above-mentioned second determination module 903 is specifically configured to determine the resource scheduling policy of the first job based on the loanable resources of the second job set, for example, executing S4031 in the above method embodiment.

Optionally, the second job set includes a second job. When the job type of the first job is a resumable transfer type, the second determination module 903 is specifically configured to assign the second job to the second job during the resource borrowing stage of the second job. The loanable resources of the job are seconded to the first job, and when the loan duration of the loanable resources expires, the second job's loanable resources are suspended to the first job; when the second job ends, the loanable resources are Schedule to the first job to continue running the first job, for example, perform S4031a in the above method embodiment.

Optionally, the above-mentioned second job set includes a second job. When the job type of the first job is terminated, the second determination module 903 is specifically configured to transfer the second job's The loanable resources are seconded to the first operation, and when the loan duration of the loanable resources expires, seconding the loanable resources of the second operation to the first operation is terminated, for example, S4031b in the above method embodiment is executed.

Optionally, the above-mentioned second job set includes a second job. When the job type of the first job is continuous, the second determination module 903 is specifically configured to, in the resource borrowing stage of the second job, change the available resource of the second job. The seconded resource is seconded to the first operation, and when the secondment period of the secondable resource expires, the resource of the second operation is continued to be seconded to the first operation, for example, S4031c in the above method embodiment is executed.

Each module of the above-mentioned management node can also be used to perform other actions in the above-mentioned method embodiment. All relevant content of each step involved in the above-mentioned method embodiment can be quoted from the functional description of the corresponding functional module, which will not be described again here.

In the case of using an integrated unit, FIG. 10 shows another possible structural diagram of the management node involved in the above embodiment. As shown in Figure 10, the management node provided by the embodiment of the present application may include: a processing module 1001 and a communication module 1002. The processing module 1001 can be used to control and manage the actions of the management node. For example, the processing module 1001 can be used to support the management node to perform S401, S402 (including S4021), S403 (including S4031, S4031a, S4031b and S4031c), and/or other processes for the techniques described herein. The communication module 1002 may be used to support communication between the management node and other network entities, for example, to support communication between the management node and a computing node. Optionally, as shown in Figure 10, the management node may also include a storage module 1003 for storing computer instructions and data.

The processing module 1001 can be a processor or a controller (for example, it can be the above-mentioned processor 301 shown in Figure 3), and the above-mentioned processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors. , a combination of DSP and microprocessor, etc. The communication module 1002 may be a communication interface (for example, it may be the above-mentioned communication interface 303 as shown in Figure 3). The storage module 1003 may be a memory (for example, it may be the above-mentioned memory 302 shown in Figure 1).

When the processing module 1001 is a processor, the communication module 1002 is a communication interface, and the storage module 1003 is a memory, the processor, transceiver and memory can be connected through a bus.

For more details on how the modules included in the above management node implement the above functions, please refer to the descriptions in the previous method embodiments, which will not be repeated here.

Each embodiment in this specification is described in a progressive manner. The same and similar parts between the various embodiments can be referred to each other. Each embodiment focuses on its differences from other embodiments.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using a software program, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted over a wired connection from a website, computer, server, or data center (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.) to another website, computer, server or data center. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center integrated with one or more available media. The available media may be magnetic media (e.g., floppy disks, magnetic disks, magnetic tapes), optical media (e.g., digital video discs (DVD)), or semiconductor media (e.g., solid state drives). SSD)) etc.

The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application shall be covered by the protection scope of the present application. . Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims (16)

1. A resource scheduling method, characterized in that it is applied to management nodes in a cluster computing system, and the method includes:

   Obtain the first job of the resource to be scheduled;

   Screening a job set of loanable resources in the cluster computing system, where the job set includes at least one job, the at least one job is a job that has allocated resources and is a loanable resource job that matches the first job;

   Determine the resource scheduling policy of the first job according to the loanable resources of the job set, and the resource scheduling policy is used to indicate the processing method of using the loanable resources to execute the first job during the resource borrowing stage. .
2. The method according to claim 1, characterized in that, screening the job set of loanable resources in the cluster computing system includes:

   According to the resource loan information of one or more jobs in the cluster computing system, a first job set matching the job type of the first job is selected from the one or more jobs; wherein the resource loan The information is used to indicate the type of operations supported by the loanable resources of the operation when the loanable resources of the operation are seconded to other operations. The operation types include: at least one of breakpoint resume type, termination type or continuous type;

   Screen a second set of jobs from the first set of jobs that can borrow resources to meet the resource requirements of the first job; the resource requirements include requirements for resource types and configuration requirements for the resource types, and the resource types include At least one of computing resources, storage resources or network resources.
3. The method of claim 2, wherein determining the resource scheduling policy of the first job based on the loanable resources of the job set includes:

   The resource scheduling policy of the first job is determined according to the loanable resources of the second job set.
4. The method according to any one of claims 1 to 3, characterized in that,

   The loan duration of the loanable resource is less than or equal to the duration of the borrowable period of the resource.
5. The method according to claim 3 or 4, characterized in that the second job set includes a second job, and when the job type of the first job is a breakpoint resume type, the second job set according to the second job The loanable resources of the job set determine the resource scheduling policy of the first job, including:

   During the resource borrowing stage of the second operation, the loanable resources of the second operation are seconded to the first operation, and when the loan duration of the loanable resources expires, the second operation is suspended. The loanable resources of the operation are seconded to the first operation;

   When the second job ends, the loanable resource is scheduled to the first job to continue running the first job.
6. The method according to claim 3 or 4, characterized in that the second job set includes a second job, and when the job type of the first job is a termination type, the job according to the second job set The secondable resources determine the resource scheduling strategy of the first job, including:

   During the resource borrowing stage of the second operation, the loanable resources of the second operation are seconded to the first operation, and when the loan duration of the loanable resources expires, the second operation is terminated. The loanable resources of the operation are seconded to the first operation.
7. The method according to claim 3 or 4, characterized in that the second job set includes a second job, and when the job type of the first job is continuous, the job according to the second job set The secondable resources determine the resource scheduling strategy of the first job, including:

   During the resource borrowing stage of the second operation, the loanable resources of the second operation are seconded to the first operation, and when the loan duration of the loanable resources expires, the second operation is continued to be borrowed. The resources of the operation are seconded to the first operation.
8. A management node is characterized by including:

   The acquisition module is used to obtain the first job of resources to be scheduled;

   The first determination module is used to screen a job set of loanable resources in the cluster computing system. The job set includes at least one job. The at least one job is an allocated resource and can be seconded that matches the first job. Resource assignments;

   The second determination module is configured to determine the resource scheduling policy of the first job according to the loanable resources of the job set, where the resource scheduling policy is used to indicate that the loanable resources are used to perform all tasks during the resource loaning stage. Describe how the first job is handled.
9. The management node according to claim 8, characterized in that:

   The first determination module is specifically configured to filter the first job that matches the job type of the first job from the one or more jobs according to the resource secondment information of one or more jobs in the cluster computing system. A collection of jobs; where the resource The secondment information is used to indicate the type of operations supported by the loanable resources of a job when the loanable resources of the job are seconded to other jobs. The job types include: at least one of breakpoint resume type, termination type, or continuous type; and Screen a second set of jobs from the first set of jobs that can borrow resources to meet the resource requirements of the first job; the resource requirements include requirements for resource types and configuration requirements for the resource types, and the resource types include At least one of computing resources, storage resources or network resources.
10. The management node according to claim 8 or 9, characterized in that,

    The second determination module is specifically configured to determine the resource scheduling policy of the first job according to the loanable resources of the second job set.
11. The management node according to any one of claims 8 to 10, characterized in that,

    The loan duration of the loanable resource is less than or equal to the duration of the borrowable period of the resource.
12. The management node according to claim 10 or 11, characterized in that the second job set includes a second job, and when the job type of the first job is a breakpoint resume type;

    The second determination module is specifically configured to secondment the loanable resources of the second job to the first job during the resource borrowing stage of the second job, and when the loan duration of the second job is When it expires, suspend the loaning of the loanable resources of the second job to the first job; when the second job ends, schedule the loanable resources to the first job to continue running. The first job.
13. The management node according to claim 10 or 11, wherein the second job set includes a second job, and when the job type of the first job is a termination type;

    The second determination module is specifically configured to secondment the loanable resources of the second job to the first job during the resource borrowing stage of the second job, and when the loan duration of the second job is Upon expiration, the loaning of the loanable resources of the second operation to the first operation is terminated.
14. The management node according to claim 10 or 11, wherein the second job set includes a second job, and when the job type of the first job is continuous;

    The second determination module is specifically configured to secondment the loanable resources of the second job to the first job during the resource borrowing stage of the second job, and when the loan duration of the second job is When it expires, continue to second the resources of the second job to the first job.
15. A management node, characterized in that it includes a memory and at least one processor connected to the memory. The memory is used to store computer program code. The computer program code includes computer instructions. When the computer instructions are described When at least one processor is executed, the management node is caused to execute the method according to any one of claims 1 to 7.
16. A cluster computing system, characterized by including a management node and at least one computing node, the management node executing the method according to any one of claims 1 to 7.