Task processing method, apparatus, device, storage medium, and program product

Abstract

This application discloses a task processing method, apparatus, device, storage medium, and program product, relating to the field of industrial internet technology. The method includes: updating the global ready state of task sets with dependencies on each terminal at scheduling time slot boundaries; calculating the scheduling priority of tasks in the global ready state and sorting them by scheduling priority to obtain a ready task list; wherein the scheduling priority of a task defines the criticality of the ready task; traversing the ready task list to generate a set of actionable actions for each ready task in the list; selecting candidate tasks based on the set of actionable actions; and selecting an execution position from the set of actionable actions of the candidate tasks and initiating execution. This solution not only ensures that critical tasks are executed first, reducing the risk of critical tasks being blocked by non-critical tasks, but also eliminates rollback and reselection due to infeasible decisions, significantly improving the stability and reliability of task scheduling in industrial scenarios and reducing end-to-end task processing latency.

Description

Technical Field

[0001] This application relates to the field of industrial internet technology, and in particular to a task processing method, apparatus, equipment, storage medium, and program product. Background Technology

[0002] In industrial internet and smart manufacturing scenarios, production sites contain numerous industrial terminals and edge devices running applications highly sensitive to end-to-end response time and stability, such as online quality inspection, equipment health diagnostics, and adaptive process optimization. These industrial applications typically consist of multiple sub-tasks with dependencies, often described using a directed acyclic graph (DAG). Due to limitations in local terminal computing power, network link fluctuations, and edge server load variations, tasks are usually executed via a cloud-edge-device collaborative approach, offloading some or all sub-tasks from the terminal to edge servers or the cloud. Therefore, continuous online scheduling decisions are necessary during task scheduling.

[0003] Existing solutions typically employ list scheduling, HEFT (Heterogeneous Earliest Finish Time) heuristics for task ordering and mapping, or optimization methods when the system state is relatively stable, or online learning strategies such as reinforcement learning to model offloading decisions. Some methods use graph neural networks to improve the ability to represent task dependencies. The core decisions of these methods include determining the start order of ready tasks and allocating them to appropriate execution locations. When dealing with resource-constrained, highly dynamic scenarios such as industrial settings, these methods have the following significant drawbacks: Critical task chains are susceptible to congestion, and end-to-end latency is difficult to guarantee: Existing methods typically focus on evaluating the immediate cost (such as computation time and transmission time) of a single task at different locations during task scheduling, while lacking effective control over the startup order of multiple ready tasks. When links are congested or edge servers are heavily queued, a non-critical task may be prioritized due to its lower immediate cost, thereby preempting the communication or computing resources necessary for the critical task chain. This hinders the progress of the critical task chain, ultimately significantly increasing the end-to-end completion time of the entire application and posing a risk of timeouts in quality control or production cycle time.

[0004] The lack of feasibility guarantees in online decision-making can easily lead to latency jitter: Existing methods fail to adequately handle multiple feasibility constraints when making online decisions, easily resulting in unexecutable decisions and causing system instability. In dynamically changing industrial environments, task execution is limited by multiple hard constraints such as dependencies, terminal resource status, network reachability, and edge server concurrency capabilities. Scheduling decisions generated by existing solutions may be found infeasible during actual execution (e.g., attempting to unload a task onto an edge node that has reached its concurrency limit), forcing a rollback and re-decision. This "decision-rollback-reselection" process introduces unnecessary decision-making overhead and processing delays, significantly increasing system response latency jitter and reducing the stability of online scheduling.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this application is to provide a task processing method, apparatus, device, storage medium, and program product, which aims to solve the technical problems of insufficient control over the task initiation sequence in dynamic industrial environments, which leads to the easy blockage of critical task chains; and the lack of explicit compliance mechanism for scheduling decisions with hard constraints, which leads to infeasible decisions and thus causes rollback and latency jitter.

[0007] To achieve the above objectives, this application proposes a task processing method, which includes: At the time slot boundaries of each scheduling time slot, update the global ready state of the task sets with dependencies on each terminal; Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. A set of actionable actions for each ready task is generated by traversing and scanning the ready task list. Candidate tasks are selected from the ready tasks based on the set of actionable actions; From the set of actionable actions for the candidate tasks, select an execution location for the candidate task and start execution.

[0008] In one embodiment, the step of calculating the scheduling priority of tasks in the globally ready state includes: The critical chain urgency of the target application is calculated based on the remaining deadline of the target application within the scheduling slot and the lower bound of the estimated execution time of the unfinished critical chain of the target application; wherein, the target application is the application to which the task in the globally ready state belongs; The scheduling priority of the task is determined based on the critical chain urgency.

[0009] In one embodiment, the step of generating a set of actionable actions for each of the ready tasks by traversing and scanning the ready task list includes: A target task is selected from the list of ready tasks in sorted order by traversing and scanning. A set of action masks is predefined for each possible execution location action of the target task; the action mask includes multiple sub-masks, and each sub-mask corresponds to the satisfaction of a constraint condition; the constraint conditions include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints; the resource channel is an abstraction of the terminal's computing resources, and the resource channel includes a local computing channel and an uplink transmission channel for task offloading; The sub-masks of each constraint condition are combined and judged to determine the feasible execution location of the target task; Combine the feasible execution locations into a set of feasible actions for the target task; If the target task is not the last ready task in the ready task list, return and execute the step of selecting a target task from the ready task list in sorted order by traversing and scanning, until the target task is the last ready task in the ready task list.

[0010] In one embodiment, the step of selecting an execution location for a candidate task from the set of actionable actions of the candidate task and initiating execution includes: A global task graph is constructed based on the dependencies between the ready tasks in the ready task list; the global task graph contains explicit markers for the candidate tasks. The global task graph is subjected to bidirectional graph attention encoding to obtain the corresponding state representation; Based on the state representation, an execution location is selected for the candidate task from the set of actionable actions for the candidate task, and execution is initiated; the execution location is at least one of the terminal local, edge server, and cloud server.

[0011] In one embodiment, the step of performing bidirectional graph attention encoding on the global task graph to obtain the corresponding state representation includes: Extract the node features of each node in the global task graph; the node features include task size, computational intensity, application to which it belongs, application remaining time, terminal busy / idle flag, and candidate flag; The node features are subjected to forward graph attention aggregation to obtain forward graph attention features, and the node features are subjected to reverse graph attention aggregation to obtain reverse graph attention features; The forward graph attention features and the reverse graph attention features are concatenated and fused to obtain the corresponding state representation.

[0012] In one embodiment, after the step of selecting an execution location for the candidate task from the set of actionable actions of the candidate task and initiating execution, the method further includes: Update the resource channel status of the target terminal to which the candidate task belongs, and the resource status of the edge server corresponding to the target terminal, and remove the candidate task from the tasks in the globally ready state; The steps of returning and executing the steps of calculating the scheduling priority of the tasks in the global ready state, sorting the tasks according to the scheduling priority, and obtaining a list of ready tasks are repeated until no candidate tasks exist.

[0013] Furthermore, to achieve the above objectives, this application also proposes a task processing apparatus, the task processing apparatus comprising: The state update module is used to update the global ready state of the task set with dependencies on each terminal at the time slot boundary of each scheduling time slot. The priority sorting module is used to calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. The action filtering module is used to generate a set of actionable actions for each ready task by traversing and scanning the ready task list. The task selection module is used to select candidate tasks from the ready tasks based on the set of actionable actions. The action selection module is used to select an execution position for the candidate task from the set of actionable actions of the candidate task and start execution.

[0014] In addition, to achieve the above objectives, this application also proposes a task processing device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the task processing method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and which, when executed by a processor, implements the steps of the task processing method described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the task processing method described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: A time-slotted control approach is adopted. At the boundary of each scheduling time slot, the global ready state of the set of dependent tasks on the terminal is updated. The scheduling priority of tasks in the global ready state is calculated and used for task sorting to form a ready task list. This ensures that critical tasks with a greater impact on end-to-end latency are given higher scheduling priority. Subsequently, candidate tasks to be launched are selected from the ready task list based on the set of actionable actions. This ensures that high-priority critical tasks are evaluated and scheduled first, effectively solving the problem of critical task chains being blocked by non-critical tasks due to insufficient launch sequence control, and ensuring the timely progress of core business processes.

[0018] Furthermore, by generating a set of actionable actions for each task in the ready task list and selecting an execution position for candidate tasks within this set, an explicit hard constraint compliance mechanism is established. Through the set of actionable actions, all infeasible options that do not meet hard constraints such as resources, network, or dependencies can be eliminated in advance before the task scheduling decision. Therefore, the final selected execution position must be a valid decision that can be executed directly, fundamentally avoiding the occurrence of unexecutable decision results during task scheduling, thereby eliminating rollback and reselection caused by infeasible decisions.

[0019] The technical solution of this application provides a guarantee for avoiding infeasible decisions and ensuring scheduling stability. It can not only ensure that critical tasks are executed first and reduce the risk of critical tasks being blocked by non-critical tasks, but also eliminate the rollback and reselection caused by infeasible decisions. It significantly improves the stability and reliability of task scheduling in industrial scenarios and reduces the processing latency of end-to-end tasks. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1This is one of the schematic flowcharts of the task processing method provided in the embodiments of this application; Figure 2 This is the second schematic diagram of the task processing method provided in the embodiments of this application; Figure 3 This is a schematic diagram of the bidirectional graph attention encoding process provided in an embodiment of this application; Figure 4 This is a schematic diagram of the process for generating a set of possible actions provided in an embodiment of this application; Figure 5 This is a schematic diagram of the module structure of the task processing device according to an embodiment of this application; Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the task processing method in the embodiments of this application.

[0023] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0025] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0026] In the context of the Industrial Internet and smart manufacturing, there are numerous industrial terminals and edge devices on the production floor, such as industrial cameras and machine vision workstations, robot controllers, PLCs (Programmable Logic Controllers) and motion controllers, workstation IPCs (Industrial Personal Computers), AGV (Automated Guided Vehicle) onboard computing units, online inspection workstations, vibration and temperature monitoring sensors, and AR remote maintenance terminals. These field terminals are limited by size, power consumption, cost, and environmental adaptability, resulting in limited local computing power and heat dissipation capabilities. Furthermore, the load on industrial field network links (Industrial Ethernet, TSN (Time-Sensitive Networking), Industrial Wi-Fi, Industrial 5G private networks, etc.) and edge servers fluctuates over time.

[0027] Industrial applications such as online quality inspection, equipment health diagnosis, anomaly alarm and traceability, adaptive process optimization, collaborative robot safety protection, and digital twin online simulation and parameter calibration are sensitive to end-to-end response time and stability (long-tail latency). Once an occasional timeout occurs, it can cause cycle jitter, control loop lag, misjudgment and missed judgment, work order delay or line stoppage risks.

[0028] Therefore, in engineering implementation, the terminal executes some tasks locally or offloads them to the production line edge server based on the link conditions and edge computing resources. When edge resources are insufficient or centralized modeling and cross-domain collaboration are required, there is also a way to forward the tasks to the cloud via edge nodes.

[0029] Industrial applications are often not single tasks, but rather composed of multiple subtasks with dependencies, such as a pipeline of "data acquisition → preprocessing → feature extraction → inference and recognition → fusion decision → control output → execution feedback". A Directed Acyclic Graph (DAG) is commonly used to describe this dependency structure: once the current task is completed, the subsequent task enters a ready state. Therefore, continuous online scheduling is necessary. The core decisions in the scheduling process include: determining the task startup order from the set of ready tasks; and selecting the execution location (local, edge, or cloud) for the tasks to be started.

[0030] Existing solutions for online industrial scenarios have the following main shortcomings in handling industrial application tasks: Critical process and control chains are highly susceptible to congestion in resource-competitive scenarios: many existing solutions focus on comparing the immediate costs of executing individual tasks at different locations, with insufficient control over the priority of ready tasks. When link congestion or edge queuing is significant, if non-critical tasks preemptively occupy uplink or edge computing resources, it will directly hinder the progress of the critical chain, leading to a substantial increase in end-to-end completion time, and consequently causing problems such as delayed quality inspection results, lagging trajectory tracking, control loop jitter, and even timeouts.

[0031] Insufficient handling of feasibility constraints is the core reason for the poor stability of online decision-making: if "selecting a task + selecting an execution location" is treated as a joint action, the action space of task scheduling decisions will expand dramatically with the increase in the number of ready tasks and edge nodes; at the same time, it also needs to deal with multiple hard constraints such as dependency constraints, channel occupancy constraints, edge reachability constraints, and edge concurrent core count constraints. Without an explicit feasibility verification mechanism, online decisions may produce unexecutable actions, requiring rollback and reselection, which will exacerbate latency jitter and reduce the stability of online scheduling decisions.

[0032] Faced with the online arrival of tasks and the dynamic fluctuations in system status, the stability of existing strategies is clearly insufficient: there are many dynamic interference factors in industrial sites, such as batch switching, order insertion, sudden changes in equipment status, and network interference; the DAG structure of a task often only appears completely when the task arrives; the link rate and edge load also fluctuate continuously over time, and strategies based on static estimation or offline solution are prone to causing high latency problems at the tail in some periods.

[0033] The parallel capabilities of terminal-side computing and transmission are not fully utilized: Many industrial terminal platforms support the parallel execution of local computing and uplink transmission (e.g., simultaneous data acquisition and uploading, simultaneous preprocessing and transmission); if the scheduling model simplifies the terminal into a single execution channel, it will serialize the computing and transmission processes, introduce unnecessary waiting time, and exacerbate the blocking effect of the critical chain.

[0034] Based on this, this application provides a solution that employs a time-slotted online control framework, dividing time into fixed lengths. The scheduling time slots are driven by a ready queue: a complete scheduling loop is executed at the boundary of each time slot. The task start order is controlled by priority sorting, and the feasibility of decisions is guaranteed by pre-generating a set of actionable actions. Under dynamic link and edge load fluctuation conditions, it can ensure that the critical chain is prioritized and output directly executable unloading location decisions. At the same time, it reduces the risk of waiting propagation and high latency at the tail, thereby solving the latency jitter problem caused by critical task chain blocking and decision rollback.

[0035] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or task processing device capable of performing the above functions. The following description uses a task processing device as an example to illustrate this embodiment and the subsequent embodiments.

[0036] Based on this, this application provides a task processing method and an online scheduler running in an industrial internet environment. The online scheduler can be deployed on an edge server or a centralized control node.

[0037] Specifically, refer to Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the task processing method of this application.

[0038] In this embodiment, the task processing method includes steps S10 to S50: Step S10: At the time slot boundary of each scheduling time slot, update the global ready state of the task set with dependencies on each terminal. Step S20: Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. Step S30: Generate a set of actionable actions for each ready task by traversing and scanning the ready task list; Step S40: Select candidate tasks from the ready tasks according to the set of actionable actions; Step S50: Select an execution location for the candidate task from the set of actionable actions of the candidate task and start execution.

[0039] First, at the boundary of each scheduling slot, the global ready state of the sets of tasks with dependencies on each terminal needs to be updated. That is, the ready state of all tasks on each terminal is updated to obtain the global ready set. It should be noted that a task only enters the ready state and meets the conditions for scheduling execution when all its predecessor tasks (i.e., the tasks it depends on) have been completed.

[0040] As a feasible implementation method, a time-driven incremental update mechanism is adopted, which updates the task readiness state by maintaining a global task graph. Specifically, the global task graph is constructed based on tasks in a set of tasks with dependencies. Tasks are used as nodes, and directed edges between nodes represent the dependencies between tasks. When a task is completed, the terminal sends an event indicating task completion. After detecting the task completion event sent by the terminal, the system iterates through the direct successor tasks of the completed task, checks whether all predecessors of these successor tasks have been completed, and if so, marks the status of the successor task as "ready" and adds it to the global ready set, thus completing the update of the global ready state.

[0041] Additionally, it should be noted that the global ready state is a dynamic Boolean set of task states that records all tasks on each terminal that can be scheduled immediately at the start of time slot t. Its source is the satisfaction of the dependencies between tasks, and it serves as the input for subsequent scheduling decisions.

[0042] A set of tasks with dependencies is typically described in the form of a directed acyclic graph (DAG). In one embodiment, each application corresponds to a DAG, whose data includes task nodes, task computation / data volume attributes, and dependency edges representing the execution order. Its purpose is to define the execution constraints between tasks and is the fundamental basis for determining the ready state.

[0043] After updating the global ready state of the task set on the terminal and obtaining the global ready set, priorities need to be assigned to the tasks in the global ready set to determine the order in which the tasks are started. This is done by calculating the scheduling priority of the tasks in the global ready state and sorting the tasks according to that priority to obtain the ready task list.

[0044] In this context, the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. The higher the priority, the greater the criticality of the ready task, and the greater the probability that the ready task will be started and executed first. As an example, the criticality of a ready task is used to characterize its impact on the application completion time.

[0045] Scheduling priority is a numerical metric used for ranking. Its calculation depends on a specific prioritization strategy; that is, it calculates the scheduling priority of tasks in the globally ready state according to a preset priority strategy. The scheduling priority of ready tasks is used to quantify the criticality of tasks, thereby guiding the scheduler to prioritize tasks that have a greater impact on the overall application completion time or system goals, thus alleviating critical chain congestion from a ranking perspective.

[0046] Furthermore, by traversing the list of ready tasks, each ready task is scanned to generate a set of actionable actions for each ready task. This set of actionable actions contains the actual executable positions that the corresponding ready task can take in the current scheduling slot, thereby eliminating infeasible execution positions in advance and avoiding the generation of infeasible decisions.

[0047] In other words, the actionable set is a subset of all possible execution locations of a ready task. This subset only includes execution locations that satisfy the resource and constraint conditions in the current scheduling slot. The generation of the actionable set depends on querying and simulating the current resource state. The role of the actionable set is to construct a feasible solution space for execution location selection, restricting subsequent task scheduling decisions to this space and fundamentally preventing the selection of unexecutable actions.

[0048] As one embodiment, a set of actionable tasks for ready tasks is generated through simulation and filtering of a resource reservation table. The resource reservation table includes a terminal resource table and an edge node resource table. The terminal resource table records the local resource occupancy status of each terminal, including the remaining occupancy time of local computing and uplink transmission channels. The edge node resource table records the resource occupancy status of edge servers, specifically including the remaining available computing cores of the edge nodes.

[0049] When traversing the list of ready tasks, for each ready task, simulate local execution, execution offloaded to an edge node, and execution offloaded to the cloud.

[0050] Simulate local execution and check whether the resource occupancy status of the local computing channel of the terminal to which the ready task belongs is idle (i.e., whether the remaining occupancy time is 0). If the local computing channel is idle, the local computing can be added to the set of actionable actions as a candidate for a feasible execution location.

[0051] The simulation involves offloading the execution to an edge node, checking whether the uplink transmission channel of the terminal is idle, and whether the edge node has any remaining computing cores. If both conditions are met, the execution at the edge node is added to the set of possible actions as a candidate for a feasible execution location.

[0052] Simulate offloading to the cloud for execution, checking if the uplink transmission channel of the terminal to which the ready task belongs is idle and if there is at least one reachable edge node (for forwarding the task to the cloud). If the conditions are met, the execution in the cloud will be added to the set of actionable actions as a candidate for a feasible execution location, and a default or optimal edge node will be associated for forwarding and offloading the ready task.

[0053] Based on the actionable sets of each generated ready task, a ready task is selected as a candidate task from the ready task list. In one feasible implementation, the candidate task is the ready task with a non-empty actionable set and the highest scheduling priority. That is, by traversing the ready task list, the ready task with the highest priority and a non-empty actionable set is selected as the candidate task according to the scheduling priority.

[0054] Further, from the set of actionable actions of the candidate task, an execution position is selected for the candidate task, and execution of the candidate task is initiated. In one embodiment, by traversing and scanning the list of ready tasks, a set of actionable actions for each ready task is generated. This process is repeated for each ready task's set of actionable actions. Based on this, according to the scheduling priority order of the ready tasks in the list, the ready task with the highest priority and a non-empty set of actionable actions is selected as the candidate task to be executed. An execution position is then selected for the candidate task from its set of actionable actions, and execution of the candidate task is initiated.

[0055] In another embodiment, steps S30-S50 can be executed cyclically, with the selection of a candidate task or the completion of scanning all ready tasks in the ready task list as the end condition for traversal, thereby realizing the cyclic dispatch of ready tasks. Specifically, refer to... Figure 2 , Figure 2 This is another flowchart illustrating the task processing method provided in the embodiments of this application. Figure 2 The task processing methods shown include: At the time slot boundaries of each scheduling time slot, update the global ready state of the task sets with dependencies on each terminal; Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. A set of actionable actions for the currently scanned ready tasks is generated by traversing and scanning the ready task list. If the set of available actions is not empty, the currently scanned ready tasks are taken as candidate tasks, and an execution position is selected for the candidate task from the set of available actions of the candidate tasks and execution is started. If the set of actionable actions is empty, return and execute the step of generating the set of actionable actions for the currently scanned ready task by traversing and scanning the ready task list, until the set of actionable actions is not empty, or the currently scanned ready task is the last task in the ready task list.

[0056] like Figure 2 As shown, after obtaining the list of ready tasks, each ready task in the list is traversed by scanning, and a set of actionable actions is generated based on the currently traversed ready task. If the generated set of actionable actions is not empty, the currently traversed ready task is taken as a candidate task, and an execution position is selected from the generated set of actionable actions to start and execute the ready task.

[0057] Understandably, if the set of available actions for the currently traversed ready task is empty, the process continues to traverse the next ready task, repeating the above process until a ready task with a non-empty set of available actions is scanned. At this point, the task is initiated as a candidate task, and the traversal ends. If no candidate task meeting the criteria is selected after traversing all ready tasks in the ready task list, the traversal and task dispatch for the current scheduling slot end, and the process moves to the next scheduling slot.

[0058] In this embodiment, a time-slotted control approach is adopted. At the boundary of each scheduling time slot, the global ready state of the set of dependent tasks on the terminal is updated. The scheduling priority of tasks in the global ready state is calculated and used for task sorting to form a ready task list. This ensures that critical tasks with a greater impact on end-to-end latency are given higher scheduling priority. Subsequently, candidate tasks to be launched are selected from the ready task list based on the set of actionable actions. This ensures that high-priority critical tasks are evaluated and scheduled first, thereby effectively solving the problem of critical task chains being blocked by non-critical tasks due to insufficient launch order control, and ensuring the timely progress of core business processes.

[0059] Furthermore, by generating a set of actionable actions for each task in the ready task list and selecting an execution position for candidate tasks within this set, an explicit hard constraint compliance mechanism is established. Through the set of actionable actions, all infeasible options that do not meet hard constraints such as resources, network, or dependencies can be eliminated in advance before the task scheduling decision. Therefore, the final selected execution position must be a valid decision that can be executed directly, fundamentally avoiding the occurrence of unexecutable decision results during the task scheduling process, thereby eliminating rollback and reselection caused by infeasible decisions.

[0060] The technical solution of this application provides a guarantee for avoiding infeasible decisions and ensuring scheduling stability. It can not only ensure that critical tasks are executed first and reduce the risk of critical tasks being blocked by non-critical tasks, but also eliminate the rollback and reselection caused by infeasible decisions. It significantly improves the stability and reliability of task scheduling in industrial scenarios and reduces the processing latency of end-to-end tasks.

[0061] In one feasible implementation, step S20 may include steps S201-S202: Step S201: Calculate the critical chain urgency of the target application based on the remaining deadline of the target application within the scheduling slot and the estimated execution time of the lower bound of the unfinished critical chain of the target application; wherein, the target application is the application to which the task in the globally ready state belongs; Step S202: Determine the scheduling priority of the task based on the critical chain urgency.

[0062] When calculating the scheduling priority of tasks in the globally ready state, the critical chain urgency of the target application is calculated based on the remaining deadline of the target application within the scheduling slot and the lower bound of the estimated execution time of the unfinished critical chain of the target application. The scheduling priority of tasks in the globally ready state is then determined based on this critical chain urgency.

[0063] The target application is the application to which the task belongs in the globally ready state.

[0064] In specific implementations, the critical chain urgency used to determine scheduling priority can be calculated in different ways. One feasible implementation is to first calculate the remaining deadline for the target application in the current scheduling slot, i.e., subtract the current time from a preset absolute deadline. Second, estimate the shortest execution time required for the incomplete critical path of the application, i.e., the lower bound estimated execution time. Finally, subtract the lower bound estimated execution time from the remaining deadline; the resulting critical chain slack characterizes the urgency of the target application: a smaller slack indicates less time buffer and a more urgent application.

[0065] For example, the calculation of critical chain urgency is defined as follows: The remaining deadline for application i in the current scheduling slot t for: (1) The lower bound of the estimated execution time for the incomplete critical link of application i is: ; Application of key chain slack As shown in Formula 2: (2) In formula 1-2, This is the deadline for applying i pre-configuration. Indicates the current moment. It is the duration of the time slot corresponding to the scheduling time slot. It is a time slot index.

[0066] Key chain slack It is the core indicator for measuring the critical chain urgency of application i, and intuitively represents the difference between the remaining available time of application i and the minimum required time. When the critical chain slack is greater than 0, it means that the application still has a buffer time in theory; when the critical chain slack is less than or equal to 0, it means that even under ideal conditions, the application has faced or exceeded its implementation, and the urgency is extremely high.

[0067] As another feasible implementation, critical chain urgency can be characterized by urgency weight values. For example, a weighted sum of the reciprocal of the remaining deadline and the lower bound of the estimated execution time can be obtained to obtain a comprehensive urgency value. The larger this value, the more urgent the application, and this approach provides flexibility in incorporating different urgency factors.

[0068] When determining the scheduling priority of specific tasks based on critical chain urgency, as an example, the application's critical chain slackness is used as the first sorting key for ascending order, ensuring that the most urgent application's tasks receive the highest priority. Then, for the same application, tasks are sorted using their level or rank in the dependency graph as the second sorting key, typically prioritizing tasks that are earlier in the hierarchy and can advance the critical chain, thereby optimizing the task scheduling order at both the global and local levels.

[0069] In another embodiment, the calculated critical chain urgency value (whether it is a slackness or urgency weight) is directly assigned to each task in the application's globally ready state as its scheduling priority value. Subsequently, all assigned tasks are globally sorted according to this scheduling priority value, with tasks having higher scheduling priority values ​​having higher priority, or vice versa. The specific configuration can be made according to actual needs and is not limited here.

[0070] Among them, the remaining deadline of the application is a time constraint target that can be used to dynamically measure the urgency of the application's completion and serves as the basic time reference for calculating urgency.

[0071] The lower bound execution estimated time is an estimate of the shortest time required to complete the remaining critical path of the application. This estimate depends on the computational workload, data volume, and best-case assumptions regarding the system's resource processing capabilities. It assesses urgency from a workload perspective, representing the theoretically fastest time required for the application to complete the remaining tasks, and is a core benchmark for determining whether a timeout is possible.

[0072] Critical chain slack is calculated from the application's remaining deadline and lower bound execution estimate. It can intuitively reflect the application's "available time" versus "minimum required time" and is used to quantify the application's timeout risk. For example, a negative or zero slack means that the application is at risk of timeout, thus requiring higher priority attention to tasks in the application's globally ready state.

[0073] By using the urgency of the application's critical chain as the basis for calculating task scheduling priority, the scheduling decision for tasks can be made from the perspective of the global application completion time, prioritizing the most urgent and critical application links to obtain computing and communication resources, thereby effectively mitigating the risk of critical task chains being blocked.

[0074] In one feasible implementation, step S30 may include steps S301 to S305: Step S301: Select a target task from the ready task list in sorted order by traversing and scanning. Step S302: Predefine a set of action masks for each possible execution location action of the target task; the action mask includes multiple sub-masks, and each sub-mask corresponds to the satisfaction of a constraint condition; the constraint conditions include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints; the resource channel is an abstraction of the terminal's computing resources, and the resource channel includes a local computing channel and an uplink transmission channel for task offloading; Step S303: Combine and judge the sub-masks of each constraint condition to determine the feasible execution location of the target task; Step S304: Combine the feasible execution locations into a set of feasible actions for the target task; Step S305: If the target task is not the last ready task in the ready task list, return and execute the step of selecting a target task from the ready task list in sorting order by traversing and scanning, until the target task is the last ready task in the ready task list.

[0075] In a specific implementation, the core of generating a set of actionable actions lies in systematically excluding execution location options that do not meet hard constraints through a structured action masking mechanism, thereby ensuring the absolute feasibility of subsequent scheduling decisions.

[0076] As a feasible implementation method, a decomposed action mask generation mechanism is adopted. By traversing and scanning, a target task is selected sequentially from the list of ready tasks, and an action mask is predefined for each possible execution location action of that target task. These multiple action masks include multiple sub-masks, each sub-mask corresponding to the satisfaction of a constraint condition. These constraints include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints.

[0077] The resource channel is an abstraction of the computing resources of the terminal to which the task belongs, specifically including the local computing channel and the uplink transmission channel used for task unloading.

[0078] The sub-masks of each constraint are combined and judged to determine the feasible execution position of the target task, and the feasible execution position is combined into a set of feasible actions for the target task. If the current target task is not the last ready task in the ready task list, the process returns and executes step S301. By traversing and scanning, a target task is selected again from the ready task list in sequence until the current target task is the last ready task in the ready task list, thus obtaining the set of feasible actions for each ready task.

[0079] In one embodiment, each action mask is composed of multiple sub-masks combined through a logical AND operation. Each sub-mask corresponds to the satisfaction of a hard constraint. By performing a logical AND operation on the sub-masks of each constraint, it is determined whether the execution position corresponding to the action mask is an actionable action. Each possible execution position action of the target task includes local execution, edge node execution, and cloud execution.

[0080] Specifically, the sub-masks of the predefined action mask for the target task include: Task dependency constraint submask: Used to indicate whether all predecessor tasks of the target task have been completed. This constraint submask is typically used to update the global ready state of the terminal task. For a ready task, this submask is always true, indicating that all predecessor tasks of the ready task have been completed.

[0081] Resource channel occupancy constraint submask: Related to the execution location, it is used to characterize the occupancy status of the terminal's resource channels, thereby determining availability. Specifically, if the action is executed locally, it is necessary to check whether the local computing channel of the terminal to which the target task belongs is idle; if the action is executed offloaded, it is necessary to check whether the uplink transmission channel of the terminal is idle.

[0082] In one embodiment, the computing resources of each terminal are abstracted into two resource channels: a local computing channel and an uplink transmission channel. The remaining busy time of the two resource channels is maintained at the time slot t boundary. : ; At the same time, maintain the start flag of the resource channel in time slot t. : .

[0083] Specifically, issuing a task start command that occupies the corresponding resource channel at the time slot t boundary indicates that the corresponding resource channel is started once. During the task start execution, the corresponding resource channel is in an occupied state, which is represented by a start flag.

[0084] If and only if and At that time, the local computing channel is in the time slot It can be started; if and only if and At that time, the uplink transmission channel is in the time slot It can be started. That is, for the local computing channel and the uplink transmission channel, it can be started if and only if the remaining busy time is 0 and the start flag in time slot t indicates that it is in an available state (or an unoccupied state).

[0085] It should be noted that each terminal's two resource channels can launch a task at most once within the same time slot. Therefore, the maximum number of launches within any time slot t is... satisfy: , The number of terminals. Furthermore, the local computing channel and the uplink transmission channel of a terminal are two independent resource channels. For the same terminal, its local computing channel and uplink transmission channel can proceed in parallel when conditions permit, thereby reducing unnecessary waiting latency and improving end-to-end latency stability.

[0086] Edge reachability constraint submask: For actions involving edge servers, if the action is executed on edge server k, or executed in the cloud but needs to be forwarded through edge server k, it is necessary to check whether edge server k is located in the set of reachable edges M(i,t) of the terminal in the current time slot.

[0087] Resource capacity constraint submask for edge servers: For actions executed on edge server k, it is necessary to check whether the number of tasks currently being executed concurrently on edge server k is less than its total available concurrent cores C(k). This submask is always true for actions executed locally or in the cloud.

[0088] The final mask of an action is true only when all relevant sub-masks for an action at a given execution location are true, meaning all constraints are met. Only then is the execution location deemed a feasible execution location. The set of all feasible execution locations constitutes the set of feasible actions for the target task. This process executes each ready task in the ready task list sequentially until all ready tasks have been traversed.

[0089] In this system, the action mask and sub-mask are Boolean identifiers used to ultimately allow or disable an execution location option. The sub-mask is the basic unit that constitutes the action mask, and each type of sub-mask corresponds to an independent decision result for a system hard constraint. The action mask decomposes the complex multi-constraint feasibility determination problem into multiple independent, parallel-computable decisions, which are then combined using logical AND operations, simplifying the decision-making process and ensuring both completeness and efficiency.

[0090] Task dependency constraints are used to ensure that task scheduling conforms to application logic and are a manifestation of DAG dependencies; resource channel occupancy constraints are used to ensure that at most one task is executing on any specific channel (local computing channel or uplink transmission channel) of a terminal at any given time, preventing resource conflicts; edge reachability constraints are determined by network topology and link quality, ensuring that scheduling decisions conform to physical connection limitations; edge server resource capacity constraints are used to ensure that the total number of tasks scheduled to the edge server does not exceed its processing capacity limit, avoiding overload.

[0091] These constraints collectively define the boundary of the scheduling solution space, and the process of generating the set of feasible actions is the process of finding feasible solutions that satisfy all constraints at the current moment. By introducing a structured action masking mechanism to explicitly encode and jointly judge various system hard constraints, it is possible to accurately and efficiently filter out all currently executable feasible action positions for each ready task, fundamentally avoiding invalid decisions that violate dependencies, conflict with resources, or exceed load capacity. This eliminates the overhead of rollback and reselection caused by infeasible decisions, significantly improving the stability and real-time performance of online scheduling.

[0092] In some embodiments, when steps S20 to S50 are executed cyclically, for the selected target task, after obtaining its set of possible actions, it is determined whether the target task is a candidate task based on whether the set of possible actions is empty. If it is a candidate task, an execution position is selected from the set of possible actions, and the candidate task is started and executed. If it is not a candidate task, the next ready task is selected from the ready task list in sorted order by traversing and scanning, and the above process is repeated until a candidate task is scanned, or the target task is the last ready task in the ready task list.

[0093] The specific implementation method is shown in the following steps: At the time slot boundaries of each scheduling time slot, update the global ready state of the task sets with dependencies on each terminal; Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. A target task is selected from the list of ready tasks in sorted order by traversing and scanning. A set of action masks is predefined for each possible execution location action of the target task; the action mask includes multiple sub-masks, and each sub-mask corresponds to the satisfaction of a constraint condition; the constraint conditions include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints; the resource channel is an abstraction of the terminal's computing resources, and the resource channel includes a local computing channel and an uplink transmission channel for task offloading; The sub-masks of each constraint condition are combined and judged to determine the feasible execution location of the target task; Combine the feasible execution locations into a set of feasible actions for the target task; If the set of possible actions is not empty, then the target task is taken as a candidate task, an execution position is selected for the candidate task from the set of possible actions, and execution is started; If the set of available actions is empty, then return and execute the step of selecting a target task from the list of ready tasks in sorted order by traversing and scanning, until the target task is the last ready task in the list of ready tasks.

[0094] In one feasible implementation, step S50 may include steps S501 to S503: Step S501: Based on the dependencies between each ready task in the ready task list, construct a global task graph; the global task graph is set with explicit markers for the candidate tasks. Step S502: Perform bidirectional graph attention encoding on the global task graph to obtain the corresponding state representation; Step S503: Based on the state representation, select an execution location for the candidate task from the set of actionable actions of the candidate task and start execution; the execution location is at least one of the terminal local, edge server and cloud server.

[0095] In a specific implementation, global dependency information is encoded based on graph neural networks to optimize intelligent decision-making for execution location selection. By constructing and analyzing a global task graph, complex dependencies between tasks are captured, thereby selecting a better execution location for candidate tasks.

[0096] Specifically, the selection of execution location is implemented based on the encoding and decision-making of a bidirectional graph attention network. First, all arrived and incomplete task instances and their dependencies are integrated to construct a unified global task graph: G(t) = (V(t), E(t)), where nodes V(t) represent task instances and directed edges E(t) represent dependencies between tasks. To focus on the current decision object, an explicit flag is set for the current candidate task in the node's feature vector; for example, one field in its feature vector is set to 1, while other tasks are set to 0. Next, the global task graph is processed through bidirectional graph attention encoding to obtain the corresponding state representation.

[0097] The bidirectional attention encoding of the global task graph involves information propagation and aggregation on the task-dependent edge E(t) and its reverse edges. Specifically, forward propagation aggregates information from predecessor tasks, representing the task's readiness conditions and history; backward propagation aggregates information from successor tasks, representing the task's downstream impact and urgency. Through multi-layered bidirectional attention aggregation, each task node ultimately obtains a state representation that integrates its local and global context information. Finally, the state representations of candidate task nodes are input into a policy network, such as a multilayer perceptron (MLP). This policy network outputs a probability distribution or Q-value distribution on the set of possible actions for the candidate task, based on which the execution position is selected and execution is initiated, such as selecting the action with the highest probability as the execution position.

[0098] A global task graph is a dynamic data structure where nodes represent all pending task instances, and directed edges represent data dependencies or execution order constraints between tasks. Its purpose is to integrate the scattered task dependencies within a terminal into a unified, machine-readable topological representation, providing input for subsequent graph-based intelligent decision-making. The global task graph encodes the contextual information of tasks and is fundamental to understanding the relationships between them.

[0099] It should be noted that the global task graph is constructed from the application dimension. Based on the dependency relationship of ready tasks in the same application in the ready task list, the global task graph of the application is constructed. The readiness and completion of task instances proceed independently according to their respective applications, without introducing cross-terminal synchronous completion semantics.

[0100] The explicit labeling of candidate tasks is used in the node feature vector of the global task graph to uniquely identify the current task to be decided (i.e., the candidate task). This enables the graph encoding network and the subsequent policy network to clearly distinguish the subject of the current decision from other tasks in the graph, thereby ensuring that the learned state representation and the final decision are focused on the current candidate task.

[0101] Bidirectional graph attention encoding is a specific operation in graph neural networks. Bidirectional encoding refers to the simultaneous propagation of information along the forward and backward directions of dependency edges to capture both the preceding and subsequent dependencies of a task. Graph attention, in information aggregation, does not treat all neighboring nodes equally, but rather assigns different weights to different neighboring nodes through a learnable attention mechanism, thereby modeling the differences in importance between tasks more precisely. The goal is to transform information such as topology and task features into a low-dimensional, dense, and semantically rich state representation.

[0102] The state representation obtained through bidirectional graph attention encoding is a vectorized representation of the candidate task's state in the global task graph. It encodes information such as the candidate task's attributes, the status of its dependent tasks, the status of its blocked successor tasks, and its criticality in the entire task flow. This state representation serves as input to the reinforcement learning policy network or heuristic evaluation function, providing a high-level decision-making basis for the intelligent selection of execution positions. The goal is to select execution positions that optimize overall application completion time and other metrics.

[0103] Furthermore, based on the encoded state representation, an execution location is selected for the candidate task from the set of actionable actions, and the candidate task is then executed. The selected execution location can be at least one of local execution on the terminal, execution on an edge server, or execution on a cloud server.

[0104] The system selects the execution location for candidate tasks from local, edge, or cloud, and masks non-executable actions through a set of actionable actions or an action mask, so that output actions can be directly sent for execution and avoid latency jitter caused by backtracking and reselection. When multiple edge nodes meet hard constraints at the same time, the system makes a preference selection based on the edge queuing waiting degree to reduce the risk of queuing tailing and improve tail latency.

[0105] By constructing a global task graph and using graph neural network techniques such as bidirectional graph attention to extract state representations rich in dependencies, the decision-making for the selection of execution location can transcend the local perspective of a single task, taking into account its position and influence in the overall task dependency graph, thus potentially making better scheduling decisions that are beneficial to the global goal.

[0106] In one embodiment, step S502 may further include steps A1 to A3: Step A1: Extract the node features of each node in the global task graph; the node features include task size, computational intensity, application to which it belongs, application remaining time, terminal busy / idle flag, and candidate flag; Step A2: Perform forward graph attention aggregation on the node features to obtain forward graph attention features, and perform reverse graph attention aggregation on the node features to obtain reverse graph attention features; Step A3: The forward graph attention features and the reverse graph attention features are concatenated and fused to obtain the corresponding state representation.

[0107] In a specific implementation, the bidirectional graph attention encoding of the global task graph first extracts the node features of each node in the global task graph. Nodes in the global task graph are task instances, and node features are task characteristics or task information. In this embodiment, each task node in the global task graph is constructed with a feature vector. This feature vector integrates the multi-dimensional information of the node. Specific node features include: task scale, such as computational load (cycles) and data transmission volume (bits); computational intensity, characterizing the task's sensitivity to computational resources; the application to which it belongs, characterized by an application identifier used to associate application-level attributes (such as application remaining deadline); application remaining time, i.e., the critical chain slack or remaining deadline of the application to which the task belongs, reflecting the overall urgency of the application; terminal busy / idle flag, indicating the busy / idle status of the local computation and uplink transmission channels of the terminal to which the task belongs; and candidate flags, which can be one-hot explicit flags used to identify or characterize candidate tasks. For example, only the current candidate task node is marked as 1, and the remaining task nodes are marked as 0. The node features collectively characterize the individual attributes, resource requirements, and dynamic context of the corresponding task.

[0108] Then, bidirectional graph attention aggregation is performed based on the extracted node features, such as... Figure 3 As shown, bidirectional graph attention encoding includes forward graph attention aggregation and reverse graph attention aggregation. Forward graph attention aggregation specifically follows the direction of dependency edges, from the predecessor node to the successor node, using a Graph Attention Network (GAT) layer for information propagation. Each node aggregates the features of all its predecessor nodes, and the attention weights are dynamically calculated based on the feature similarity between the predecessor and current nodes, thus obtaining forward graph attention features that encapsulate the historical execution context. Reverse graph attention aggregation specifically follows the reverse direction of dependency edges, from the successor node to the predecessor node, also using a GAT layer. Each node aggregates the features of all its successor nodes, dynamically calculating the attention weights, thus obtaining reverse graph attention features that reflect the impact and urgency of downstream congestion.

[0109] Finally, the forward and reverse graph attention features are concatenated and fused to obtain the corresponding state representation. For each node in the graph, its learned forward and reverse graph attention feature vectors are concatenated to form a fusion vector with doubled dimensions. This fusion vector is the final state representation of the node, which simultaneously encodes the completion status of upstream dependencies and downstream critical pressures, providing comprehensive topological context information for decision-making.

[0110] The extracted node features are digital descriptions of each task node in the graph. Static attributes such as task size and computational intensity are derived from the task's definition or historical performance analysis data, used to estimate the task's execution overhead. The application it belongs to and the remaining application time provide application-level urgency context. Terminal busy / idle status can be maintained by monitoring the terminal's resource channel status. The determination is a direct manifestation of dynamic resource constraints; candidate labels can be dynamically set when constructing the current decision context.

[0111] In some embodiments, such as Figure 3 As shown, the encoded state representation can be a graph representation. The encoded output of this diagram represents This is used for decision-making regarding the execution location.

[0112] By defining specific node features and a structured feature fusion process for bidirectional attention aggregation, the abstract bidirectional graph attention encoding is transformed into practically operable steps. This enables the automatic and efficient extraction of deep features from complex global task dependency graphs, laying the foundation for generating high-quality execution location decisions.

[0113] In one feasible implementation, after step S50, steps S601-S602 may also be included: Step S601: Update the resource channel status of the target terminal to which the candidate task belongs, and the resource status of the edge server corresponding to the target terminal, and remove the candidate task from the tasks in the global ready state. Step S602: Return and execute the step of calculating the scheduling priority of the tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a list of ready tasks, until the candidate tasks no longer exist.

[0114] After selecting an execution location for a candidate task and initiating execution, an immediate status update is performed: First, the resource channel status of the target terminal to which the candidate task belongs is updated. Specifically, based on the selected execution location of the candidate task, the remaining busy time of the corresponding resource channel of the terminal is increased. If the candidate task is executed locally on the terminal, the estimated execution time of the task is added to the local remaining busy time; if the task is offloaded for execution, the estimated uplink transmission time of the task is added to the local remaining busy time. Simultaneously, the start flag 'b' of the corresponding resource channel is set to a characteristic value representing channel occupancy, indicating that the channel is already occupied in this time slot.

[0115] Secondly, if a task is offloaded to a specific edge server k for execution, the resource status of that edge server needs to be updated, typically by incrementing its current concurrent task count counter. Finally, the candidate task is removed from the list of tasks in the globally ready state, as it has already been scheduled and started execution and is no longer in the ready state.

[0116] After updating the resource channels and resource status, return to and re-execute step S20 to recalculate the scheduling priorities of tasks in the globally ready state and sort the globally ready tasks by priority to obtain a new list of ready tasks. At this point, since both the globally ready tasks and resource status have been updated, the priorities of the globally ready tasks will be recalculated, and a new set of actionable actions for ready tasks will be generated. An attempt will then be made to start the next candidate task. This loop continues until, in a certain iteration, no candidate task with a non-empty set of actionable actions can be found, or the list of ready tasks is empty (i.e., no candidate task exists). At this point, task dispatch for the current time slot ends, and the process waits for the next time slot boundary.

[0117] Updating resource channel status is the cornerstone of maintaining accurate task scheduling. The operation targets the abstract local computing channel and uplink transmission channel of each terminal, and the data to be updated is the terminal's remaining busy time and the current time slot start flag. Its purpose is to: record resource occupancy; the increase in remaining busy time quantifies the busyness of the corresponding resource channel in the future, used for determining the availability of subsequent time slots; and implement mutual exclusion within time slots, setting the current time slot start flag b to ensure that at most one task is started on each resource channel within the same time slot. This is a key constraint for achieving finite-step iteration and avoiding resource conflicts.

[0118] Updating the resource status of edge servers typically refers to updating the number of available concurrent computing cores. When a task is scheduled or offloaded to an edge server for execution, its available core count decreases accordingly. Updating the resource status of the outermost edge servers can reflect their remaining processing capacity in real time, ensuring that subsequent scheduling decisions adhere to their resource capacity constraints and preventing overload.

[0119] By looping task dispatch and execution within the same time slot, as many tasks as possible can be started within a time slot, filling newly released resource gaps, avoiding resource idleness, and thus maximizing resource utilization within the time slot.

[0120] Furthermore, initiating a task may decouple multiple subsequent tasks, immediately making them ready and enabling a chain reaction of decision-making. Reordering after each iteration ensures that newly ready high-priority tasks are considered promptly, thereby accelerating the critical chain and reducing the risk of blockages in critical links.

[0121] By establishing a state update mechanism after task initiation and a cyclical mechanism for task dispatch and initiation, a closed-loop, self-consistent online scheduling state machine is constructed. This not only ensures the real-time accuracy of resources but also drives the scheduler to actively and orderly utilize available resources within a time slot through the initiation-update-re-decision cycle, thereby significantly improving system throughput and response efficiency.

[0122] The following example illustrates in detail the task processing method provided in this application.

[0123] Specifically, the task processing method provided in this application adopts a time-slotted control approach, making task scheduling decisions at the boundary of each time slot. After a task starts, it runs in a non-preemptive manner until completion. At the boundary of each scheduling time slot t, the online scheduler executes the following process: 1. Update the global ready task set. For the application of terminal i, a Directed Acyclic Graph (DAG) is used to represent dependencies. Nodes are task instances T(i,j), and edges represent predecessor constraints and data dependencies. A task instance is only allowed to enter the ready task set R(t) and start after all its predecessor task instances have been completed. The ready condition can be expressed as: the ready condition is satisfied if and only if for any predecessor task T(i,p), the following conditions are met. , Where TF(i,p) represents the completion time of task T(i,p); This represents the boundary time of time slot t.

[0124] 2. Initialize the start flag of the terminal resource channel in this time slot: For each terminal i, initialize the start flag of its local computing channel in time slot t. and the start flag of the uplink transmission channel. ; 3. Regarding The task is dynamically prioritized and sorted to form a list of ready tasks. ; 4. From The list header begins scanning for ready tasks one by one. ; 5. Generate ready tasks using an action masking mechanism. Set of possible actions ); 6. If If empty, continue scanning for the next ready task; if not empty, then scan for the ready task. Set as a candidate task End of scan.

[0125] 7. If no candidate task is found after the scan, the task dispatch for this time slot ends, and the process proceeds to the next time slot; for candidate tasks... In its set of actions Under the constraints, select and execute the position action and initiate it; 8. After the task starts, update the terminal's remaining busy time and the resource channel's start flag, and... from Remove from; 9. After completing the update in step 8, return to step 3 to re-sort the priorities and regenerate the action set according to steps 4-5. Continue the iterative dispatch of ready tasks in the same time slot until step 6-7 triggers the iteration termination.

[0126] When generating the set of possible actions for ready tasks, candidate tasks The execution location action set A is defined as: Local, Cloud, Edge(1), and Edge(K). Local represents local execution; Edge(k) represents execution offloaded to edge server k; and Cloud represents execution on the cloud. If Cloud needs to be forwarded via an edge server, the forwarding edge nodes are determined by a preset rule that defines the set of reachable edge nodes on the terminal. Internal determination.

[0127] The hard constraints of task execution are decomposed into sub-masks and combined using a logical AND operation to obtain the action mask. .For example, The following submasks can be combined using a logical AND: dependency and ready submasks. Channel occupancy and time slot start restriction sub-mask Accessibility mask Edge concurrent capacity submask .

[0128] Where, if a is Local, then The condition for being 1 is and ; If a is Edge(k) or Cloud, then The condition for being 1 is and ; If a is Edge(k) or Cloud forwarding via Edge(k), then The condition for being 1 is ; If a is Edge(k), then The condition for a value of 1 is the number of concurrently executed tasks.

[0129] If and only if all sub-masks are 1, then... Only when any submask is 1; when any submask is 0, The set of possible actions is 0. For the action set A that satisfies The set of all actions.

[0130] After each task is started, the terminal resource channel q and b are updated. Based on the updated q and b, the channel-related submask is recalculated, and the action set Avalid is updated. Then, the next round of "sorting - scanning the first ready task - starting" is entered. This order is used to avoid online jitter caused by outputting unexecutable actions and backtracking and reselecting.

[0131] When multiple edge nodes are feasible at the same time, a preference rule can be introduced to prioritize those with shorter queuing times, such as selecting edge nodes with shorter expected queuing times.

[0132] The update of the remaining busy time q of the terminal resource channel is achieved through the busy time evolution of the terminal resource channel. When moving from time slot t to time slot t+1, the busy time decays and is superimposed with the busy time increment caused by the tasks started in time slot t, as shown below: (3) (4) in, This indicates taking the maximum value. This represents the busy time increment introduced when a task is started at the time slot t boundary, targeting... One estimation method is given: If task If executed locally, the estimated execution time is: (5) If task T(i,j) is unloaded and executed, its uplink transmission time is estimated to be: (6) And can be taken: (7) (8) Where ceil represents rounding up; cycles(i,j) is the computational cost (CPU cycles) of task T(i,j); bits(i,j) is the amount of data (bits) that task T(i,j) needs to transmit uplink; f(i,t) is the available CPU frequency or equivalent computing power of the terminal; and R(i,t) is the effective uplink rate of the terminal.

[0133] based on Remaining busy time for terminal resource channels Updates are performed via busy-hour increments. The busy level of the terminal resource channel in the future period after the task is started is quantified, which is used to determine the availability of the terminal resource channel in subsequent time slots.

[0134] As an example, refer to Figure 4 The illustrated action set generation process requires input parameters including ready tasks. Remaining busy time for local computing channels Uplink transmission channel remaining busy time The reachable edge set M(i,t) and the initiation flag within the time slot (including the initiation flag of the local computing channel). and the start flag of the uplink transmission channel ), and action set A.

[0135] A decomposition masking mechanism is employed to generate corresponding sub-masks for the constraints of the input parameters. Then, based on constraint determination rules, the constraint satisfaction status of each sub-mask is determined, resulting in an action mask. The generated action mask is used to filter available actions, resulting in a set of available actions. Then, in the set of possible actions If not empty, the ready task will be... As a candidate task In the action mask Under constraints, actions are selected from the set of available actions through scheduling decisions. Finally, output the selected action. At the corresponding execution location, start and execute the candidate task. At the same time, update the candidate tasks. The remaining busy time of the terminal and the start flag within time slot t. The remaining busy time is updated according to the method shown in Formula 3-8 above.

[0136] In one implementation, the execution location selection can be achieved using a heuristic strategy or a reinforcement learning strategy. If a reinforcement learning strategy is used, the action can be output under the constraint of the action mask, so that the strategy selects the action only within the set of available actions Avalid. The training phase can be optimized by combining timeout penalties, latency and energy consumption shaping.

[0137] In this embodiment, firstly, to address the issue of critical links being blocked by non-critical links, a mandatory dynamic priority sorting and sequential scanning selection mechanism ensures that highly critical tasks have priority in being scheduled and processed, suppressing resource preemption by non-critical tasks and providing priority guarantees for the smooth progress of the critical chain. Secondly, to address the issue of rollbacks and latency jitter caused by infeasible decisions, a feasible action set pre-generation mechanism is introduced. Before making a decision, the feasibility of the decision is filtered, and all hard constraints such as task dependencies, channel occupancy, and resource capacity are encoded. This ensures that the scheduler only makes selections from a fully feasible action space, logically avoiding the possibility of outputting invalid or conflicting instructions. This eliminates the additional latency and uncertainty jitter caused by rollback and reselection, greatly enhancing the reliability of single decisions and the stability of online decisions.

[0138] Furthermore, by decoupling the terminal's computing and communication capabilities through a dual-channel abstract model, local computing and data uploading can proceed in parallel when conditions permit, reducing idle waiting in the traditional serial execution mode and compressing the end-to-end latency of a single task. On the other hand, the iterative dispatch mechanism within the time slot boundaries creates a tight scheduling window, allowing resources released by the initiation of a task (if downstream tasks are ready) or newly exposed resource slots within the same time slot to be immediately discovered and utilized by subsequent iterations. This drives the scheduler to continuously utilize available resources, thereby significantly improving the resource throughput within each scheduling time slot and the overall efficiency of the computing task completion.

[0139] The system selects the execution location for candidate tasks across local, edge, or cloud environments, and masks inexecutable actions using a set of feasible actions or an action mask. This allows output actions to be directly executed and avoids latency jitter caused by rollback and reselection. When multiple edge nodes simultaneously meet hard constraints, the system can select based on edge queuing levels to reduce queuing tail risk and improve tail latency. After task startup, a non-preemptive approach is used until completion. Each task startup occupies the corresponding resource channel until completion, improving state consistency and reducing runtime control overhead.

[0140] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the task processing method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0141] This application also provides a task processing device, please refer to... Figure 5 The task processing device includes: The state update module 10 is used to update the global ready state of the task set with dependencies on each terminal at the time slot boundary of each scheduling time slot. The priority sorting module 20 is used to calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. The action filtering module 30 is used to generate a set of actionable actions for each ready task by traversing and scanning the ready task list. The task selection module 40 is used to select candidate tasks from the ready tasks according to the set of actionable tasks; The action selection module 50 is used to select an execution position for the candidate task from the set of actionable actions of the candidate task and start execution.

[0142] In one embodiment, the priority sorting module 20 is further configured to: The critical chain urgency of the target application is calculated based on the remaining deadline of the target application within the scheduling slot and the lower bound of the estimated execution time of the unfinished critical chain of the target application; wherein, the target application is the application to which the task in the globally ready state belongs; The scheduling priority of the task is determined based on the critical chain urgency.

[0143] In one embodiment, the action filtering module 30 is further configured to: A target task is selected from the list of ready tasks in sorted order by traversing and scanning. A set of action masks is predefined for each possible execution location action of the target task; the action mask includes multiple sub-masks, and each sub-mask corresponds to the satisfaction of a constraint condition; the constraint conditions include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints; the resource channel is an abstraction of the terminal's computing resources, and the resource channel includes a local computing channel and an uplink transmission channel for task offloading; The sub-masks of each constraint condition are combined and judged to determine the feasible execution location of the target task; Combine the feasible execution locations into a set of feasible actions for the target task; If the target task is not the last ready task in the ready task list, return and execute the step of selecting a target task from the ready task list in sorted order by traversing and scanning, until the target task is the last ready task in the ready task list.

[0144] In one embodiment, the action selection module 50 is further configured to: A global task graph is constructed based on the dependencies between the ready tasks in the ready task list; the global task graph contains explicit markers for the candidate tasks. The global task graph is subjected to bidirectional graph attention encoding to obtain the corresponding state representation; Based on the state representation, an execution location is selected for the candidate task from the set of actionable actions for the candidate task, and execution is initiated; the execution location is at least one of the terminal local, edge server, and cloud server.

[0145] In one embodiment, the action selection module 50 is further configured to: Extract the node features of each node in the global task graph; the node features include task size, computational intensity, application to which it belongs, application remaining time, terminal busy / idle flag, and candidate flag; The node features are subjected to forward graph attention aggregation to obtain forward graph attention features, and the node features are subjected to reverse graph attention aggregation to obtain reverse graph attention features; The forward graph attention features and the reverse graph attention features are concatenated and fused to obtain the corresponding state representation.

[0146] In one embodiment, the task processing device further includes a resource update module, used for: Update the resource channel status of the target terminal to which the candidate task belongs, and the resource status of the edge server corresponding to the target terminal, and remove the candidate task from the tasks in the globally ready state; The steps of returning and executing the steps of calculating the scheduling priority of the tasks in the global ready state, sorting the tasks according to the scheduling priority, and obtaining a list of ready tasks are repeated until no candidate tasks exist.

[0147] The task processing apparatus provided in this application, employing the task processing method described in the above embodiments, can solve the technical problems of insufficient control over the task initiation sequence in dynamic industrial environments, leading to easy blockage of critical task chains; and the lack of explicit compliance mechanisms for scheduling decisions with hard constraints, resulting in infeasible decisions and thus triggering rollback and latency jitter. Compared with the prior art, the beneficial effects of the task processing apparatus provided in this application are the same as those of the task processing method provided in the above embodiments, and other technical features in the task processing apparatus are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0148] This application provides a task processing device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the task processing method in Embodiment 1 above.

[0149] The following is for reference. Figure 6 The diagram illustrates a structural schematic of a task processing device suitable for implementing embodiments of this application. The task processing device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The task processing device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0150] like Figure 6 As shown, the task processing device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the task processing device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the task processing device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows task processing devices with various systems, it should be understood that it is not required to implement or possess all of the systems shown. More or fewer systems may be implemented alternatively.

[0151] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0152] The task processing device provided in this application, employing the task processing method described in the above embodiments, can solve the technical problems of insufficient control over the task initiation sequence in dynamic industrial environments, leading to easy blockage of critical task chains; and the lack of explicit compliance mechanisms for scheduling decisions with hard constraints, resulting in infeasible decisions and thus triggering rollback and latency jitter. Compared with the prior art, the beneficial effects of the task processing device provided in this application are the same as those of the task processing method provided in the above embodiments, and other technical features in this task processing device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0153] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0155] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the task processing method described in the above embodiments.

[0156] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0157] The aforementioned computer-readable storage medium may be included in the task processing device or may exist independently and not assembled into the task processing device.

[0158] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by a task processing device, cause the task processing device to: At the time slot boundaries of each scheduling time slot, update the global ready state of the task sets with dependencies on each terminal; Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. A set of actionable actions for each ready task is generated by traversing and scanning the ready task list. Candidate tasks are selected from the ready tasks based on the set of actionable actions; From the set of actionable actions for the candidate tasks, select an execution location for the candidate task and start execution.

[0159] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0160] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0161] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0162] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described task processing method. This addresses the technical problems of insufficient control over the task initiation sequence in dynamic industrial environments, leading to the easy blocking of critical task chains; and the lack of explicit compliance mechanisms for scheduling decisions with hard constraints, resulting in infeasible decisions and consequently rollback and latency jitter. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the task processing method provided in the above embodiments, and will not be repeated here.

[0163] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the task processing method described above.

[0164] The computer program product provided in this application can solve the technical problems of insufficient control over the task initiation sequence in dynamic industrial environments, which leads to the easy blockage of critical task chains; and the lack of explicit compliance mechanism for scheduling decisions with hard constraints, resulting in infeasible decisions and thus triggering rollback and latency jitter. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the task processing method provided in the above embodiments, and will not be repeated here.

[0165] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A task processing method, characterized in that, The task processing method includes: At the time slot boundaries of each scheduling time slot, update the global ready state of the task sets with dependencies on each terminal; Calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. A set of actionable actions for each ready task is generated by traversing and scanning the ready task list. Candidate tasks are selected from the ready tasks based on the set of actionable actions; From the set of actionable actions for the candidate tasks, select an execution location for the candidate task and start execution.

2. The task processing method as described in claim 1, characterized in that, The step of calculating the scheduling priority of tasks in the globally ready state includes: The critical chain urgency of the target application is calculated based on the remaining deadline of the target application within the scheduling slot and the lower bound of the estimated execution time of the unfinished critical chain of the target application; wherein, the target application is the application to which the task in the globally ready state belongs; The scheduling priority of the task is determined based on the critical chain urgency.

3. The task processing method as described in claim 1, characterized in that, The step of generating a set of actionable actions for each ready task by traversing and scanning the ready task list includes: A target task is selected from the list of ready tasks in sorted order by traversing and scanning. A set of action masks is predefined for each possible execution location action of the target task; the action mask includes multiple sub-masks, and each sub-mask corresponds to the satisfaction of a constraint condition; the constraint conditions include task dependency constraints, resource channel occupancy constraints, edge reachability constraints, and edge server resource capacity constraints; the resource channel is an abstraction of the terminal's computing resources, and the resource channel includes a local computing channel and an uplink transmission channel for task offloading; The sub-masks of each constraint condition are combined and judged to determine the feasible execution location of the target task; Combine the feasible execution locations into a set of feasible actions for the target task; If the target task is not the last ready task in the ready task list, return and execute the step of selecting a target task from the ready task list in sorted order by traversing and scanning, until the target task is the last ready task in the ready task list.

4. The task processing method as described in claim 1, characterized in that, The step of selecting an execution location and initiating execution for a candidate task from the set of actionable actions for the candidate task includes: A global task graph is constructed based on the dependencies between the ready tasks in the ready task list; the global task graph contains explicit markers for the candidate tasks. The global task graph is subjected to bidirectional graph attention encoding to obtain the corresponding state representation; Based on the state representation, an execution location is selected for the candidate task from the set of actionable actions for the candidate task, and execution is initiated; the execution location is at least one of the terminal local, edge server, and cloud server.

5. The task processing method as described in claim 4, characterized in that, The step of performing bidirectional graph attention encoding on the global task graph to obtain the corresponding state representation includes: Extract the node features of each node in the global task graph; the node features include task size, computational intensity, application to which it belongs, application remaining time, terminal busy / idle flag, and candidate flag; The node features are subjected to forward graph attention aggregation to obtain forward graph attention features, and the node features are subjected to reverse graph attention aggregation to obtain reverse graph attention features; The forward graph attention features and the reverse graph attention features are concatenated and fused to obtain the corresponding state representation.

6. The task processing method according to any one of claims 1 to 5, characterized in that, After the step of selecting an execution location for a candidate task from the set of actionable actions for the candidate task and initiating execution, the method further includes: Update the resource channel status of the target terminal to which the candidate task belongs, and the resource status of the edge server corresponding to the target terminal, and remove the candidate task from the tasks in the global ready state; The steps of returning and executing the steps of calculating the scheduling priority of the tasks in the global ready state, sorting the tasks according to the scheduling priority, and obtaining a list of ready tasks are repeated until no candidate tasks exist.

7. A task processing device, characterized in that, The task processing device includes: The state update module is used to update the global ready state of the task set with dependencies on each terminal at the time slot boundary of each scheduling time slot. The priority sorting module is used to calculate the scheduling priority of tasks in the global ready state, and sort the tasks according to the scheduling priority to obtain a ready task list; the scheduling priority of each ready task in the ready task list is used to define the criticality of the ready task. The action filtering module is used to generate a set of actionable actions for each ready task by traversing and scanning the ready task list. The task selection module is used to select candidate tasks from the ready tasks based on the set of actionable actions. The action selection module is used to select an execution position for the candidate task from the set of actionable actions of the candidate task and start execution.

8. A task processing device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the task processing method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the task processing method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the task processing method as described in any one of claims 1 to 6.

Images