Parameter configuration method, device, equipment and medium

Abstract

The embodiment of the invention provides a parameter configuration method and device, equipment and a medium, which are used for solving the problem that a parameter optimization method in related technologies is easy to fall into a local optimal solution and cannot improve the task execution efficiency. In the embodiment of the invention, the electronic equipment randomly generates the plurality of parameter value groups, so that the possibility of exploring different configuration combinations is provided, the parameter configuration with better performance can be found, the collected information is learned and analyzed by utilizing the strengthening model, the target parameter value of each parameter is automatically output, and the efficiency of the electronic equipment is improved. According to the method, the subjectivity and uncertainty of manual judgment are reduced, so that the parameter value of each parameter of the target tool can be accurately and effectively determined, the parameters are configured, and the subsequent execution efficiency of the task is improved.

Description

Technical Field

[0001] The present application relates to the technical field of big data performance tuning, and in particular to a parameter configuration method, device, equipment and medium. Background Art

[0002] Hive and Flink are important tools in the field of big data processing, used for batch processing and stream processing tasks respectively. Their performance depends largely on the optimization of parameter configuration. For example, to execute the same task at 2 pm every day, Hive parameters include column storage parameters, Map side parameters, Reduce side parameters, optimization parameters, and storage parameters. However, these parameters are numerous and have complex interdependencies. Manual parameter tuning requires rich expert knowledge and practical experience, and the knowledge cost is high. For users who lack relevant experience, manual tuning is not only inefficient, but also can only be tried within a limited search space, making it difficult to find the global optimal configuration.

[0003] Existing parameter optimization methods mainly rely on empirical rules or heuristic algorithms, which are usually based on empirical summaries in specific scenarios and are prone to fall into local optimal solutions. Summary of the invention

[0004] The embodiments of the present application provide a parameter configuration method, apparatus, device and medium, which are used for the parameter optimization method in the related art, which is prone to fall into the problem of local optimal solution and unable to improve the efficiency of task execution.

[0005] In a first aspect, an embodiment of the present application provides a parameter configuration method, the method comprising:

[0006] If an instruction for parameter configuration of a target tool is received, each parameter to be configured of the target tool saved in advance is obtained; a preset number of parameter values ​​of each parameter are randomly generated, and the parameter values ​​of each parameter generated at any time are combined into a corresponding parameter value group, wherein the target tool is Hive or Flink;

[0007] For each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0008] Each parameter value group, the corresponding execution time and resource cost information are input into a pre-trained reinforcement model to obtain a target parameter value for each parameter output by the reinforcement model.

[0009] In a second aspect, an embodiment of the present application further provides a parameter configuration device, the device comprising:

[0010] A processing module, configured to, if receiving an instruction for parameter configuration of a target tool, obtain each parameter to be configured of the target tool that is saved in advance; randomly generate a preset number of parameter values ​​of each parameter, and form a corresponding parameter value group with the parameter values ​​of each parameter generated at any time, wherein the target tool is Hive or Flink;

[0011] A determination module, for determining, for each parameter value group, when each parameter of the target tool is determined to be a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction;

[0012] The acquisition module is used to input each parameter value group, the corresponding execution time and resource cost information into a pre-trained reinforcement model to obtain the target parameter value of each parameter output by the reinforcement model.

[0013] In a third aspect, an embodiment of the present application further provides an electronic device, including:

[0014] A memory for storing program instructions;

[0015] The processor is used to call the program instructions stored in the memory and execute the steps included in any one of the above parameter configuration methods according to the obtained program instructions.

[0016] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a computer, the computer executes a parameter configuration method as described in any one of the above items.

[0017] In the embodiment of the present application, if the electronic device receives an instruction for parameter configuration of a target tool, it obtains each parameter to be configured that is saved in advance; randomly generates a preset number of parameter values ​​for each parameter, and obtains the corresponding parameter value group; for each parameter value group, when it is determined that each parameter of the target tool is the parameter value of the parameter value group, it executes the task corresponding to the task identifier carried in the instruction, and obtains the corresponding execution time and resource cost information; each parameter value group, the corresponding execution time and resource cost information are input into the pre-trained reinforcement model, and the target parameter value of each parameter output by the reinforcement model is obtained. Since the electronic device randomly generates multiple parameter value groups, it provides the possibility of exploring different configuration combinations, which helps to find parameter configurations with better performance, and uses the reinforcement model to learn and analyze the collected information, and automatically outputs the target parameter value of each parameter, which reduces the subjectivity and uncertainty of human judgment, so that the parameter value of each parameter of the target tool can be accurately and effectively determined, and then the parameters can be configured, thereby improving the efficiency of subsequent execution of the task. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A process diagram of a parameter configuration method provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of a process for determining key parameters provided in an embodiment of the present application;

[0021] Figure 3 A process for generating parameter optimization suggestions provided in an embodiment of the present application;

[0022] Figure 4 A detailed process diagram of parameter configuration provided in an embodiment of the present application;

[0023] Figure 5 A structural diagram of a parameter configuration device provided in an embodiment of the present application;

[0024] Figure 6 A structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of this application.

[0026] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0027] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0028] The words "connected", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "plurality" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. The terms "first", "second", "third" and the like involved in this application are merely used to distinguish similar objects, and do not represent a specific ordering of objects.

[0029] In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0030] In order to solve the problem that the parameter optimization method in the related art is prone to fall into the local optimal solution and cannot improve the efficiency of task execution, the embodiments of the present application provide a parameter configuration method, device, equipment and medium.

[0031] The parameter configuration method includes: if the electronic device receives an instruction for parameter configuration of a target tool, then each parameter of the target tool to be configured is obtained in advance; a preset number of parameter values ​​for each parameter are randomly generated, and the parameter value of each parameter generated at any time is formed into a corresponding parameter value group, wherein the target tool is Hive or Flink; for each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution time and resource cost information of the task corresponding to the task identifier carried in the execution instruction are input; each parameter value group, the corresponding execution time and resource cost information are input into a pre-trained reinforcement model, and the target parameter value of each parameter output by the reinforcement model is obtained.

[0032] Embodiment 1:

[0033] Figure 1 A process diagram of a parameter configuration method provided in an embodiment of the present application, the process includes the following steps:

[0034] S101: If an instruction for parameter configuration of a target tool is received, each parameter to be configured of the target tool that is pre-saved is obtained; a preset number of parameter values ​​of each parameter are randomly generated, and the parameter values ​​of each parameter generated at any time are combined into a corresponding parameter value group, wherein the target tool is Hive or Flink.

[0035] The parameter configuration method provided in the embodiment of the present application is applied to an electronic device, which may be a smart device such as a PC or a server.

[0036] In order to configure parameters accurately and effectively, when business personnel need to configure parameters, they can select the target tool through the preset page of the device they use or the preset device, and click the preset button, such as the "Configure" button. At this time, the electronic device can receive instructions for parameter configuration of the target tool, where the target tool is Hive or Flink.

[0037] In order to configure parameters accurately and effectively, the electronic device locally saves each parameter to be configured for the tool in advance, and the electronic device can obtain each parameter to be configured of the pre-saved target tool. Among them, Hive parameters include column storage parameters, Map side parameters, Reduce side parameters, optimization parameters, storage parameters, etc. Flink parameters include ports, number of retries, concurrency, etc.

[0038] For each parameter to be configured, the electronic device randomly generates a preset number of parameter values. The randomly generated parameter values ​​of each parameter are combined into a corresponding parameter value group. Specifically, the electronic device uses a loop, nested loop or iteration method in a programming language to traverse the parameter value of each parameter, and combines each parameter value of each parameter with the parameter values ​​of other parameters to form a parameter value group.

[0039] S102: For each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the execution instruction is executed.

[0040] In order to accurately and effectively configure parameters, the instructions received by the electronic device carry a task identifier, wherein different tasks have different identifiers.

[0041] For each parameter value group determined, the electronic device can determine that when each parameter of the target tool is a parameter value in the parameter value group, execute the task corresponding to the task identifier, and determine the execution time and resource cost information of executing the task. Specifically, the electronic device can determine the resource information and obtain the resource cost information saved corresponding to the resource information. Among them, the resource cost information is the cost information corresponding to the usage of resources such as CPU, memory, and disk. Taking the Grafana monitoring platform as an example, it is used to obtain the execution time of the task and the usage of resources such as CPU, memory, and disk, and combine the unit resource cost to obtain the total resource cost. Taking the Yarn cluster as an example, the resource cost involves CPU resources and memory resources such as JobManager and TaskManager.

[0042] S103: Input each parameter value group, the corresponding execution time and resource cost information into a pre-trained reinforcement model to obtain a target parameter value of each parameter output by the reinforcement model.

[0043] In order to configure parameters accurately and effectively, the electronic device locally stores a pre-trained enhancement model. The electronic device inputs each parameter value group, the execution time corresponding to each parameter value group, and the resource cost information into the pre-trained enhancement model, and obtains the output of the enhancement model. The output of the enhancement model is the target parameter value of each parameter.

[0044] The electronic device may configure the parameters of the target tool based on the acquired target parameter value of each parameter.

[0045] The electronic equipment implements intelligent parameter tuning of Hive / Flink tasks based on reinforcement learning algorithms. Compared with traditional manual tuning methods, intelligent parameter tuning reduces the reliance on professional knowledge and practical experience, significantly improves tuning efficiency and performance, and reduces labor costs.

[0046] In the embodiment of the present application, if the electronic device receives an instruction for parameter configuration of a target tool, it obtains each parameter to be configured that is saved in advance; randomly generates a preset number of parameter values ​​for each parameter, and obtains the corresponding parameter value group; for each parameter value group, when it is determined that each parameter of the target tool is the parameter value of the parameter value group, it executes the task corresponding to the task identifier carried in the instruction, and obtains the corresponding execution time and resource cost information; each parameter value group, the corresponding execution time and resource cost information are input into the pre-trained reinforcement model, and the target parameter value of each parameter output by the reinforcement model is obtained. Since the electronic device randomly generates multiple parameter value groups, it provides the possibility of exploring different configuration combinations, which helps to find parameter configurations with better performance, and uses the reinforcement model to learn and analyze the collected information, and automatically outputs the target parameter value of each parameter, which reduces the subjectivity and uncertainty of human judgment, so that the parameter value of each parameter of the target tool can be accurately and effectively determined, and then the parameters can be configured, thereby improving the efficiency of subsequent execution of the task.

[0047] Embodiment 2:

[0048] In order to accurately and effectively configure parameters, based on the above embodiments, in the embodiment of the present application, the randomly generated preset number of parameter values ​​of each parameter includes:

[0049] For each parameter, a preset number of parameter values ​​within the parameter value range are randomly generated according to a pre-stored parameter value range corresponding to the parameter.

[0050] Since each parameter has a corresponding parameter value range, if the corresponding parameter value is randomly generated, it may not be possible to accurately and effectively perform tasks based on the generated parameter value. Therefore, in an embodiment of the present application, the electronic device generates the parameter value based on the parameter value range corresponding to the parameter.

[0051] Specifically, the electronic device can consult and obtain a pre-stored parameter value range corresponding to each parameter to be configured, and then randomly generate a preset number of parameter values ​​within the parameter value range, the number of which is preset and adjustable to meet different configuration requirements and application scenarios.

[0052] Since the electronic device randomly generates parameter values ​​by following the preset parameter value range, it ensures that the generated parameter values ​​are both logical and fit the actual application scenario, thereby improving the effectiveness and accuracy of task execution.

[0053] Embodiment 3:

[0054] In order to accurately and effectively configure parameters, based on the above embodiments, in the embodiment of the present application, when determining that each parameter of the target tool is a parameter value in the parameter value group, after executing the execution time and resource cost information of the task corresponding to the task identifier carried in the instruction, before inputting each parameter value group, the corresponding execution time and resource cost information into the pre-trained reinforcement model, the method further includes:

[0055] According to each parameter value group and the corresponding execution time, a first-order Sobol index and a first overall effect corresponding to each parameter are determined respectively; and according to each parameter value group and the corresponding resource cost information, a second-order Sobol index and a second overall effect corresponding to each parameter are determined respectively;

[0056] For each parameter, determine the impact value of the parameter on the execution time and resource cost information according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter, and if the impact value is lower than a threshold, delete the parameter value corresponding to the parameter in each parameter value group;

[0057] For each retained parameter, perform a subsequent step of inputting each parameter value group, the corresponding execution time, and the resource cost information into the pre-trained reinforcement model.

[0058] In actual application environments, there are many parameters to be configured, some of which have little impact on task performance. In order to improve configuration efficiency and optimize resource utilization, electronic devices can screen the parameters to be configured to determine which of the many parameters to be configured are "key parameters" that have a significant impact on task performance.

[0059] Specifically, after determining the execution time and resource cost information of the task corresponding to the task identifier carried in the execution instruction, the electronic device can determine the first-order Sobol index and the overall effect corresponding to each parameter according to each parameter value group and the corresponding execution time. For the sake of distinction, the first-order Sobol index can be called the first first-order Sobol index, and the overall effect can be called the first overall effect; and according to each parameter value group and the corresponding resource cost information, the first-order Sobol index and the overall effect corresponding to each parameter can be determined respectively. For the sake of distinction, the first-order Sobol index can be called the second first-order Sobol index, and the overall effect can be called the second overall effect. Among them, the overall effect can also be called the total effect Sobol index.

[0060] In the embodiment of the present application, the electronic device uses the Sobol analysis method to perform sensitivity analysis on the parameters involved in the task execution process, identify and filter out the key parameters that have a significant impact on the task performance. By using the task execution time and resource cost as evaluation indicators, the complexity and computational overhead of parameter tuning are reduced, and the scientificity and accuracy of tuning are improved.

[0061] The embodiment of the present application is equivalent to using Sobol Sensitivity Analysis to obtain key parameters. Sobol Sensitivity Analysis is a global sensitivity analysis method used to evaluate the impact of input parameters on output. The basic idea is to determine the contribution of each input parameter and its interaction to the output by decomposing the variance of the output variable. The first-order Sobol Index is used to measure the direct contribution of a single input parameter to the output variable, the total-effect Sobol Index is used to measure the contribution of the input parameter to the output variable considering all interactions, and the total effect corresponds to the degree of influence of the current parameter and other parameters on the output variable.

[0062] For each parameter to be configured, the impact value of the parameter on the execution time and resource cost information is determined according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter.

[0063] In a possible implementation, the electronic device can determine, for each parameter to be configured, a target first-order Sobol index according to the first-order Sobol index, the second-order Sobol index, and the corresponding weights of the parameter, and determine a target overall effect according to the first overall effect, the second overall effect, and the corresponding weights. After determining the target first-order Sobol index and the target overall effect, the impact value of the parameter on the execution time and resource cost information can be determined according to the target first-order Sobol index, the target overall effect, and the corresponding weights.

[0064] In another possible implementation, the electronic device may determine the corresponding first-order Sobol index and the overall effect only according to the execution time, and determine the impact value according to the first-order Sobol index and the overall effect.

[0065] If the impact value is lower than the threshold, it is determined that the parameter has little impact on the execution time and resource cost information, and the parameter value corresponding to the parameter can be deleted in each parameter value group; if the impact value is not lower than the threshold, it is determined that the parameter has a significant impact on the execution time and resource cost information, and the parameter is determined to be a key parameter. The threshold can be 0.2.

[0066] The embodiment of the present application comprehensively considers the independent influence and synergistic influence of the parameters, calculates the weighted value according to the first-order Sobol index and the overall effect, and screens out the key parameters according to the set threshold. The method provided in the embodiment of the present application is used to extract the parameters with strong correlation with Hive / Flink performance from each parameter to be configured, and removes those unimportant parameters, which can effectively shorten the generation time and resource cost of the parameter optimization strategy.

[0067] Figure 2 A schematic diagram of a process for determining key parameters provided in an embodiment of the present application.

[0068] Depend on Figure 2 It can be seen that the electronic device can first randomly generate parameter values ​​within the value range to obtain multiple groups of parameter value groups, that is, parameter configuration combinations, and perform central processing unit (Central Processing Unit, CPU) intensive / input output (Input Output, IO) intensive tests on the simulation tasks based on the big data testing tool. The execution time of the task corresponding to each parameter value group configuration is obtained, that is, the execution time. The first-order Sobol index and overall effect of each parameter are obtained based on Sobol analysis. The electronic device performs weighted summation based on the first-order Sobol index and the overall effect to obtain the parameter importance weight, that is, the impact value described in the embodiment of the present application. If the impact value is less than the threshold value t, the parameter is discarded. If it is not less than the threshold value t, the parameter is determined as a key parameter.

[0069] Figure 2 In the example, determining the impact value of the parameter based on the execution time of the task is taken, and the specific method of determining the impact value of the parameter is not limited here.

[0070] It should be noted that the electronic device can determine the target parameter values ​​of key parameters based on the method provided in the embodiments of the present application, and determine the parameter values ​​of parameters that have less impact on the execution time and resource cost information as pre-saved preset parameter values.

[0071] Embodiment 4:

[0072] In order to configure the parameters accurately and effectively, based on the above embodiments, in the embodiment of the present application, the method further includes:

[0073] When determining that each parameter of the target tool is the target parameter value output by the reinforcement model, the target execution time and target resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0074] Determining a target reward value according to the target execution time, the target resource cost information and the corresponding weights;

[0075] Based on the target reward value and the pre-stored standard reward value, the reinforcement parameters of the reinforcement model are adjusted.

[0076] In order to configure the parameters accurately and effectively, the electronic device can adjust the enhanced parameters of the enhanced model.

[0077] Specifically, after determining each target parameter value output by the enhancement model, the electronic device can determine that when each parameter of the target tool is the target parameter value output by the enhancement model, the execution time and resource cost information of the task corresponding to the task identifier carried in the execution instruction. For ease of distinction, the execution time can be referred to as the target execution time, and the resource cost information can be referred to as the target resource cost information.

[0078] The electronic device can determine the target reward value according to the target execution time, the target resource cost information and the corresponding weights; in a possible implementation, the electronic device can determine the target time range to which the target execution time belongs according to each pre-saved time range, and determine the first reward value saved for the target time range. The electronic device can also determine the target resource cost range to which the target resource cost information belongs according to each pre-saved resource cost range, and determine the second reward value saved for the target resource cost range; determine the target reward value according to the first reward value, the second reward value and the corresponding weights. It should be noted that if the execution time is the same, the smaller the resource cost, the larger the target reward value; if the resource cost is the same, the shorter the execution time, the larger the target reward value.

[0079] After determining the target reward value, the electronic device may adjust the reinforcement parameters of the reinforcement model based on the target reward value and the pre-stored standard reward value.

[0080] In one possible implementation, after the reinforcement model outputs the target parameter value for each parameter, the target parameter value for each parameter can be sent to a preset device or a device used by the business personnel themselves. If the business personnel decides to adopt the target parameter value, they click a preset button, such as an "OK" button. At this time, the business personnel updates the parameter configuration according to the suggestion of the electronic device, and the electronic device generates a new task and calculates the reward, optimizing the reinforcement model in real time. If the business personnel does not adopt the suggestion of the electronic device, there is no need to optimize the reinforcement model based on the obtained target parameter value.

[0081] The embodiment of the present application is equivalent to adjusting the parameters of the reinforcement model through online reinforcement learning (Online Reinforcement Learning, Online RL). Among them, online reinforcement learning is a machine learning method that dynamically adjusts strategies to optimize decisions by interacting with the environment in real time during the operation of the system. Unlike offline reinforcement learning, online reinforcement learning continuously applies the learned strategies while training the model, updates and improves the model in real time to adapt to the ever-changing environment and task requirements, and is suitable for scenarios that require rapid response and dynamic adjustment.

[0082] In the embodiment of the present application, the electronic device can be based on the parameter tuning method of the Twin Delayed Deep Deterministic policy gradient (TD3) online reinforcement learning model. By deploying the trained reinforcement model in the task processing process, combined with the data indicators of the task monitoring platform, that is, the parameter value group mentioned in the embodiment of the present application, the corresponding execution time and resource cost information, it can perform parameter performance evaluation and analysis on offline or real-time tasks, and automatically configure the parameters of Hive / Flink when the parameter configuration can be optimized. TD3 is an online (on-line) off-policy deep reinforcement learning algorithm for solving continuous control problems, which is improved by Scott Fujimoto et al. on the Deep Deterministic Policy Gradient (DDPG) algorithm. In essence, the TD3 algorithm integrates the idea of ​​the Double Q-Learning algorithm into the DDPG algorithm.

[0083] It should be noted that due to the complexity and diversity of big data processing tasks, it is extremely challenging to find parameter configurations that can perform well in different tasks and environments. Although researchers have explored parameter tuning methods based on deep learning technology, these methods cannot be dynamically adjusted according to changes in workload during actual runtime. Different types of tasks and data levels may have very different requirements for parameter configurations, and static optimization methods are difficult to cope with such changing requirements, resulting in unstable performance. In addition, existing methods often require a large number of parameter tuning experiments before task execution, and cannot adjust parameter configurations in real time during task execution. This method, which lacks real-time responsiveness, cannot quickly adapt to environmental changes, affecting overall efficiency.

[0084] Based on the online reinforcement model, parameter tuning of the Flink real-time processing framework is achieved. Electronic devices obtain environmental feedback in real time during operation, and dynamically adjust parameter configuration according to the state, quickly respond to environmental changes, and continuously optimize task execution performance. The reinforcement learning model is used to interact with the environment in real time, and the parameter configuration is gradually adjusted to maximize key performance indicators such as job execution time and resource cost, and intelligent parameter tuning suggestions are generated. Compared with existing methods, this method realizes automated, cross-task, and real-time response parameter optimization, and has better adaptability and tuning effects. This enables electronic devices to achieve better performance under different workloads and environmental conditions, effectively saving cluster computing resources of the Hive / Flink framework while improving overall performance, reducing development costs.

[0085] In order to solve the problems in the related art, a parameter tuning method that is compatible with offline and real-time business scenarios and can adapt to multiple task types is needed. The method provided in the embodiment of the present application should be able to monitor task performance and resource usage in real time during task execution, dynamically adjust parameter configuration, optimize job execution time and resource cost, and improve overall performance. At the same time, it should also have cross-task adaptability, and be able to maintain good performance under different types of tasks and environmental conditions, thereby improving the resource utilization efficiency of the Hive / Flink framework and reducing development and operation costs. The Hive / Flink online parameter tuning method based on reinforcement learning proposed in the embodiment of the present application realizes automated, cross-task, and real-time responsive parameter optimization, so that electronic devices can interact with the task environment in real time under different workloads and environmental conditions, and provide optimal parameter tuning suggestions. Parameter tuning across task types is achieved by using task metadata and model noise mechanisms. During task execution, metadata information such as operation set, task type, and data magnitude is collected as task knowledge. Combined with the TD3 reinforcement learning algorithm, the noise mechanism is used to improve the adaptability of the model to multiple tasks and improve the accuracy and stability of parameter optimization.

[0086] The parameter configuration process of the embodiment of the present application is described below in conjunction with a specific embodiment:

[0087] Implementation step 1: The electronic device generates n groups (configurable by business personnel, 20 groups by default) of random parameter configurations, and uses big data testing tools to generate simulation data of various task types. It starts tasks in the cluster based on each group of parameter configurations and counts the execution time.

[0088] Implementation step 2: The electronic device uses the Sobol analysis method to evaluate the impact of each parameter on the task execution time and resource cost information. The first-order Sobol index and the overall effect score are calculated, and the impact value of the parameter is calculated based on them. The parameters with weights greater than or equal to the threshold t are retained as the key parameter set for subsequent tuning.

[0089] Implementation step three: Initialize the TD3 reinforcement learning model, determine the reinforcement learning model state as task execution time, resource cost and task metadata, where the task metadata is each parameter value group described in the embodiment of this application. The model action is the key parameter set configuration to be tuned, and the reward function is obtained based on the task execution time and resource cost.

[0090] Implementation step 4: The reinforcement learning agent interacts with the simulated data, explores different parameter configurations, and continuously adjusts the parameters based on feedback to maximize the reward function value. After pre-training is completed, the reinforcement learning model is deployed to the actual environment and online parameter tuning of the Hive / Flink job is started.

[0091] Implementation step 5: When a new Hive or Flink task is submitted, the electronic device automatically collects the task metadata information. After the task is running, the agent provides real-time parameter optimization suggestions based on the state monitored from the environment. If the user updates the parameter configuration according to the suggestions, a new task is generated and the reward is calculated to optimize the model in real time. Otherwise, it continues to monitor the environment state.

[0092] Embodiment 5:

[0093] In order to accurately and effectively determine the target reward value, based on the above embodiments, in the embodiment of the present application, the target reward value is determined according to the target execution time, the target resource cost information and the corresponding weights, including:

[0094] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0095] If the first score is less than the pre-saved target score, obtaining the second score determined last time;

[0096] A target reward value is determined according to a ratio of the second score to the first score.

[0097] In an embodiment of the present application, the electronic device may determine a corresponding score based on the target execution time, target resource cost information, and the corresponding weights. For ease of distinction, the score may be referred to as a first score.

[0098] The electronic device locally stores a target score, wherein the target score can be a score determined by the execution time and resource cost information corresponding to the parameter value output by the reinforcement model for the first time. If the first score is less than the pre-saved target score, the electronic device obtains a second score corresponding to the execution time and resource cost information determined last time. After determining the second score, the electronic device can determine the ratio of the first score to the second score, and determine the target reward value based on the ratio. In a possible implementation, the ratio can be determined as the target reward value.

[0099] Specifically, the electronic device may determine the first score using the following formula:

[0100]

[0101] Among them, x represents the current task, T t (x) represents the task execution time at time t, Represents all resource costs involved in the task. w is the weight of the execution time, and 1-w is the weight of the resource cost.

[0102] Embodiment 6:

[0103] In order to accurately and effectively determine the target reward value, based on the above embodiments, in the embodiment of the present application, the target reward value is determined according to the target execution time, the target resource cost information and the corresponding weights, including:

[0104] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0105] If the first score is equal to the pre-saved target score, the target reward value is determined to be a preset first reward value; if the first score is greater than the target score, the target reward value is determined to be a preset second reward value; wherein the second reward value is greater than the first reward value.

[0106] In an embodiment of the present application, the electronic device may determine a corresponding score based on the target execution time, target resource cost information, and the corresponding weights. For ease of distinction, the score may be referred to as a first score.

[0107] The electronic device locally stores a target score, wherein the target score may be a score determined by the execution time and resource cost information corresponding to the parameter value outputted for the first time by the reinforcement model. If the first score is equal to the pre-saved target score, the electronic device may determine the preset first reward value as the target reward value. If the first score is greater than the target score, the preset second reward value may be determined as the target reward value. The second reward value is greater than the first reward value. The second reward value may be 0, and the first reward value may be -50.

[0108] Embodiment 7:

[0109] In order to accurately and effectively determine the target reward value, based on the above embodiments, in the embodiment of the present application, before adjusting the reinforcement parameters of the reinforcement model based on the target reward value and the pre-saved standard reward value, the method further includes:

[0110] If the parameter value of any parameter is not within the parameter value range corresponding to the parameter, the target reward value is determined to be a preset third reward value; wherein the third reward value is less than the second reward value corresponding to when the first score is equal to the pre-saved target score.

[0111] In the embodiment of the present application, if the parameter value of any parameter is not within the parameter value range corresponding to the parameter, it can be determined that the parameter value output by the reinforcement model is abnormal, and it is determined that the current performance of the reinforcement model is poor, and the electronic device can determine the preset third reward value as the target reward value. The third reward value is less than the second reward value corresponding to the first score being lower than the target score. The third reward value can be -100.

[0112] Specifically, the electronic device may use the following formula to determine the target reward value:

[0113]

[0114] Among them, reward is the target reward value, P t is the first score, P t-1 is the second score, P init is the target score.

[0115] Figure 3 A generation process of parameter optimization suggestions provided in an embodiment of the present application is specifically:

[0116] Depend on Figure 3It can be seen that at each time step, the model obtains the current state from the online monitoring platform, that is, the task execution time, resource cost, and task original data, and generates the initial recommended parameter configuration based on the actor network. The critic network will evaluate the quality of the generated parameter configuration. The parameter configuration adapts to changes in task type and data magnitude through feedback of reward signals. The task is performed according to the recommended parameter configuration, and the model observes the reward and stores the experience in the experience replay buffer. The model continuously randomly extracts samples from the experience replay buffer, adjusts the parameters of the network by maximizing the system performance, and accurately estimates the long-term return of each state-action pair. After training, the model can generate intelligent parameter optimization suggestions for the current task, which can be automatically updated to the parameter configuration of the task.

[0117] The method provided in the embodiment of the present application is used to solve problems such as machine resource overload or poor operating efficiency caused by parameter configuration when large network data is processed and analyzed offline, as well as event security risks caused by delays in real-time associated task alarms, thereby providing guarantees for task execution efficiency and reliability.

[0118] During the real-time parameter tuning process, the online job monitoring component is used to monitor the task running environment, including the running environment monitoring platform and task metadata. Task metadata obtains task-related information such as operation set, task type, and data level from Hive / Flink tasks. The operation set refers to each parameter value group, and the reinforcement model involves three parts: state, action, and reward. The state is obtained from the online job monitoring component, including performance indicators collected by the monitoring platform and task metadata information. The performance indicators are used to optimize the parameter tuning strategy. The task metadata is used to enable the reinforcement model to learn task-related information to achieve cross-task parameter tuning; the action is defined as the set of Hive / Flink key parameters that need to be optimized; the reward function is calculated based on the task execution time and resource cost. Lower execution time and resource cost correspond to higher rewards, and vice versa.

[0119] Figure 4 A detailed process diagram of parameter configuration provided for an embodiment of the present application.

[0120] Depend on Figure 4 It can be seen that the electronic equipment can first collect parameter configurations. Specifically, based on the Hive / Fink task, random parameter configurations are generated to obtain the Hive / Fink parameter set. Then, through Sobol analysis, the first-order Sobol index and overall effect are determined, and key parameters are extracted based on the first-order Sobol index and overall effect.

[0121] And, by Figure 4It can be seen that the electronic device also performs real-time parameter tuning. Specifically, online job monitoring monitors the task running environment, obtains the task execution time and resource cost through the monitoring platform, and obtains task metadata such as operation set, task type, data level, etc. A "parameter optimization strategy" is generated based on the TD3 reinforcement learning agent. This step involves multiple electronic device components, such as electronic devices, states, actor networks, intelligent parameters, etc. A parameter strategy designed to optimize the performance of the job will be generated based on the information of these components, as well as preset algorithms and rules. After generating the parameter optimization strategy, the electronic device will apply it to the "actor network" for intelligent parameter optimization. The actor network is a network that can output specific actions according to the parameter strategy. It continuously learns and adjusts its own parameters through interaction with the environment to find the best parameter combination. In the process of intelligent parameter optimization, the electronic device will also use the "critic network" to evaluate the quality of the strategy. The critic network will give an evaluation score based on the current strategy and state, that is, the target reward described in the embodiment of the present application. This score reflects the pros and cons of the strategy in the current state. The electronic device will adjust the strategy according to the evaluation score to improve the performance of the job.

[0122] Specifically, the embodiments of the present application mainly solve the following three problems:

[0123] 1. Manual tuning of Hive / Flink parameters is inefficient and produces suboptimal results: The performance of Hive and Flink is highly dependent on the choice of parameter configuration. However, these parameters are numerous and have complex interdependencies. Manual tuning requires extensive expert knowledge and practical experience, and the testing cost is high. For users who lack relevant experience, manual tuning is not only inefficient, but can only be tried within a limited search space, making it difficult to find the global optimal configuration.

[0124] 2. It is difficult to implement a cross-task parameter tuning strategy: Hive / Flink involves multiple workloads such as CPU-intensive and I / O-intensive workloads, as well as different data levels. The sensitivity and optimal configuration of parameters for different task types and data levels vary greatly. Currently, most parameter tuning methods can only target specific task types and cannot take into account multiple types of workloads.

[0125] 3. Offline parameter tuning methods are not suitable for real-time stream processing frameworks such as Flink: Most of the existing parameter tuning methods are performed offline. Offline tuning methods cannot dynamically adjust parameter strategies according to real-time changes in data streams, and cannot guarantee the optimal performance of real-time stream processing frameworks.

[0126] Embodiment 8:

[0127] Based on the same inventive concept, the present application embodiment provides a parameter configuration device, please refer to Figure 5 , the device comprises:

[0128] The processing module 501 is used for obtaining each parameter to be configured of the target tool that is saved in advance if an instruction for parameter configuration of the target tool is received; randomly generating a preset number of parameter values ​​of each parameter, and forming a corresponding parameter value group with the parameter values ​​of each parameter generated at any time, wherein the target tool is Hive or Flink;

[0129] A determination module 502 is used to determine, for each parameter value group, when each parameter of the target tool is a parameter value in the parameter value group, execute the execution time and resource cost information of the task corresponding to the task identifier carried in the instruction;

[0130] The acquisition module 503 is used to input each parameter value group, the corresponding execution time and resource cost information into the pre-trained enhancement model, and obtain the target parameter value of each parameter output by the enhancement model.

[0131] In a possible implementation, the processing module 501 is specifically configured to randomly generate a preset number of parameter values ​​within the parameter value range for each parameter according to a pre-saved parameter value range corresponding to the parameter.

[0132] In a possible implementation, the acquisition module 503 is also used to determine the first-order Sobol index and the first overall effect corresponding to each parameter according to each parameter value group and the corresponding execution time; and determine the second-order Sobol index and the second overall effect corresponding to each parameter according to each parameter value group and the corresponding resource cost information; for each parameter, determine the impact value of the parameter on the execution time and resource cost information according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter, if the impact value is lower than the threshold, delete the parameter value corresponding to the parameter in each parameter value group; for each retained parameter, perform the subsequent step of inputting each parameter value group, the corresponding execution time and resource cost information into the pre-trained reinforcement model.

[0133] In a possible implementation, the processing module 501 is also used to determine that when each parameter of the target tool is the target parameter value output by the reinforcement model, the target execution time and target resource cost information of the task corresponding to the task identifier carried in the instruction are executed; the target reward value is determined according to the target execution time, the target resource cost information and the corresponding weights; and the reinforcement parameters of the reinforcement model are adjusted based on the target reward value and the pre-saved standard reward value.

[0134] In a possible implementation, the processing module 501 is specifically used to determine the corresponding first score based on the target execution time, the target resource cost information and the corresponding weights; if the first score is less than the pre-saved target score, obtain the last determined second score; and determine the target reward value based on the ratio of the second score to the first score.

[0135] In one possible implementation, the processing module 501 is specifically used to determine a corresponding first score based on the target execution time, the target resource cost information and the corresponding weights; if the first score is equal to a pre-saved target score, the target reward value is determined to be a preset first reward value; if the first score is greater than the target score, the target reward value is determined to be a preset second reward value; wherein the second reward value is greater than the first reward value.

[0136] In one possible implementation, the processing module 501 is specifically used to determine that the target reward value is a preset third reward value if the parameter value of any parameter is not within the parameter value range corresponding to the parameter; wherein the third reward value is less than the second reward value corresponding to the first score being equal to the pre-saved target score.

[0137] Embodiment 9:

[0138] Based on the same inventive concept, the present application embodiment provides an electronic device, which can implement the function of the parameter configuration described above. Please refer to Figure 6 The device includes a processor 601, a memory 603 and a communication bus 604, wherein the processor 601, the communication interface 602 and the memory 603 communicate with each other through the communication bus 604.

[0139] The memory 603 stores a computer program. When the program is executed by the processor 601, the processor 601 performs the following steps:

[0140] If an instruction for parameter configuration of a target tool is received, each parameter to be configured of the target tool saved in advance is obtained; a preset number of parameter values ​​of each parameter are randomly generated, and the parameter values ​​of each parameter generated at any time are combined into a corresponding parameter value group, wherein the target tool is Hive or Flink;

[0141] For each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0142] Each parameter value group, the corresponding execution time and resource cost information are input into a pre-trained reinforcement model to obtain a target parameter value for each parameter output by the reinforcement model.

[0143] In a possible implementation manner, the randomly generating a preset number of parameter values ​​for each parameter includes:

[0144] For each parameter, a preset number of parameter values ​​within the parameter value range are randomly generated according to a pre-stored parameter value range corresponding to the parameter.

[0145] In a possible implementation, when determining that each parameter of the target tool is a parameter value in the parameter value group, after executing the execution time and resource cost information of the task corresponding to the task identifier carried in the instruction, and before inputting each parameter value group, the corresponding execution time and resource cost information into the pre-trained reinforcement model, the method further includes:

[0146] According to each parameter value group and the corresponding execution time, a first-order Sobol index and a first overall effect corresponding to each parameter are determined respectively; and according to each parameter value group and the corresponding resource cost information, a second-order Sobol index and a second overall effect corresponding to each parameter are determined respectively;

[0147] For each parameter, determine the impact value of the parameter on the execution time and resource cost information according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter, and if the impact value is lower than a threshold, delete the parameter value corresponding to the parameter in each parameter value group;

[0148] For each retained parameter, perform a subsequent step of inputting each parameter value group, the corresponding execution time, and the resource cost information into the pre-trained reinforcement model.

[0149] In a possible implementation, the method further includes:

[0150] When determining that each parameter of the target tool is the target parameter value output by the reinforcement model, the target execution time and target resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0151] Determining a target reward value according to the target execution time, the target resource cost information and the corresponding weights;

[0152] Based on the target reward value and the pre-stored standard reward value, the reinforcement parameters of the reinforcement model are adjusted.

[0153] In a possible implementation manner, determining the target reward value according to the target execution time, the target resource cost information, and the corresponding weights respectively includes:

[0154] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0155] If the first score is less than the pre-saved target score, obtaining the second score determined last time;

[0156] A target reward value is determined according to a ratio of the second score to the first score.

[0157] In a possible implementation manner, determining the target reward value according to the target execution time, the target resource cost information, and the corresponding weights respectively includes:

[0158] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0159] If the first score is equal to the pre-saved target score, the target reward value is determined to be a preset first reward value; if the first score is greater than the target score, the target reward value is determined to be a preset second reward value; wherein the second reward value is greater than the first reward value.

[0160] In a possible implementation manner, before adjusting the reinforcement parameters of the reinforcement model based on the target reward value and the pre-stored standard reward value, the method further includes:

[0161] If the parameter value of any parameter is not within the parameter value range corresponding to the parameter, the target reward value is determined to be a preset third reward value; wherein the third reward value is less than the second reward value corresponding to when the first score is equal to the pre-saved target score.

[0162] The communication bus mentioned in the above server can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0163] The communication interface 602 is used for communication between the electronic device and other devices.

[0164] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0165] The above-mentioned processor can be a general-purpose processor, including a central processing unit, a network processor (Network Processor, NP), etc.; it can also be a digital signal processing processor (Digital Signal Processing, DSP), an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc.

[0166] Embodiment 10:

[0167] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program executable by an electronic device. When the program runs on the electronic device, the electronic device implements the following steps when executing:

[0168] If an instruction for parameter configuration of a target tool is received, each parameter to be configured of the target tool saved in advance is obtained; a preset number of parameter values ​​of each parameter are randomly generated, and the parameter values ​​of each parameter generated at any time are combined into a corresponding parameter value group, wherein the target tool is Hive or Flink;

[0169] For each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0170] Each parameter value group, the corresponding execution time and resource cost information are input into a pre-trained reinforcement model to obtain a target parameter value for each parameter output by the reinforcement model.

[0171] In a possible implementation manner, the randomly generating a preset number of parameter values ​​for each parameter includes:

[0172] For each parameter, a preset number of parameter values ​​within the parameter value range are randomly generated according to a pre-stored parameter value range corresponding to the parameter.

[0173] In a possible implementation, when determining that each parameter of the target tool is a parameter value in the parameter value group, after executing the execution time and resource cost information of the task corresponding to the task identifier carried in the instruction, and before inputting each parameter value group, the corresponding execution time and resource cost information into the pre-trained reinforcement model, the method further includes:

[0174] According to each parameter value group and the corresponding execution time, a first-order Sobol index and a first overall effect corresponding to each parameter are determined respectively; and according to each parameter value group and the corresponding resource cost information, a second-order Sobol index and a second overall effect corresponding to each parameter are determined respectively;

[0175] For each parameter, determine the impact value of the parameter on the execution time and resource cost information according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter, and if the impact value is lower than a threshold, delete the parameter value corresponding to the parameter in each parameter value group;

[0176] For each retained parameter, perform a subsequent step of inputting each parameter value group, the corresponding execution time, and the resource cost information into the pre-trained reinforcement model.

[0177] In a possible implementation, the method further includes:

[0178] When determining that each parameter of the target tool is the target parameter value output by the reinforcement model, the target execution time and target resource cost information of the task corresponding to the task identifier carried in the instruction are executed;

[0179] Determining a target reward value according to the target execution time, the target resource cost information and the corresponding weights;

[0180] Based on the target reward value and the pre-stored standard reward value, the reinforcement parameters of the reinforcement model are adjusted.

[0181] In a possible implementation manner, determining the target reward value according to the target execution time, the target resource cost information, and the corresponding weights respectively includes:

[0182] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0183] If the first score is less than the pre-saved target score, obtaining the second score determined last time;

[0184] A target reward value is determined according to a ratio of the second score to the first score.

[0185] In a possible implementation manner, determining the target reward value according to the target execution time, the target resource cost information, and the corresponding weights respectively includes:

[0186] Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights;

[0187] If the first score is equal to the pre-saved target score, the target reward value is determined to be a preset first reward value; if the first score is greater than the target score, the target reward value is determined to be a preset second reward value; wherein the second reward value is greater than the first reward value.

[0188] In a possible implementation manner, before adjusting the reinforcement parameters of the reinforcement model based on the target reward value and the pre-stored standard reward value, the method further includes:

[0189] If the parameter value of any parameter is not within the parameter value range corresponding to the parameter, the target reward value is determined to be a preset third reward value; wherein the third reward value is less than the second reward value corresponding to when the first score is equal to the pre-saved target score.

[0190] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0191] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0194] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A parameter configuration method, characterized in that: The method comprises: If an instruction for parameter configuration of a target tool is received, each parameter to be configured of the target tool saved in advance is obtained; a preset number of parameter values ​​of each parameter are randomly generated, and the parameter values ​​of each parameter generated at any time are combined into a corresponding parameter value group, wherein the target tool is Hive or Flink; For each parameter value group, when it is determined that each parameter of the target tool is a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction are executed; Each parameter value group, the corresponding execution time and resource cost information are input into a pre-trained reinforcement model to obtain a target parameter value for each parameter output by the reinforcement model.

2. The method according to claim 1, characterized in that The randomly generating a preset number of parameter values ​​of each parameter includes: For each parameter, a preset number of parameter values ​​within the parameter value range are randomly generated according to a pre-stored parameter value range corresponding to the parameter.

3. The method according to claim 1, characterized in that When determining that each parameter of the target tool is a parameter value in the parameter value group, after executing the execution time and resource cost information of the task corresponding to the task identifier carried in the instruction, and before inputting each parameter value group, the corresponding execution time and resource cost information into the pre-trained reinforcement model, the method further includes: According to each parameter value group and the corresponding execution time, a first-order Sobol index and a first overall effect corresponding to each parameter are determined respectively; and according to each parameter value group and the corresponding resource cost information, a second-order Sobol index and a second overall effect corresponding to each parameter are determined respectively; For each parameter, determine the impact value of the parameter on the execution time and resource cost information according to the first-order Sobol index, the first overall effect, the second-order Sobol index, the second overall effect and the corresponding weights of the parameter, and if the impact value is lower than a threshold, delete the parameter value corresponding to the parameter in each parameter value group; For each retained parameter, perform a subsequent step of inputting each parameter value group, the corresponding execution time, and the resource cost information into the pre-trained reinforcement model.

4. The method according to claim 1, characterized in that: The method further comprises: When determining that each parameter of the target tool is the target parameter value output by the reinforcement model, the target execution time and target resource cost information of the task corresponding to the task identifier carried in the instruction are executed; Determining a target reward value according to the target execution time, the target resource cost information and the corresponding weights; Based on the target reward value and the pre-stored standard reward value, the reinforcement parameters of the reinforcement model are adjusted.

5. The method according to claim 4, characterized in that The determining of the target reward value according to the target execution time, the target resource cost information and the corresponding weights respectively includes: Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights; If the first score is less than the pre-saved target score, obtaining the second score determined last time; A target reward value is determined according to a ratio of the second score to the first score.

6. The method according to claim 4, characterized in that The determining of the target reward value according to the target execution time, the target resource cost information and the corresponding weights respectively includes: Determine a corresponding first score according to the target execution time, the target resource cost information and their corresponding weights; If the first score is equal to the pre-saved target score, the target reward value is determined to be a preset first reward value; if the first score is greater than the target score, the target reward value is determined to be a preset second reward value; wherein the second reward value is greater than the first reward value.

7. The method according to claim 5 or 6, characterized in that: Before adjusting the reinforcement parameters of the reinforcement model based on the target reward value and the pre-stored standard reward value, the method further includes: If the parameter value of any parameter is not within the parameter value range corresponding to the parameter, the target reward value is determined to be a preset third reward value; wherein the third reward value is less than the second reward value corresponding to when the first score is equal to the pre-saved target score.

8. A parameter configuration device, characterized in that: The device comprises: A processing module, configured to, if receiving an instruction for parameter configuration of a target tool, obtain each parameter to be configured of the target tool that is saved in advance; randomly generate a preset number of parameter values ​​of each parameter, and form a corresponding parameter value group with the parameter values ​​of each parameter generated at any time, wherein the target tool is Hive or Flink; A determination module, for determining, for each parameter value group, when each parameter of the target tool is determined to be a parameter value in the parameter value group, the execution duration and resource cost information of the task corresponding to the task identifier carried in the instruction; The acquisition module is used to input each parameter value group, the corresponding execution time and resource cost information into a pre-trained reinforcement model to obtain the target parameter value of each parameter output by the reinforcement model.

9. An electronic device, characterized in that: include: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory, and execute the steps included in the parameter configuration method according to any one of claims 1 to 7 according to the obtained program instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a computer, the computer executes the parameter configuration method according to any one of claims 1 to 7.

Images