Root cause positioning method and device, equipment and storage medium

Abstract

The invention discloses a root cause positioning method and device, equipment and a storage medium. Comprising the following steps: constructing a proxy model structure based on a pre-trained generative language model; a reinforcement learning network in the proxy model structure is trained to generate a root cause positioning model, and the root cause positioning model comprises a generative language model and a tool selection model; and obtaining an input task, and inputting the input task into the root cause positioning model to obtain an output root cause positioning result. The strong understanding and generating capability of the language model can be fully utilized, complex fault information is analyzed more accurately, and the accuracy of root cause positioning is improved. The root cause analysis is carried out by adopting a method of combining the reinforcement learning network and the generative language model, so that the root cause positioning model can be adaptively adjusted according to different input tasks and system environments, complex and changeable fault conditions can be better handled, the efficiency and accuracy of root cause positioning are further improved, and the fault solving period is shortened.

Description

Technical Field

[0001] The present invention relates to the field of root cause analysis, and in particular to a root cause locating method, device, equipment and storage medium. Background Art

[0002] In today's complex information technology environment, the information systems of enterprises are becoming increasingly large and complex. With the continuous expansion of business and the continuous updating of technology, the possibility of failure in the system is also increasing. When a failure occurs, it is crucial to quickly and accurately find its root cause to reduce losses, restore the normal operation of the system, and prevent similar problems from happening again.

[0003] The existing technology uses the fault location method DejaVu, which establishes a mapping relationship between fault modes and program features by studying and analyzing a large number of known fault cases. Then, when faced with new faults, it can quickly and accurately locate the possible fault location. The existing technology solution directly uses the generative language model without additional training, lacks professional knowledge, and has poor accuracy. When there is additional data training, private domain information can only be injected by fine-tuning the generative language model, but fine-tuning requires a lot of data and resources, a large workload, and will affect the model's existing natural language understanding ability. Summary of the invention

[0004] The present invention provides a root cause location method, device, equipment and storage medium, which enhance the accuracy of root cause location by combining the decision-making ability of reinforcement learning and the text understanding ability of a generative language model.

[0005] According to one aspect of the present invention, a root cause location method is provided, the method comprising:

[0006] Build a proxy model structure based on a pre-trained generative language model;

[0007] The reinforcement learning network in the agent model structure is trained to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model;

[0008] An input task is obtained and input into the root cause localization model to obtain an output root cause localization result.

[0009] Optionally, building a proxy model structure based on a pre-trained generative language model includes: obtaining the pre-trained generative language model as a base model; obtaining a reinforcement learning network and a toolset, and combining the reinforcement learning network, the toolset, and the base model to generate a proxy model structure.

[0010] Optionally, the reinforcement learning network in the agent model structure is trained to generate a root cause localization model, including: designing a reward function, wherein the reward function includes task completion and step efficiency; deploying an initial environment, and injecting faults into the initial environment based on a chaos engineering tool to generate a training environment; training the reinforcement learning network based on the training environment and the reward function to generate a tool selection model; fixing the weight of a generative language model, and combining the generative language model with the tool selection model to generate a root cause localization model.

[0011] Optionally, the input task is input into the root cause localization model to obtain an output root cause localization result, including: summarizing the current status of the input task based on the generative language model; judging whether the current status summary is completed, and if so, directly outputting the root cause localization result based on the generative language model; otherwise, determining the target tool through the tool selection model, and determining the root cause localization result according to the target tool through the generative language model.

[0012] Optionally, the target tool is determined through a tool selection model, including: determining a tool selection method, wherein the tool selection method includes random selection or selection by tool value; when the tool selection method is random selection, randomly selecting the target tool from the tool set; when the tool selection method is selection by tool value, determining the tool value of each tool in the tool set, and selecting the tool with the highest tool value as the target tool.

[0013] Optionally, a root cause location result is determined according to a target tool through a generative language model, including: extracting tool parameters of the target tool; executing the tool based on the tool parameters and obtaining the tool execution result, and determining a tool completion status based on the tool execution result; when the tool completion status is completed, generating a root cause location result according to the tool execution result.

[0014] Optionally, after determining the completion status of the tool based on the tool execution result, the method further includes: when the tool completion status is incomplete, performing iterative calculations through a root cause location model until a root cause location result is generated.

[0015] According to another aspect of the present invention, a root cause locating device is provided, the device comprising:

[0016] The proxy model structure building module is used to build the proxy model structure based on the pre-trained generative language model;

[0017] A root cause localization model generation module is used to train the reinforcement learning network in the agent model structure to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model;

[0018] The root cause localization result output module is used to obtain an input task and input the input task into the root cause localization model to obtain an output root cause localization result.

[0019] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0020] at least one processor;

[0021] and a memory communicatively coupled to the at least one processor;

[0022] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a root cause location method described in any embodiment of the present invention.

[0023] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a root cause location method described in any embodiment of the present invention when executed.

[0024] The technical solution of the embodiment of the present invention can make full use of the powerful understanding and generation capabilities of the language model to perform more accurate analysis of complex fault information and improve the accuracy of root cause location. The root cause analysis is performed by combining the reinforcement learning network and the generative language model, so that the root cause location model can be adaptively adjusted according to different input tasks and system environments, better cope with complex and changeable fault conditions, further improve the efficiency and accuracy of root cause location, and shorten the fault resolution cycle.

[0025] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 is a flow chart of a root cause location method provided according to Embodiment 1 of the present invention;

[0028] Figure 2 is a flowchart of another root cause locating method provided according to Embodiment 1 of the present invention;

[0029] Figure 3 is a flowchart of another root cause location method provided according to Embodiment 2 of the present invention;

[0030] Figure 4 is a schematic diagram of the structure of a root cause locating device provided according to Embodiment 3 of the present invention;

[0031] Figure 5 It is a structural schematic diagram of an electronic device for implementing a root cause locating method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] Embodiment 1

[0035] Figure 1 A flowchart of a root cause location method is provided for the first embodiment of the present invention. This embodiment is applicable to root cause analysis scenarios. The method can be performed by a root cause location device. The root cause location device can be implemented in the form of hardware and / or software. The root cause location device can be configured in a computer controller. Figure 1 As shown, the method includes:

[0036] S110. Build a proxy model structure based on the pre-trained generative language model.

[0037] Among them, the pre-trained generative language model refers to a model that has been pre-trained on massive text data, such as GPT-3, BERT, etc. The pre-trained generative language model can generate new text content based on the learned language rules and semantic relationships, and has strong language understanding and generation capabilities, providing basic text processing capabilities for the subsequent construction of complex models. The proxy model structure refers to an intermediate architecture built based on existing models or components. In this embodiment, it refers to the use of pre-trained generative language models to integrate other elements to form a new architecture as the basic framework for subsequent training customization, simplifying the difficulty of directly building a new complex model.

[0038] Specifically, the proxy model structure is built based on the pre-trained generative language model because the generative language model has learned common language patterns, grammatical rules and semantic associations on massive text data. Using its parameters and architecture, the initial framework adapted to the root cause location task can be quickly built, reducing the huge amount of data and computing power required to train the model from scratch.

[0039] Optionally, building a proxy model structure based on a pre-trained generative language model includes: obtaining the pre-trained generative language model as a base model; obtaining a reinforcement learning network and a toolset, and combining the reinforcement learning network, the toolset, and the base model to generate a proxy model structure.

[0040] Among them, the use of pre-trained generative language models as the basis can quickly give the agent model preliminary text understanding and processing capabilities, avoid the high cost and long period of training from scratch, and use its converged parameters to lay a good foundation for subsequent customized fine-tuning. The toolset can be a series of external programs, algorithm libraries, or data query interfaces. For example, the knowledge graph query tool, when the model processes text, if structured knowledge needs to be supplemented, it can use it to quickly retrieve entity relationships; or computing tools in specific fields, such as risk assessment algorithms in the financial field, are used to assist in making accurate decisions.

[0041] In a specific embodiment, the generative language model includes an LLaMA base, which is used to encode the input text summary and generate a high-dimensional vector representation, in which the semantic information of the input text is encoded. The reinforcement learning network includes a Transformer layer, which is used to further process the vector encoding generated by the LLaMA base. The reinforcement learning network can adopt a fully connected neural network structure, and the output is a k-dimensional vector, where k is the number of tools, and each value represents the value of the corresponding tool.

[0042] S120: training the reinforcement learning network in the agent model structure to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model.

[0043] Among them, the reinforcement learning network refers to a network system based on intelligent agents, environments, and reward feedback mechanisms. The intelligent agent performs actions in the environment, and the environment gives rewards or penalties based on the effects of the actions. The intelligent agent uses this to adjust its strategy, repeatedly iterates and optimizes its decisions, and finally learns the best way to achieve its goals. The root cause location model refers to a model used to determine the root cause of the problem. It integrates the generative language model and the tool selection model. The former processes text and sorts out clues, while the latter calls external tools as needed to assist and jointly lock the root cause. In root cause location, the generative language model can interpret the input, describe the fault status, and help derive the root cause. In root cause location, the tool selection model is responsible for evaluating the applicability and value of different tools, selecting appropriate tools from the toolset according to the progress of the task, assisting in digging deep into the problem, and expanding the model's ability to solve complex scenarios.

[0044] Specifically, the reinforcement learning network introduces a reward function mechanism, allowing the model to optimize its decision-making through continuous trial and error. In the root cause location scenario, the intelligent model performs a series of actions, such as screening possible causes and evaluating the relevance of different factors. Each action will receive a reward from environmental feedback. If the cause of the location is more accurate, the reward value will be higher; otherwise, it will receive a lower reward or even a penalty.

[0045] S130: Obtain an input task, and input the input task into a root cause localization model to obtain an output root cause localization result.

[0046] The input task refers to the original problem description provided by the user to the root cause location model, which is mostly in the form of text and contains relevant information of various types of faults to be investigated, such as system errors and process anomalies. The root cause location result refers to the final output of the root cause location model, which includes the core cause of the problem and provides a key basis for subsequent repair and rectification operations.

[0047] Figure 2 A flowchart of a root cause location method is provided for the first embodiment of the present invention. Step S130 mainly includes the following steps S131 to S134:

[0048] S131. Summarize the current status of the input task based on the generative language model.

[0049] S132, determine whether the current status summary is completed, if so, execute S133, otherwise, execute S134.

[0050] S133. Output the root cause location result directly based on the generative language model.

[0051] S134. Determine a target tool through a tool selection model, and determine a root cause location result based on the target tool through a generative language model.

[0052] Among them, the generative language model can summarize the current state of the input task, that is, the model will parse the text, identify the key entities, events and the logical relationship between them, and then use the language patterns it has learned to integrate the fragmented information into the current state summary, and determine whether the current state summary is complete. "Complete" means that the generative language model's initial understanding and summary of the task alone is sufficient to accurately locate the root cause. If the current state summary is clear enough and highly directional, and all fault-related clues are fully presented, the current state can be determined to be "completed". At this time, the generative language model can rely on the massive knowledge and pattern recognition capabilities accumulated in the previous pre-training to directly extract the corresponding root cause information from the summarized state and output accurate positioning results. If the current state summary is not "completed", it means that the existing information is not enough to lock the root cause and additional assistance is needed. That is, at this time, the target tool can be determined through the tool selection model, and then the root cause location process can be improved through the target tool.

[0053] Optionally, the target tool is determined through a tool selection model, including: determining a tool selection method, wherein the tool selection method includes random selection or selection by tool value; when the tool selection method is random selection, randomly selecting the target tool from the tool set; when the tool selection method is selection by tool value, determining the tool value of each tool in the tool set, and selecting the tool with the highest tool value as the target tool.

[0054] It should be noted that in the root cause location process, the tool selection model needs to decide what strategy to use to select the right tool from the tool set. The tool selection methods include random selection or selection by tool value. When the random selection method is used, it does not rely on complex evaluation and sorting, and purely relies on probability to extract tools from the tool set. In some scenarios, especially when the estimated differences in tool effects are not large, or when exploring the problem domain for the first time and lacking sufficient prior knowledge, random selection can quickly give a usable direction to try, avoiding the need for too much effort in the complex evaluation process in the early stage. When the tool value selection method is used, the tool selection model will make decisions based on the potential value that each tool can bring to solving the current problem. That is, at this time, the tool selection model can determine the tool value of each tool in the tool set according to the reward function in the reinforcement learning network, and then use the tool with the highest tool value as the target tool to ensure that in the current root cause location task, the most valuable target tool is called to assist in digging deep into the root cause of the problem.

[0055] Optionally, a root cause location result is determined according to a target tool through a generative language model, including: extracting tool parameters of the target tool; executing the tool based on the tool parameters and obtaining the tool execution result, and determining a tool completion status based on the tool execution result; when the tool completion status is completed, generating a root cause location result according to the tool execution result.

[0056] It should be noted that because different tools have different requirements for input parameter formats and types, the generative language model will use its powerful text understanding ability to convert the fuzzy requirements in the task into the parameters required by the target tool to ensure the smooth execution of the tool. Specifically, after selecting the target tool, the generative language model will extract the parameters required for the startup and operation of the target tool. Taking the database query tool as an example, the generative language model will extract key parameter information such as query keywords, screening conditions, time range, etc. from the task description and previous analysis results.

[0057] Furthermore, after executing the tool based on the tool parameters, the generative language model will also determine the tool execution results and the tool completion status. For example, call the system performance monitoring tool, configure the extracted parameters, and the tool starts to collect real-time data such as CPU usage and memory usage as the tool execution results. To determine whether the tool has completed the task, it depends on whether the execution result meets the expected goal. For simple tools, the results are clear and clear, such as file search tools, which are completed when the target file is found; for more complex tools, such as tools that simulate software fault reproduction, it is necessary to consider whether the fault is fully reproduced and whether the relevant error reports are accurately output. It is considered completed only when all preset indicators are met. When the tool completion status is completed, the generative language model can take over the analysis of the tool execution results. For example, if it is the data packet information given by the network packet capture tool, the model will dig deep into the abnormal traffic direction, the source of the frequently retransmitted data packets, and other clues, and combine the knowledge accumulated in the early stage with the task summary to sort out the cause and effect chain, and finally output the root cause location results such as "due to network congestion between server A and server B, data transmission delays, causing business jams".

[0058] Optionally, after determining the completion status of the tool based on the tool execution result, the method further includes: when the tool completion status is incomplete, performing iterative calculations through a root cause location model until a root cause location result is generated.

[0059] Among them, when the tool completion status is incomplete, it means that the currently called tool has not achieved the expected goal and has not provided enough effective information to identify the root cause. At this time, the root cause location model will start iterative calculations and repeat the above steps. That is, the generative language model will update the task status summary based on the existing results and re-sort out the key information of the problem. The tool selection model will also intervene again to reflect on whether the previous tool selection was wrong and re-evaluate the potential value of each tool in the toolset. The newly selected tool is run again with the updated parameters to obtain a new round of execution results, and then the completion status of this tool is judged. By repeating the above steps and performing iterative calculations repeatedly, each round of iteration narrows the scope of the problem and makes the information more accurate until the root cause location results that meet the requirements are successfully output.

[0060] The technical solution of the embodiment of the present invention can make full use of the powerful understanding and generation capabilities of the language model to perform more accurate analysis of complex fault information and improve the accuracy of root cause location. The root cause analysis is performed by combining the reinforcement learning network and the generative language model, so that the root cause location model can be adaptively adjusted according to different input tasks and system environments, better cope with complex and changeable fault conditions, further improve the efficiency and accuracy of root cause location, and shorten the fault resolution cycle.

[0061] Embodiment 2

[0062] Figure 3 This is a flowchart of a root cause location method provided in the second embodiment of the present invention. This embodiment adds a specific process of training the reinforcement learning network in the proxy model structure to generate a root cause location model based on the above-mentioned first embodiment. Among them, the specific contents of steps S210 and S260 are roughly the same as steps S110 and S130 in the first embodiment, so they will not be repeated in this embodiment. Figure 3 As shown, the method includes:

[0063] S210. Build a proxy model structure based on the pre-trained generative language model.

[0064] Optionally, building a proxy model structure based on a pre-trained generative language model includes: obtaining the pre-trained generative language model as a base model; obtaining a reinforcement learning network and a toolset, and combining the reinforcement learning network, the toolset, and the base model to generate a proxy model structure.

[0065] S220. Design a reward function, wherein the reward function includes task completion and step efficiency.

[0066] The reward function consists of two parts: task completion and step efficiency. Task completion means that if the agent successfully completes the task, a positive reward R_completed is given. Step efficiency means that a negative reward R_step penalty is given for each step. R_step penalty is a positive number representing the penalty for each step. The formula of the reward function is: R = a*R_completed-b*R_step penalty*N_steps. Among them, N_steps represents the number of steps from the beginning to now, a and b represent the weights of different components. Note that R_completed needs to be much larger than R_step penalty, such as 1000 and 1.

[0067] It can be known that in the root cause location task, if the model successfully finds the real root cause of the fault, it indicates that the task completion is high and a high reward should be given. For example, when investigating the cause of a software system crash, the model accurately points out that it is a memory leak problem in a certain section of code, which is close to perfect completion of the task and corresponds to a high reward value. On the contrary, if the given cause is far from the actual cause, the reward is extremely low or even negative. Step efficiency refers to the number of reasoning steps taken by the model to achieve its goal. Efficient models can lock the root cause with fewer steps and should receive more rewards. For example, one method finds the root cause after 5 analysis steps, and the other takes 10 steps. Even if both are correct, the former has an advantage in step efficiency because it has fewer steps.

[0068] S230. Deploy the initial environment and inject faults into the initial environment based on chaos engineering tools to generate a training environment.

[0069] Specifically, when deploying the initial environment, you can first build an infrastructure that is close to the real application scenario, such as simulating an e-commerce transaction system, including front-end web pages, back-end servers, databases and other components. Each component runs according to the standard configuration to form a stable initial state as the starting point for subsequent changes. Then you can use chaos engineering tools, such as chaosmesh tools, to create faults in the initial environment randomly or in a specific pattern. For example, suddenly interrupt the database connection and cause some servers to overload and freeze. This generates a training environment rich in various fault samples, which helps the reinforcement learning network accumulate experience in dealing with real problems.

[0070] S240: Train the reinforcement learning network based on the training environment and the reward function to generate a tool selection model.

[0071] Specifically, in a training environment full of faults, the reinforcement learning network starts to act. Every time it chooses to call a diagnostic tool or execute a specific troubleshooting strategy, the environment will give corresponding feedback, combined with a pre-designed reward function to evaluate the pros and cons of each action. If the tool is called to accurately locate the fault and a high reward is obtained, the network will strengthen the relevant decision path; conversely, low rewards prompt it to adjust its strategy. With countless such interactions and iterative optimizations, the network gradually learns how to choose tools intelligently, and eventually generates a tool selection model that is good at calling external resources on demand.

[0072] It can be known that training the reinforcement learning network based on a specific training environment and reward function to generate a tool selection model can accurately train the model to make more reasonable tool selection decisions and improve the accuracy and efficiency of root cause location.

[0073] S250, fixing the weight of the generative language model, and combining the generative language model with the tool selection model to generate a root cause location model.

[0074] It should be noted that the generative language model has been pre-trained on large-scale data and has mature language understanding and generation capabilities. Fixing its weights can prevent subsequent fine-tuning from destroying the model structure, ensuring stable output of high-quality text parsing and preliminary inference. Ultimately, the text processing advantages of the generative language model can be combined with the tool selection model's ability to flexibly call external resources and make accurate decisions. When faced with the input of the root cause location task, the generative language model first sorts out the semantics and mines key information, and then the tool selection model decides which tool to enable to assist in deep mining. The two work together to form a complete root cause location model.

[0075] S260: Obtain an input task, and input the input task into a root cause localization model to obtain an output root cause localization result.

[0076] Optionally, the input task is input into the root cause localization model to obtain an output root cause localization result, including: summarizing the current status of the input task based on the generative language model; judging whether the current status summary is completed, and if so, directly outputting the root cause localization result based on the generative language model; otherwise, determining the target tool through the tool selection model, and determining the root cause localization result according to the target tool through the generative language model.

[0077] Optionally, the target tool is determined through a tool selection model, including: determining a tool selection method, wherein the tool selection method includes random selection or selection by tool value; when the tool selection method is random selection, randomly selecting the target tool from the tool set; when the tool selection method is selection by tool value, determining the tool value of each tool in the tool set, and selecting the tool with the highest tool value as the target tool.

[0078] Optionally, a root cause location result is determined according to a target tool through a generative language model, including: extracting tool parameters of the target tool; executing the tool based on the tool parameters and obtaining the tool execution result, and determining a tool completion status based on the tool execution result; when the tool completion status is completed, generating a root cause location result according to the tool execution result.

[0079] Optionally, after determining the completion status of the tool based on the tool execution result, the method further includes: when the tool completion status is incomplete, performing iterative calculations through a root cause location model until a root cause location result is generated.

[0080] The technical solution of the embodiment of the present invention, by designing a reward function that includes task completion and step efficiency, can clearly guide the reinforcement learning network to learn in the direction of completing tasks more efficiently and accurately, avoid blind exploration, and improve training effects. By deploying the initial environment and injecting faults using chaos engineering tools, the generated training environment is closer to the actual complex and changeable system operation conditions, making the trained model more robust and adaptable. The weights of the generative language model are fixed to avoid unnecessary changes in training, and at the same time, it is combined with the tool selection model to generate a root cause location model, achieving a balance between stability and flexibility, ensuring that the model has accurate tool selection capabilities based on the use of existing language understanding capabilities, thereby improving the overall root cause location performance.

[0081] Embodiment 3

[0082] Figure 4 This is a schematic diagram of the structure of a root cause location device provided in Embodiment 3 of the present invention. Figure 4 As shown, the apparatus includes: a proxy model structure construction module 310, for constructing a proxy model structure based on a pre-trained generative language model;

[0083] A root cause localization model generation module 320 is used to train the reinforcement learning network in the agent model structure to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model;

[0084] The root cause localization result output module 330 is used to obtain an input task, and input the input task into the root cause localization model to obtain an output root cause localization result.

[0085] Optionally, the proxy model structure construction module 310 is specifically used to: obtain a pre-trained generative language model as a basic model; obtain a reinforcement learning network and a toolset, and combine the reinforcement learning network, the toolset and the basic model to generate a proxy model structure.

[0086] Optionally, the root cause localization model generation module 320 is specifically used to: design a reward function, wherein the reward function includes task completion and step efficiency; deploy an initial environment, and inject faults into the initial environment based on a chaos engineering tool to generate a training environment; train a reinforcement learning network based on the training environment and the reward function to generate a tool selection model; and fix the weights of a generative language model, and combine the generative language model with the tool selection model to generate a root cause localization model.

[0087] Optionally, the root cause localization result output module 330 is specifically used to: summarize the current status of the input task based on the generative language model; determine whether the current status summary is completed, and if so, directly output the root cause localization result based on the generative language model; otherwise, determine the target tool through the tool selection model, and determine the root cause localization result according to the target tool through the generative language model.

[0088] Optionally, the root cause location result output module 330 specifically includes: a target tool determination unit, used to: determine a tool selection method, wherein the tool selection method includes random selection or selection by tool value; when the tool selection method is random selection, randomly select the target tool from the tool set; when the tool selection method is selection by tool value, determine the tool value of each tool in the tool set, and select the tool with the highest tool value as the target tool.

[0089] Optionally, the root cause location result output module 330 specifically includes: a root cause location result determination unit, which is used to: extract tool parameters of the target tool; execute the tool based on the tool parameters, and obtain the tool execution result, and determine the tool completion status based on the tool execution result; when the tool completion status is completed, generate a root cause location result according to the tool execution result.

[0090] Optionally, the root cause localization result output module 330 further includes: an iterative calculation unit, which is used to, after determining the tool completion status based on the tool execution result, when the tool completion status is incomplete, perform iterative calculation through the root cause localization model until a root cause localization result is generated.

[0091] The technical solution of the embodiment of the present invention can make full use of the powerful understanding and generation capabilities of the language model to perform more accurate analysis of complex fault information and improve the accuracy of root cause location. The root cause analysis is performed by combining the reinforcement learning network and the generative language model, so that the root cause location model can be adaptively adjusted according to different input tasks and system environments, better cope with complex and changeable fault conditions, further improve the efficiency and accuracy of root cause location, and shorten the fault resolution cycle.

[0092] A root cause locating device provided in an embodiment of the present invention can execute a root cause locating method provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0093] Embodiment 4

[0094] Figure 5A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0095] like Figure 5 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0096] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0097] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a root cause location method.

[0098] In some embodiments, a root cause location method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of a root cause location method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform a root cause location method in any other appropriate manner (e.g., by means of firmware).

[0099] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0100] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0101] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0102] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0103] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0104] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0105] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0106] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A root cause location method, characterized in that: include: Build a proxy model structure based on a pre-trained generative language model; Training the reinforcement learning network in the proxy model structure to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model; An input task is obtained, and the input task is input into the root cause localization model to obtain an output root cause localization result.

2. The method according to claim 1, characterized in that The construction of the proxy model structure based on the pre-trained generative language model includes: Get a pre-trained generative language model as the base model; A reinforcement learning network and a tool set are obtained, and the reinforcement learning network, the tool set and the base model are combined to generate the proxy model structure.

3. The method according to claim 1, characterized in that The step of training the reinforcement learning network in the agent model structure to generate a root cause location model includes: Designing a reward function, wherein the reward function includes task completion and step efficiency; Deploy an initial environment, and inject faults into the initial environment based on a chaos engineering tool to generate a training environment; Training the reinforcement learning network based on the training environment and the reward function to generate a tool selection model; The weight of the generative language model is fixed, and the generative language model is combined with the tool selection model to generate a root cause localization model.

4. The method according to claim 2, characterized in that: The step of inputting the input task into the root cause localization model to obtain an output root cause localization result includes: Summarizing a current state of the input task based on the generative language model; Determine whether the current state summary is completed, and if so, directly output a root cause location result based on the generative language model; Otherwise, a target tool is determined by the tool selection model, and a root cause location result is determined according to the target tool by the generative language model.

5. The method according to claim 4, characterized in that Determining the target tool through the tool selection model includes: Determining a tool selection method, wherein the tool selection method includes random selection or selection based on tool value; When the tool selection mode is random selection, a target tool is randomly selected from the tool set; When the tool selection method is selection by tool value, the tool value of each tool in the tool set is determined, and the tool with the highest tool value is used as the target tool.

6. The method according to claim 4, characterized in that Determining the root cause location result according to the target tool by using the generative language model includes: Extracting tool parameters of the target tool; Executing a tool based on the tool parameters, obtaining a tool execution result, and determining a tool completion status based on the tool execution result; When the tool completion status is completion, a root cause location result is generated according to the tool execution result.

7. The method according to claim 6, characterized in that After determining the tool completion status based on the tool execution result, the method further includes: When the tool completion status is incomplete, iterative calculation is performed through the root cause location model until a root cause location result is generated.

8. A root cause locating device, characterized in that: include: The proxy model structure building module is used to build the proxy model structure based on the pre-trained generative language model; A root cause localization model generation module, used for training the reinforcement learning network in the proxy model structure to generate a root cause localization model, wherein the root cause localization model includes a generative language model and a tool selection model; The root cause localization result output module is used to obtain an input task, and input the input task into the root cause localization model to obtain an output root cause localization result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The computer storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method according to any one of claims 1 to 7 when executed.

Images