A multi-agent system failure log root cause positioning method and system

Abstract

The application belongs to the technical field of artificial intelligence and computer software, and discloses a multi-agent system failure log root cause positioning method and system. The method comprises the following steps: obtaining a failure track by analyzing a failure log; identifying an error step in reverse order; backtracking to analyze an error influence relationship to form a candidate root cause step set; detecting whether the candidate root cause step is affected by a more previous step, and replacing and updating if so; detecting whether the error influence of the candidate root cause step is corrected in subsequent execution, and determining the candidate root cause step whose error influence continues to task failure as a root cause step; and outputting a positioning result. The application can complete root cause positioning based on a failure track without relying on a virtual correction track, has the advantages of low data requirement, strong interpretability, no need for manual intervention, independent encapsulation and deployment, and is suitable for rapid fault diagnosis and performance optimization of a multi-agent system.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and computer software technology, specifically to a method and system for root cause localization of failure logs in multi-agent systems. Background Technology

[0002] With the development of large language model technology, multi-agent systems (MAS) built on large language models (LLMs) can complete complex tasks through collaborative interaction among multiple agents with different roles. During task execution in a MAS, each agent performs task planning, information processing, tool invocation, and result generation according to a certain execution order, forming an execution trajectory that includes action sequences, intermediate outputs, communication content, and state changes. When an error occurs at a step in a certain execution trajectory, this error may be received by subsequent agents and continue to propagate, gradually affecting subsequent execution and ultimately leading to task failure. Therefore, accurately identifying the root cause step and corresponding agent from the failed execution trajectory is of great significance for fault analysis and performance optimization of multi-agent systems.

[0003] Existing failure attribution methods typically rely on comparing the failure trajectory with multiple dummy corrective trajectories to pinpoint the root cause. However, since these dummy corrective trajectories do not exist in reality, interpretable failure attribution is impossible. Furthermore, synthesizing such corrective trajectories requires manual analysis of the potential correct substitutions at each step of the failure trajectory and trajectory regeneration via a multi-agent system, severely hindering the efficiency of failure attribution. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for root cause localization of failure logs in multi-agent systems. This invention aims to liberate failure attribution methods from their dependence on potential corrective trajectories, performing root cause localization solely based on available failure trajectories.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for root cause localization of failure logs in multi-agent systems is provided, including: Obtain and parse the failure logs generated by the multi-agent system during task execution to obtain the failure trajectory; By identifying erroneous steps in the failure trajectory in reverse order, a set of erroneous steps is obtained. Based on the failure trajectory, backtracking forward, analyzing the error influence relationship between the erroneous step and its predecessor steps, identifying the predecessor steps that can cause the erroneous step to fail, and forming a candidate root cause step set; Detect whether the candidate root cause step is affected by an error in the preceding step. If the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. Detect whether the error impact of the candidate root cause steps in the updated candidate root cause step set is corrected in subsequent execution, and identify the candidate root cause steps whose error impact persists until the task fails as root cause steps. The root cause localization result is output based on the root cause step.

[0006] Secondly, a multi-agent system failure log root cause localization system is provided, including: The acquisition and parsing module is configured to: acquire and parse the failure logs generated by the multi-agent system in executing tasks to obtain the failure trajectory; The error step extraction module is configured to: identify the error steps in the failure trajectory in reverse order, and obtain a set of error steps; The candidate root cause step backtracking module is configured to: backtrack forward based on the failure trajectory, analyze the error influence relationship between the erroneous step and its predecessor steps, determine the predecessor steps that can cause the erroneous step to make a mistake, and form a candidate root cause step set; The candidate root cause step verification module is configured to: detect whether the candidate root cause step is affected by an error in a preceding step; if the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. The root cause localization module is configured to: detect whether the error impact of the candidate root cause steps in the updated set of candidate root cause steps is corrected in subsequent execution, and identify the candidate root cause steps whose error impact persists until the task fails as root cause steps. The output module is configured to output the root cause localization result based on the root cause steps.

[0007] Thirdly, an electronic device is also provided, comprising: Memory is used to store computer-readable instructions in a non-transitory manner. Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0008] Fourthly, a computer-readable storage medium is provided having a program stored thereon that, when executed by a processor, implements the method described in the first aspect above.

[0009] Fifthly, a computer program product is provided, employing the following technical solution: A computer program product includes software code, wherein a program in the software code performs the steps of the method described in the first aspect of the present invention.

[0010] The above technical solution has the following advantages or beneficial effects: (1) The theoretical basis of this invention is shifted from the optimization problem that relies on the comparison of failure trajectory and potential correction trajectory to the necessary condition of the root cause step. The attribution process no longer depends on the participation of virtual correction trajectory. The execution trajectory is the only data requirement of the technical solution of this invention, which ensures that this invention works on the basis of complete data.

[0011] (2) The design of each part of the present invention is based on a complete theoretical analysis of the necessary conditions for the root cause step, which ensures the interpretability of the present invention in the attribution process.

[0012] (3) The reasoning process of this invention does not require human intervention and can be encapsulated into any multi-agent system as an independent plug-in. No modifications are required to the original multi-agent system. In addition, no other data collection is required except for the obtainable failure trajectory. The above advantages support the autonomous and rapid response of this invention.

[0013] (4) The present invention does not have virtual data dependency, and the strong interpretability of the reasoning process supports high prediction accuracy. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figure 1 This is a flowchart of the multi-agent system failure log root cause localization method in a specific embodiment of the present invention. Detailed Implementation

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] In this embodiment of the invention, "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of this invention, "multiple" refers to two or more.

[0019] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0020] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0021] All data acquisition in this embodiment is carried out in compliance with laws and regulations and with user consent, and the data is used legally.

[0022] Example 1 Failure attribution is based on a failure trajectory The process involves locating the root cause step that led to task failure, the corresponding time step, and the active agent at that time step. Indicates a time step. Indicates at time step For active agents, existing methods often achieve failure attribution by solving the following mathematical problems: (1); in, For the root cause step, This represents the time index corresponding to the root cause step. This refers to the agent that is active at that time step, i.e., the agent responsible for the failure. This indicates that the trajectory is a failed trajectory, and vice versa. This indicates that the trajectory is a successful trajectory; Representation and failure trajectory The corresponding corrected trajectory is obtained at the time step. , will activate intelligent agents The original behavior is replaced with a potential optimization scheme, and all steps of MAS after that time step are re-executed.

[0023] Guided by the mathematical characterization of the failure attribution problem described above, existing technical solutions mainly identify and locate the cause through two methods: Method 1: Directly call an LLM for failure attribution, which is divided into three detection methods: All-at-Once, Step-by-Step and Binary Search. (1) The All-at-Once method provides all failure trajectories to the LLM for evaluation at once, requiring the LLM to directly output the root cause step index, the agent directly responsible for the failure, and the reason for the judgment. (2) The Step-by-Step method simulates the process of manually reviewing the log line by line, allowing the LLM to check each step in the failure trajectory in chronological order. Once the root cause step is detected, it is immediately output and the detection of subsequent steps is terminated. (3) The Binary Search method adopts a binary search strategy. Starting from the first half of the provided failure trajectory, the LLM is required to determine whether the erroneous step occurred in this part, reduce the search range to half the length of the original trajectory, and then continue to use the above strategy to reduce the search range of the remaining trajectory. This process continues until the interval is reduced to a single step, which is the failure attribution result.

[0024] Method Two: The inference results are progressively refined through iterative interaction between a discriminator and three specialized evaluators. First, the discriminator proposes candidate erroneous steps based on the complete trajectory. Then, the three specialized evaluators independently review the results from three dimensions: "Did an error occur in this step?", "Is the error in this step a decisive error (if there is an optimized alternative that can guarantee that MAS can successfully resolve the query after re-execution from this step)?", and "Is this the earliest decisive error?", and provide confidence scores to assist the discriminator in correcting the prediction results in the next iteration. Finally, through multiple iterations of the above process, the attribution results are progressively corrected.

[0025] Existing technical solutions all rely on unavailable potential corrective trajectories for failure attribution, failing to generate reliable inference results. The failure attribution theoretical framework shown in Equation (1) relies on the comparison between failure trajectories and potential corrective trajectories for root cause localization, but the latter does not exist in real-world scenarios. Since the correct alternatives to agent behavior are often difficult to determine, constructing such corrective trajectories in practice is impractical.

[0026] To address the aforementioned issues, based on the fundamental differences in nature between the root cause step and other steps in the failure trajectory, this embodiment summarizes three necessary conditions that it must satisfy: (1) Error propagation condition: Errors generated in the root cause step usually propagate in subsequent interactions, causing errors in downstream steps. Therefore, there must be at least one subsequent erroneous step. The error is caused by the root cause step. Caused by, among which, This indicates the time step corresponding to the subsequent error step. This indicates the agents that are active at that time step. and These represent the time step and the active agent corresponding to the root cause step, respectively. This condition excludes cases where errors will not propagate to subsequent steps.

[0027] (2) Condition of independent causality: The root cause step is the origin of the error propagation chain, not the propagation error caused by any preceding step. Therefore, for any The steps corresponding to the failure trajectory None of these are direct causes of errors in the root cause step; that is, there is no preceding step. ,in This indicates the time step corresponding to the preceding step. This indicates the agent that is active at that time step. This condition distinguishes the actual source of error in a failure trajectory from its propagation symptoms.

[0028] (3) Persistent Effect Condition: The erroneous effect of the root cause step persists in subsequent interactions and eventually leads to task failure. In other words, no subsequent step can completely correct this erroneous effect. This condition excludes localized errors, whose erroneous effects are cut off in subsequent execution.

[0029] Based on the above three necessary conditions, the root cause steps are redefined in this embodiment as follows: Definition (Root Cause Step): For a given failure trajectory, The root cause step is a step if and only if: Error propagation: ; Causal independence: ; Lasting impact: ; in, This indicates the impact of errors between steps in the trajectory. This indicates the termination time step of the failure trajectory. This indicates an agent that is active at the termination time step.

[0030] Based on the aforementioned necessary conditions and the definition of the root cause step, such as Figure 1 As shown, this embodiment provides a method for root cause localization of failure logs in a multi-agent system, including: S101: Obtain and parse the failure logs generated by the multi-agent system in executing tasks to obtain the failure trajectory; S102: Identify the erroneous steps in the failure trajectory in reverse order to obtain the set of erroneous steps; S103: Based on the failure trajectory, backtrack forward and analyze the error influence relationship between the erroneous step and its predecessor steps to determine the predecessor steps that can cause the erroneous step to make an error, and form a candidate root cause step set. S104: Detect whether the candidate root cause step is affected by the error of the preceding step. If the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. S105: Detect whether the error impact of the candidate root cause steps in the updated candidate root cause step set is corrected in the subsequent execution process, and determine the candidate root cause steps whose error impact continues until the task fails as root cause steps. S106: Output the root cause localization result according to the root cause step.

[0031] Based on an in-depth analysis of the characteristics that distinguish the root cause step from other steps in the failure trajectory, this embodiment summarizes the necessary conditions that the root cause step should meet, and designs a root cause localization method for multi-agent system failure logs under the guidance of these conditions.

[0032] Step S101 is as follows: In this embodiment, the multi-agent system consists of multiple agents with different task roles. Each agent interacts in multiple rounds according to a preset cooperation process during the execution of the target task. The failure log records the execution information generated during the period from the start of the target task to its failure.

[0033] The failure log content is parsed, and the JSON-formatted failure logs are deconstructed and reassembled in chronological order. The specific parsing process includes the following steps: (1) Read the problem description of the target task of the original multi-agent system and the interaction records between agents according to the "question" and "history" keys in the JSON file; (2) In the interaction record, extract the agent identifier and speech content corresponding to each step in the trajectory by using the "name" and "content" keys; (3) Combine the extracted agent identifiers and speech content into a key-value correspondence of “agent identifier: speech content” to intuitively reflect the interaction process between agents; (4) Number the reorganized dialogue process starting from 1 according to the interaction sequence to obtain the following failure trajectory: "Step 1": "Agent Identifier 1: Speech Content 1"; Step 2: "Agent Identifier 2: Speech Content 2"; ... “Step T”: “Agent Identifier T: Speech Content T”.

[0034] Step S102 is as follows: Based on the error propagation conditions defined in the root cause step definition, identifying the error step is a prerequisite for root cause localization. In this stage, this embodiment designs an error step extractor. Identify up to [number] failures from the failure trajectory in reverse order (execution time from back to front). There are 10 error steps, and they are stored in the error step set. In this process, the extraction procedure takes the following form: (2); in, This represents the time step-agent pair corresponding to each step in the trajectory, where ; and These represent the steps after the analysis. and The resulting set of error steps; It is a threshold used to distinguish whether an error has occurred in a step, when the error step extractor... An error is considered to have occurred when the estimated probability of an error in a certain step is higher than this value. express The incorrect step with the lowest score. For set Errors in the process, when there are The erroneous step has been detected, and Compared to When a higher error score is achieved, the former will replace the latter to ensure the set... The system always stores the score with the highest error score. One step.

[0035] The elements in the error step set include the error step index, the name of the agent that caused the error, and a description of the error in natural language format.

[0036] Step S103 is as follows: This step in the embodiment is designed based on the error propagation conditions of the root cause step, aiming to achieve a set The error steps in the process are traced back to the reasons that caused them to fail. This embodiment uses a tracker agent. To execute this process, the tracker agent searches backwards from the execution time of the failed trajectory, retrieving the preceding steps. The input is an incorrect step. And its prerequisite steps, namely, the failure trajectory that satisfies Steps The output is the result of determining whether there is an erroneous influence relationship between the two. If it exists, output 1; otherwise, output 0. The steps tracked in this stage are identified as candidate root cause steps and stored in the candidate root cause step set. middle: (3); The elements in the candidate root cause step set include: candidate root cause step index, name of the agent predicting the candidate root cause, and explanation in natural language format.

[0037] Step S104 is as follows: Guided by the causal independence condition of the root cause steps, this step tests the causal independence of the candidate root cause steps to evaluate the set of candidate root cause steps. To perform the update, this embodiment uses a verifier agent. Steps for testing candidate root causes Each candidate root cause step To determine if it's an independent error source, examine the preceding steps backward from the execution time of the failure trajectory. If a preceding step exists, meaning the failure trajectory satisfies... Steps Leading to candidate root cause steps If an error occurs, the latter step is replaced with the former step in the candidate root cause step set; otherwise, the original candidate root cause steps are retained, and the candidate root cause step set is updated. (4); Step S105 is as follows: For each candidate root cause step in the updated set of candidate root cause steps, the error impact of each candidate root cause step is detected backward according to the execution time of the failure trajectory; the root cause localization agent is then utilized. Determine whether the impact of the error is corrected in subsequent execution steps; if the impact of the error is not corrected and continues until the task fails, the corresponding candidate root cause step is identified as the root cause step, and the root cause step is used as the localization result.

[0038] In this embodiment, the error step extractor, tracker agent, validator agent, and root cause localizer agent are all specialized agents built based on a large language model. The error step extractor, tracker agent, validator agent, and root cause localizer agent all use the DeepSeek-V4-Pro model. It should be understood that those skilled in the art may also choose other large language models.

[0039] Example 2 This embodiment provides a multi-agent system failure log root cause localization system, including: The acquisition and parsing module is configured to: acquire and parse the failure logs generated by the multi-agent system in executing tasks to obtain the failure trajectory; The error step extraction module is configured to: identify the error steps in the failure trajectory in reverse order, and obtain a set of error steps; The candidate root cause step backtracking module is configured to: backtrack forward based on the failure trajectory, analyze the error influence relationship between the erroneous step and its predecessor steps, determine the predecessor steps that can cause the erroneous step to make a mistake, and form a candidate root cause step set; The candidate root cause step verification module is configured to: detect whether the candidate root cause step is affected by an error in a preceding step; if the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. The root cause localization module is configured to: detect whether the error impact of the candidate root cause steps in the updated set of candidate root cause steps is corrected in subsequent execution, and identify the candidate root cause steps whose error impact persists until the task fails as root cause steps. The output module is configured to output the root cause localization result based on the root cause steps.

[0040] It should be noted that the acquisition and parsing module, error step extraction module, candidate root cause step backtracking module, candidate root cause step verification module, root cause localization module, and output module mentioned above correspond to steps S101 to S106 in Embodiment 1. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0041] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0042] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0043] Example 3 This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0044] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0045] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0046] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0047] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0048] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0049] Example 4 Embodiment 4 of the present invention provides a computer-readable storage medium.

[0050] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the method as described in Embodiment 1 of the present invention.

[0051] The detailed steps are the same as those provided in Example 1, and will not be repeated here.

[0052] Example 5 Embodiment 5 of the present invention provides a computer program product.

[0053] A computer program product includes software code, wherein the program in the software code performs the steps described in Embodiment 1 of the present invention.

[0054] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for root cause localization of failure logs in a multi-agent system, characterized in that, include: Obtain and parse the failure logs generated by the multi-agent system during task execution to obtain the failure trajectory; By identifying erroneous steps in the failure trajectory in reverse order, a set of erroneous steps is obtained. Based on the failure trajectory, backtracking forward, analyzing the error influence relationship between the erroneous step and its predecessor steps, identifying the predecessor steps that can cause the erroneous step to fail, and forming a candidate root cause step set; Detect whether the candidate root cause step is affected by an error in the preceding step. If the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. Detect whether the error impact of the candidate root cause steps in the updated candidate root cause step set is corrected in subsequent execution, and identify the candidate root cause steps whose error impact persists until the task fails as root cause steps. The root cause localization result is output based on the root cause step.

2. The multi-agent system failure log root cause localization method as described in claim 1, characterized in that, The root cause step satisfies three necessary conditions: error propagation condition, causal independence condition, and persistent effect condition. The error propagation condition means that the error effect generated by the root cause step is propagated in subsequent interactions. The causal independence condition means that the root cause step is the origin of the error propagation chain, rather than a propagation error caused by any preceding step. The persistent effect condition means that the error effect of the root cause step persists in subsequent interactions and eventually leads to task failure.

3. The multi-agent system failure log root cause localization method as described in claim 1, characterized in that, The step of identifying erroneous steps in the failure trajectory in reverse order to obtain a set of erroneous steps is specifically as follows: the execution steps in the failure trajectory are analyzed sequentially from back to front according to the execution time; the erroneous step extraction agent is used to judge the correctness of the current execution step and determine the corresponding error degree; when the error degree reaches the preset error judgment threshold, the current execution step is determined to be an erroneous step. Add the error step index, the agent that caused the error, and the reason for the error to the error step set.

4. The multi-agent system failure log root cause localization method as described in claim 1, characterized in that, The process of backtracking backward based on the failure trajectory, analyzing the error impact relationship between the erroneous step and its predecessor steps, identifying the predecessor steps that could cause the erroneous step to err, and forming a candidate root cause step set specifically includes: for the erroneous step in the set of erroneous steps, searching for predecessor steps forward according to the execution time of the failure trajectory; using a tracker agent to analyze the error impact relationship between the erroneous step and each predecessor step, and when it is determined that the error generated by the predecessor step can cause the erroneous step to err, identifying the predecessor step as a candidate root cause step, and forming the candidate root cause step set.

5. The multi-agent system failure log root cause localization method as described in claim 1, characterized in that, The step of detecting whether a candidate root cause step is affected by an error in a preceding step, and when such a preceding step exists, replacing the corresponding candidate root cause step with the preceding step and updating the candidate root cause step set, specifically includes: for candidate root cause steps in the candidate root cause step set, detecting preceding steps forward according to the execution time of the failure trajectory; using a validator agent to determine whether the candidate root cause step is affected by an error in the preceding step; when it is determined that the candidate root cause step is affected by an error in the preceding step, removing the candidate root cause step from the candidate root cause step set and adding the preceding step to the candidate root cause step set.

6. The multi-agent system failure log root cause localization method as described in claim 1, characterized in that, The process of detecting whether the error impact of candidate root cause steps in the updated candidate root cause step set is corrected in subsequent execution, and identifying candidate root cause steps whose error impact persists until task failure, specifically includes: for candidate root cause steps in the updated candidate root cause step set, detecting the error impact corresponding to the candidate root cause steps backward according to the execution time of the failure trajectory; using a root cause localization agent to determine whether the error impact is corrected in subsequent execution steps; and when the error impact is not corrected and persists until task failure, identifying the corresponding candidate root cause step as the root cause step.

7. A multi-agent system failure log root cause localization system, characterized in that, include: The acquisition and parsing module is configured to: acquire and parse the failure logs generated by the multi-agent system in executing tasks to obtain the failure trajectory; The error step extraction module is configured to: identify the error steps in the failure trajectory in reverse order, and obtain a set of error steps; The candidate root cause step backtracking module is configured to: backtrack forward based on the failure trajectory, analyze the error influence relationship between the erroneous step and its predecessor steps, determine the predecessor steps that can cause the erroneous step to make a mistake, and form a candidate root cause step set; The candidate root cause step verification module is configured to: detect whether the candidate root cause step is affected by an error in a preceding step; if the preceding step exists, replace the corresponding candidate root cause step with the preceding step and update the candidate root cause step set. The root cause localization module is configured to: detect whether the error impact of the candidate root cause steps in the updated set of candidate root cause steps is corrected in subsequent execution, and identify the candidate root cause steps whose error impact persists until the task fails as root cause steps. The output module is configured to output the root cause localization result based on the root cause steps.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the multi-agent system failure log root cause localization method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps in the multi-agent system failure log root cause localization method as described in any one of claims 1-6.

10. A computer program product, comprising software code, characterized in that, The program in the software code executes the steps in the multi-agent system failure log root cause localization method as described in any one of claims 1-6.

Images