Server fault processing method and device and medium

Abstract

The invention discloses a server fault processing method and device and a medium, and relates to the technical field of computers. And the server fault analysis problem is converted to obtain vector information, and the vector information is input to the pre-training model for reasoning analysis, so that accurate fault positioning is realized, and the search accuracy of fault analysis is improved. The fault analysis problem comprises a potential fault analysis problem, so that unknown faults are predicted and reasoned, and the fault processing delay is reduced. In inference analysis of a pre-training model, corresponding target inference strategies are matched based on priorities of different fault modes and association degrees of multiple faults instead of being directly processed one by one according to the priorities of the faults, and the association degrees of the multiple faults and whether the corresponding inference strategies are conflicted or not are concerned firstly; according to the method, the situation that secondary faults are caused by contradictory occurrence of heavy faults under the condition that the faults with low priorities are processed in a delayed mode is avoided, the fault maintenance progress is further accelerated, and the reliability of the server is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a server failure processing method, device and medium. Background Art

[0002] During the matching process of various technology stack combinations on the server, the server fault handling process relies on manual search based on the mapping relationship corresponding to the pre-stored fault analysis. Since conventional fault handling methods are all performed after a server fault has occurred, the normal processing progress of the server is delayed to a certain extent. Fault analysis for simple problems can be found immediately, but when multiple faults intersect, the priority of the fault is directly determined by the urgency of the fault to implement a fault strategy solution. There will be delays in handling faults with lower priority. At the same time, when multiple faults occur, their corresponding fault resolution strategies may conflict with each other. If they are resolved one by one, secondary server failures may occur during the resolution process, further reducing the repair progress of the fault and reducing the reliability of the server.

[0003] Therefore, how to speed up the fault repair progress to ensure the normal operation of the server is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0004] The purpose of the present invention is to provide a server fault handling method, device and medium to solve the problems of delayed fault repair progress and reduced reliability.

[0005] To solve the above technical problems, the present invention provides a server failure handling method, comprising:

[0006] Receiving a fault analysis question; wherein the fault analysis question includes a potential fault analysis question;

[0007] Converting the fault analysis problem into vector information; and inputting the vector information into a pre-trained model for reasoning analysis;

[0008] In the reasoning analysis of the pre-trained model, the vector information is divided into faults to obtain a corresponding target fault mode and the degree of correlation of multiple faults under the target fault mode; the target reasoning strategy corresponding to the pre-trained model is matched according to the priority of the corresponding target fault mode and the degree of correlation of the multiple faults;

[0009] After completing the reasoning analysis of the pre-training model, the target reasoning strategy is output as the response information of the pre-training model.

[0010] On the one hand, the fault mode includes at least explicit faults, occasional faults and latent faults; the fault classification of the vector information is performed to obtain the corresponding target fault mode and the correlation degree of multiple faults under the target fault mode, including:

[0011] Pre-acquire the vector knowledge database and the server's current vulnerability information;

[0012] Matching the vector information, the current vulnerability information, and the vector knowledge database to determine a target failure mode of the vector information;

[0013] determining the number of faults corresponding to the vector information under the same target fault mode;

[0014] In response to the instruction that the number of faults is greater than 1, the degree of correlation of the multiple faults is determined based on multiple relationships of the corresponding faults under the same target fault mode.

[0015] On the other hand, the correlation degree of multiple faults is determined based on the multiple relationships of the corresponding faults under the same target fault mode, including:

[0016] Obtaining the fault phenomena, time series, causal relationships, and dependency relationships of the multiple faults;

[0017] The correlation degree of the multiple faults is determined based on the fault phenomenon, the time sequence, the causal relationship and the dependency relationship.

[0018] On the other hand, matching the target reasoning strategy corresponding to the pre-trained model according to the priority of the corresponding target fault mode and the correlation degree of the multiple faults includes:

[0019] Establish a mapping relationship between failure modes and corresponding reasoning strategies in advance;

[0020] In response to an instruction that the target fault mode is one, matching a corresponding target reasoning strategy according to the mapping relationship and the degree of association of the multiple faults;

[0021] In response to the instruction that the target fault mode is multiple, the corresponding mapping relationship and the corresponding correlation degree of the multiple faults are determined one by one according to the priority of the target fault mode to match the corresponding target reasoning strategy.

[0022] On the other hand, when the target fault mode is an explicit fault, matching a corresponding target reasoning strategy according to the mapping relationship and the correlation degree of the multiple faults includes:

[0023] In response to the instruction that multiple faults occur on the same server, a knowledge graph corresponding to server hardware components, middleware, fault types, fault symptoms, and handling strategies is pre-built, and the relationships between entities in the knowledge graph are represented in the form of triples of a resource description framework; the vector information of the multiple faults is matched using regular expressions to determine a target problem template;

[0024] Convert the target question template into a corresponding query statement; and determine a corresponding query result based on the query statement and the knowledge graph database;

[0025] Generate a corresponding target reasoning strategy based on the entity relationship of the knowledge graph, the corresponding query results and the correlation degree of the multiple faults;

[0026] Alternatively, in response to an instruction that multiple faults occur on multiple servers, a knowledge graph corresponding to server hardware components, middleware, fault types, fault symptoms, and processing strategies is pre-built in parallel, so as to represent the relationships between entities of the knowledge graph in the form of triples of a resource description framework;

[0027] Performing matching processing on the vector information of the multiple faults by using regular expressions to determine a target problem template;

[0028] Convert the target question template into a corresponding query statement; and determine a corresponding query result based on the query statement and the knowledge graph database;

[0029] Generate a corresponding target reasoning strategy based on the entity relationships of the knowledge graph and the corresponding query results.

[0030] On the other hand, when the target failure mode is an occasional failure, matching a corresponding target reasoning strategy according to the mapping relationship and the correlation degree of the multiple failures includes:

[0031] Acquire historical case data containing multiple faults; and construct a fault correlation map based on the historical case data;

[0032] Combining the multiple faults to generate a first embedding vector;

[0033] Determining a similarity between the first embedding vector and a historical fault embedding vector;

[0034] Filtering target historical case data similar to the vector information based on the relationship between the similarity and the similarity threshold;

[0035] Extracting features from the target historical case data to obtain historical correlation degrees of multiple faults;

[0036] determining a matching degree of the association degree based on the historical association degree and the association degree;

[0037] Determining a corresponding root cause hypothesis from the historical case data according to the matching degree;

[0038] The root cause hypothesis is verified to determine the cause corresponding to the root cause hypothesis as the target reasoning strategy.

[0039] On the other hand, when the target fault mode is a potential fault, matching a corresponding target reasoning strategy according to the mapping relationship and the correlation degree of the multiple faults includes:

[0040] Constructing a fault correlation graph based on the correlation degree of the multiple faults;

[0041] Determine fault nodes and edges based on the multiple faults; wherein the node represents a fault; an edge represents an association relationship between faults, and the weight of the edge represents the degree of association;

[0042] Modeling the fault association graph according to the fault nodes and edges to determine a graph neural network model;

[0043] Calculate the abnormal probability of the faulty node through forward propagation;

[0044] Rank the root causes of the abnormal probabilities of the faulty nodes to determine potential faults;

[0045] The processing strategy corresponding to the potential fault is used as the target reasoning strategy.

[0046] On the other hand, when the response information outputted in response to the training process of the pre-trained model does not conform to the instruction, the method further includes:

[0047] Obtain actual response information for historical fault analysis questions;

[0048] The pre-training model is subjected to reinforcement learning according to the degree of deviation between the response information and the actual response information to obtain a new pre-training model.

[0049] To solve the above technical problems, the present invention further provides a server fault handling device, comprising:

[0050] memory for storing computer programs;

[0051] A processor is used to implement the steps of the server failure handling method when executing the computer program.

[0052] To solve the above technical problems, the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the server failure handling method as described above are implemented.

[0053] The beneficial effects of the present invention are as follows: first, the server fault analysis problem is converted into vector information, and its vector information is input into the pre-trained model for reasoning analysis. Compared with the conventional technical solution that uses a pre-stored fault mapping relationship for manual search, the present invention uses the reasoning mechanism of the pre-trained model to obtain the target reasoning strategy, achieves accurate fault location, and improves the accuracy of fault analysis. Second, the corresponding fault analysis problem in the present invention includes a potential fault analysis problem. The pre-trained model can be used to determine the corresponding fault mode and then match the target reasoning strategy to achieve predictive reasoning for unknown faults, so as to prevent the occurrence of faults in advance and reduce the delay of fault processing. Third, for multiple faults, the present invention performs fault partitioning processing on the vector information in the reasoning analysis of the pre-trained model to determine which fault mode it is and whether it is multiple faults under the same fault mode. At this time, based on the partitioning process, the corresponding target fault mode and the degree of correlation of multiple faults under the model are determined, so as to facilitate the priority reasoning corresponding to different modes in the subsequent matching reasoning process, so that faults in the explicit mode are given priority, which speeds up fault processing while achieving normal operation of the server as soon as possible. Fourth, when matching the corresponding target reasoning strategy based on the priority of different fault modes and the correlation degree of multiple faults, taking into account multiple faults, we do not directly process them one by one according to the priority of the faults. Instead, we first focus on the correlation degree of multiple faults and whether the corresponding reasoning strategies are conflicting. We then comprehensively consider and match them to obtain the corresponding target reasoning strategy, thereby avoiding delayed processing due to faults with lower priority, and preventing the contradiction of serious faults from causing secondary faults. This further speeds up the repair progress of the faults and improves the reliability of the server.

[0054] Secondly, after matching and determining the target fault pattern, the multiple fault relationships corresponding to the same fault pattern are determined to determine the correlation between the multiple faults. This allows the correlation between the multiple faults to influence subsequent reasoning strategies, avoiding secondary failures caused by conflicting resolution methods. Multiple relationships can comprehensively determine the correlation between multiple faults within a fault, helping to develop more effective fault diagnosis and strategies, and reducing the impact of the faults on the system. Based on the matching of the correlation between multiple faults under different fault patterns, the priority order of the fault patterns can be determined to improve the orderliness of the generation of target reasoning strategies. Predefined SPARQL rules are used to match comprehensive solutions for the cross-cutting of multiple faults within a server from the knowledge graph. Considering scenarios where multiple faults occur on one or multiple servers, the correlation between multiple faults can be comprehensively considered to generate a more comprehensive and efficient target reasoning strategy, avoiding simple processing based on chronological or priority order, and improving the efficiency of subsequent fault resolution.

[0055] In addition, the present invention also provides a server fault processing device and a computer-readable storage medium, which have the same beneficial effects as the above-mentioned server fault processing method. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention, the following is a brief introduction to the drawings required for use in 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 any creative work.

[0057] Figure 1 A flow chart of a server failure handling method provided by the present invention;

[0058] Figure 2 A flow chart of a server fault diagnosis method provided by an embodiment of the present invention;

[0059] Figure 3 A structural diagram of a server fault handling device provided by an embodiment of the present invention;

[0060] Figure 4 A structural diagram of a server fault handling device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] The core of the present invention is to provide a server fault handling method, device and medium to solve the problems of delayed fault repair progress and reduced reliability.

[0063] In order to enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0064] During the deep integration and transformation of full-stack adaptation, high technical barriers between departments and a lack of information sharing mechanisms are common in multiple technical areas, such as architecture optimization, operating system kernel tuning, and middleware interface specifications. The technical standards of the hardware design team and the operating system adaptation group are not fully aligned, resulting in a prolonged verification cycle for the efficiency of underlying resource scheduling. The software compatibility testing department and the quality control team lack real-time collaboration tools, resulting in a delayed feedback loop for issues. The reuse rate of technical solutions across different product lines is low, and similar adaptation cases have not yet formed a standardized knowledge base. This fragmented collaboration model leads to duplicated R&D resources and prone to deviations in the transmission of key technical parameters, seriously affecting the integration efficiency of the entire product and the hardware and software ecosystem. Especially when it comes to the rapid adaptation of multiple technology stacks, the traditional collaboration method that relies on manual communication and document silos is no longer able to support efficient iteration. It is necessary to improve overall R&D efficiency by building a cross-domain collaboration platform and integrating data and workflows.

[0065] Conventional fault handling relies on manual mapping of fault-handling solutions. However, servers involve a wide range of fault types, including those related to hardware, software, operating systems, and user applications. These faults within the server are often isolated and often interconnected, leading to secondary faults and hindering repair progress. The server fault handling method provided by the present invention addresses this technical issue.

[0066] Figure 1 A flow chart of a server failure handling method provided by the present invention, such as Figure 1 As shown, the method includes:

[0067] S11: receiving a fault analysis question; wherein the fault analysis question includes a potential fault analysis question;

[0068] S12: Convert the fault analysis problem into vector information; and input the vector information into the pre-trained model for reasoning analysis;

[0069] S13: In the reasoning analysis of the pre-trained model, the vector information is divided into faults to obtain the corresponding target fault mode and the correlation degree of multiple faults under the target fault mode; the target reasoning strategy corresponding to the pre-trained model is matched according to the priority of the corresponding target fault mode and the correlation degree of multiple faults;

[0070] S14: After completing the reasoning analysis of the pre-trained model, the target reasoning strategy is output as the response information of the pre-trained model.

[0071] Specifically, fault analysis encompasses all server fault presentation issues. It should be noted that this also includes potential fault analysis, which is used to inform the inference and prediction mechanisms of subsequent pre-trained models, allowing for the foreseeing and proactive resolution of faults. It can also encompass incidental fault analysis, which is not a limitation here.

[0072] The scope of the fault analysis problem here can be a fault analysis for a certain architecture of the server, or a comprehensive analysis of multiple faults, or multiple fault analyses for multiple architectures, etc., which is not limited here.

[0073] Converting the fault analysis problem into vector information mainly involves splitting the fault analysis problem to obtain tag information corresponding to at least one tag; and mapping the tag information into vector information.

[0074] Specifically, the segmentation here can be done using word segmentation, dividing the text into tokens, building a vocabulary, and assigning a unique index. The token sequence can be converted into a numerical vector. Regarding the textual presentation of fault analysis problems, multimodal data delivery can be provided, such as descriptive text, images, and text files. In natural language processing (NLP) and artificial intelligence (AI) dialogue systems, a token typically refers to a unit in text, which can be a word, a punctuation mark, a number, or even a subword.

[0075] This embodiment provides the use of multimodal data to increase the scope of reasoning during fault handling and improve the diversity and flexibility of handling methods.

[0076] In step S12, the vector information is input into a pre-trained model for inference analysis. The inference analysis here refers to inferences across different models. That is, a pre-trained model is not limited to a large language model; it can also be based on multiple models within the model. Pre-trained models are a general term. Large language models are language models containing tens of billions (or more) of parameters. These models are trained on large amounts of text data and are designed to understand and generate human language. Examples of large language models include the Generative Pre-trained Transformer 3 (GPT-3), the Generative Pre-trained Transformer 4 (GPT-4), the Pathways Language Model (PaLM), the Galactica model, and the Large Language Model Application (LLaMA).

[0077] The different models of the pre-trained model of the present invention are based on corresponding failure modes. For example, the explicit failure mode corresponds to model A of the pre-trained model, the occasional failure mode corresponds to model B of the pre-trained model, and the potential failure mode corresponds to model C of the pre-trained model.

[0078] In step S13, explicit faults are faults that have actually occurred and been detected. These faults usually lead to performance degradation, functional abnormalities, or even complete failure of the system or equipment. Occasional faults are mainly the occasional shutdown of service applications on the server. Occasional shutdown of service applications means that one or some service applications on the server (such as Web services, database services, application servers, etc.) suddenly stop working or shut down during normal operation, and this shutdown is not caused by active operations of the system administrator, but due to certain abnormal conditions. This phenomenon is usually sporadic, that is, it does not occur frequently, but once it occurs, it may affect business operations. Potential faults are faults that have not actually occurred but may exist. These faults may exist due to design defects, configuration errors, or potential hardware problems, but have not yet been triggered under current conditions.

[0079] Fault classification is performed on the vector information to determine the corresponding target failure mode. This fault classification is performed by directly comparing the current work log. Existing faults are easily visible, as they are caused by the server not functioning properly and can be directly observed through the monitoring system, log records, or user feedback. For occasional faults, it is necessary to check the current work log for markers, such as the number of cycles they have occurred. Potential faults can be identified by simulating or regularly checking the corresponding data indicators in the work log, provided there is no direct fault alarm or normal operation.

[0080] To determine the degree of correlation between multiple faults within a target fault pattern, we can first identify the multiple faults within that target fault pattern. Based on the degree of correlation between the multiple faults, we can then conduct a comprehensive analysis to determine the relationship between the multiple faults, using causal relationships or mutual influences. For example, a server's central processing unit (CPU) overheating may cause frequent system restarts, while the hard drive may experience read and write errors due to high temperatures. A network device interface failure may cause partial traffic loss, while a configuration error may lead to routing disruptions.

[0081] Regarding the relationship between multiple faults under the same target fault mode, it can be a composite relationship, an independent relationship, a cross relationship, a concurrent relationship, a coupling relationship, etc., which is not limited here. The corresponding degree of association is determined based on different multiple relationships. Here, the multiple relationships can be digitized to correspond to different weights, and the corresponding degree of association is determined based on different weights.

[0082] The target reasoning strategy is matched based on the priority of the target failure mode and the degree of correlation between multiple failures. The consideration here is that different failure modes have different priorities: explicit failures > occasional failures > latent failures. Prioritization is based on the severity of the failures present on the server. Failures can also be prioritized based on their importance, but this is not limited here. After determining the corresponding priority based on actual conditions, different failure modes can be addressed in order of priority.

[0083] The target reasoning strategy for matching multiple faults within the same target failure mode comprehensively considers the degree of correlation between the different faults to match a fault handling strategy that avoids conflicts and prevents secondary server failures. This comprehensive matching process corresponds to the different faults and reasoning strategies under different failure modes, so it is considered individually based on the actual situation to enhance the uniqueness of the target reasoning strategy.

[0084] After obtaining the respective target reasoning strategies, they can be output as response information.

[0085] The beneficial effects of the embodiments of the present invention are as follows: first, the server fault analysis problem is converted into vector information, which is then input into a pre-trained model for reasoning analysis. Compared to the conventional technical solution that uses a pre-stored fault mapping relationship for manual search, the present invention uses the reasoning mechanism of the pre-trained model to obtain a target reasoning strategy, achieving accurate fault location and improving the accuracy of fault analysis. Second, the corresponding fault analysis problem in the present invention includes a potential fault analysis problem. The pre-trained model can be used to determine the corresponding fault mode and then match the target reasoning strategy to achieve predictive reasoning for unknown faults, thereby preventing faults from occurring in advance and reducing the delay of fault processing. Third, for multiple faults, the present invention performs fault partitioning on the vector information during the reasoning analysis of the pre-trained model to determine the fault mode and whether it is multiple faults under the same fault mode. At this time, based on the partitioning process, the corresponding target fault mode and the degree of correlation of multiple faults under the model are determined, so as to facilitate priority reasoning corresponding to different modes in the subsequent matching reasoning process, so that faults in the explicit mode are prioritized, accelerating fault processing while achieving normal server operation as soon as possible. Fourth, when matching the corresponding target reasoning strategy based on the priority of different fault modes and the correlation degree of multiple faults, taking into account multiple faults, we do not directly process them one by one according to the priority of the faults. Instead, we first focus on the correlation degree of multiple faults and whether the corresponding reasoning strategies are conflicting. We then comprehensively consider and match them to obtain the corresponding target reasoning strategy, thereby avoiding delayed processing due to faults with lower priority, and preventing the contradiction of serious faults from causing secondary faults. This further speeds up the repair progress of the faults and improves the reliability of the server.

[0086] In some embodiments, the fault mode includes at least explicit faults, occasional faults, and latent faults; fault classification of the vector information to obtain the corresponding target fault mode and the correlation degree of multiple faults under the target fault mode includes:

[0087] Pre-acquire the vector knowledge database and the server's current vulnerability information;

[0088] Matching the vector information, current vulnerability information, and the vector knowledge database to determine the target failure mode of the vector information;

[0089] Determine the number of faults corresponding to the vector information under the same target fault mode;

[0090] In response to an instruction with a fault quantity greater than 1, the degree of correlation of the multiple faults is determined based on multiple relationships of the corresponding faults under the same target fault mode.

[0091] Specifically, an open interface is established based on the manufacturer's existing bug library, and a vector knowledge database is established through information such as hardware sensor data, operating system kernel events, middleware performance indicators and log texts, involved fields, processing solutions, and final results fed back from the historical bug library. The current vulnerability information of the server can be viewed through the work log, and the vector information, current vulnerability information and vector knowledge database are matched and processed to determine the target failure mode.

[0092] Determine the number of faults corresponding to the vector information for the same target fault pattern. If the number of faults is greater than 1, multiple faults exist, and further determination of the degree of correlation is required. This determination of the degree of correlation is based on the multiple relationships between the faults. Within these relationships, corresponding complex processes are weighted using weight coefficients. The highest weight coefficient indicates the highest degree of correlation.

[0093] After the matching provided in this embodiment determines the target fault mode, the multiple fault relationships corresponding to the same fault mode are determined to determine the correlation degree of the multiple faults, so as to realize the influence of the correlation degree under the multiple faults on the subsequent reasoning strategy and avoid the occurrence of secondary faults due to the contradiction between the solutions of the faults.

[0094] In some embodiments, determining the degree of correlation of multiple faults based on multiple relationships of corresponding faults under the same target fault mode includes:

[0095] Obtain fault phenomena, time series, causal relationships, and dependencies of multiple faults;

[0096] Determine the degree of correlation between multiple faults based on fault phenomena, time series, causal relationships, and dependencies.

[0097] Specifically, the fault symptoms are detailed records of each fault, including error messages, log records, system status, etc. Analyze the similarities and differences between multiple fault symptoms. If multiple faults exhibit similar symptoms, they may be related.

[0098] Time series analysis is based on timestamp analysis or time interval analysis. Timestamp analysis determines the order in which faults occur. If one fault occurs before another, there is temporal continuity between the two, suggesting a causal relationship. The time interval between faults is calculated; if the interval is short, it indicates a mutual relationship.

[0099] Causality involves constructing a causal graph, treating each fault as a node and using directed edges to represent the causal relationships between faults. For example, if fault A causes fault B, this can be represented by a directed edge from A to B. Alternatively, a fault tree analysis method can be used to decompose the fault from the top level (system fault) downwards to identify all possible underlying faults (elementary faults). By analyzing the structure of the fault tree, the degree of correlation between faults can be determined.

[0100] Dependencies: If the components involved in multiple failures have dependencies in the system architecture, there may be a correlation. Check the system configuration files to determine if there are any configuration errors or conflicts that may cause multiple failures to occur simultaneously.

[0101] Through the above multiple relationships, the priority weights of the multiple relationships can be set first, and then the weights can be set based on the specific situation of each relationship. Finally, the weighted sum of the weights corresponding to the multiple faults can be calculated to obtain the corresponding correlation degree.

[0102] The above-mentioned multiple relationships provided in this embodiment can comprehensively determine the degree of correlation between multiple faults in a fault, which helps to infer more effective fault diagnosis and strategies, and reduce the impact of faults on the system.

[0103] In some embodiments, matching a target reasoning strategy corresponding to a pre-trained model according to the priority of the corresponding target fault mode and the degree of correlation of multiple faults includes:

[0104] Establish a mapping relationship between failure modes and corresponding reasoning strategies in advance;

[0105] In response to an instruction with one target fault mode, the corresponding target reasoning strategy is matched according to the mapping relationship and the correlation degree of multiple faults;

[0106] In response to an instruction with multiple target fault modes, the corresponding mapping relationship and the corresponding correlation degree of multiple faults are determined one by one according to the priority of the target fault mode to match the corresponding target reasoning strategy.

[0107] Specifically, the mapping between failure modes and corresponding reasoning strategies is pre-defined. For explicit failures, the corresponding reasoning strategy uses the knowledge graph to accurately match remediation plans. For sporadic failures, the corresponding reasoning strategy uses a large model to perform semantic retrieval and analogical reasoning on similar historical cases to generate root cause hypotheses. For latent failures, the corresponding reasoning strategy uses a large model to perform joint probabilistic modeling on multi-source data and output a ranked list of the most confident root causes.

[0108] It should be noted that, based on the fault analysis problem, there can be one or more fault modes involved, and this is not limited here. When there is a single target fault mode, only the mapping relationship corresponding to that fault mode and the target reasoning strategy corresponding to the correlation degree matching of multiple faults under that fault mode are considered. When there are multiple target fault modes, the priorities corresponding to different fault modes can be considered. The reasoning strategy for the correlation degree matching of multiple faults corresponding to the highest priority fault mode can be first performed, followed by the reasoning strategy for the correlation degree matching of multiple faults corresponding to the second priority fault mode, and so on.

[0109] The correlation degree matching of multiple faults under different fault modes provided in this embodiment can be determined by the priority order of the fault modes to improve the orderliness of the generation of the target reasoning strategy.

[0110] In some embodiments, when the target fault mode is an explicit fault, matching the corresponding target reasoning strategy according to the mapping relationship and the degree of correlation of multiple faults includes:

[0111] In response to multiple failures occurring on the same server, a knowledge graph is pre-built for server hardware components, middleware, failure types, symptoms, and handling strategies. The relationships between entities in the knowledge graph are represented in the form of triples using the Resource Description Framework. Regular expressions are used to match the vector information of multiple failures to determine the target problem template.

[0112] Convert the target question template into a corresponding query statement; and determine the corresponding query result based on the query statement and the knowledge graph database;

[0113] Generate the corresponding target reasoning strategy based on the entity relationships in the knowledge graph, the corresponding query results, and the correlation degree of multiple faults;

[0114] Alternatively, in response to an instruction that multiple faults have occurred on multiple servers, a knowledge graph corresponding to server hardware components, middleware, fault types, fault symptoms, and handling strategies is pre-built in parallel to represent the relationships between entities of the knowledge graph in the form of triples of a resource description framework;

[0115] Use regular expressions to match the vector information of multiple faults to determine the target problem template;

[0116] Convert the target question template into a corresponding query statement; and determine the corresponding query result based on the query statement and the knowledge graph database;

[0117] Generate the corresponding target reasoning strategy based on the entity relationships of the knowledge graph and the corresponding query results.

[0118] Specifically, considering that multiple failures may occur in one server or in multiple servers, when there are multiple servers, the failures are processed synchronously through parallel processing of the multiple servers.

[0119] Taking multiple failures occurring on a server as an example, a knowledge graph is constructed that includes server failure information, the association relationship between failures, and the treatment plan. The entities of this knowledge graph can include server hardware components, middleware, failure types, symptoms, treatment methods, etc. The relationship between entities is represented by the triple form of the resource description framework. For example, the entities: server S1, CPU failure, memory failure, disk failure, restart service, replace hardware, etc. Relationship: Server S1 has a CPU failure, server S1 has a memory failure, the treatment method for CPU failure is to replace hardware, and the treatment method for memory failure is to restart the service. Association relationship: CPU failure may cause memory failure, and memory failure may affect disk performance.

[0120] Regular expressions are used to match the vector information to determine which predefined problem template the fault description belongs to. Based on the matching problem template, the fault description is converted into a corresponding query statement. The query results are determined based on the query statement and the knowledge graph database. This involves sending the generated database query language (SPARQL Protocol and RDF Query Language, SPARQL) query statement to the knowledge graph database (such as the Apache Jena Fuseki service) to obtain the query results.

[0121] According to the entity relationships of the knowledge graph, the corresponding query results and the correlation degree of multiple faults, the corresponding target reasoning strategy is generated.

[0122] Regarding multiple failures corresponding to multiple servers, similar to the above embodiment, they only need to be handled in parallel.

[0123] This embodiment uses predefined SPARQL rules to match a comprehensive solution for multiple server faults within a knowledge graph. Considering scenarios where multiple faults occur on one or more servers, it comprehensively considers the relationships between multiple faults to generate a more comprehensive and efficient target reasoning strategy, avoiding simple chronological or priority-based processing and improving the efficiency of subsequent fault resolution.

[0124] In some embodiments, the entity relationships of the knowledge graph, the corresponding query results, and the correlation degree of multiple faults are used to generate a corresponding target reasoning strategy, including:

[0125] In response to mutually exclusive instructions of multiple failures, a corresponding target framework is selected in sequence using the underlying frameworks of the servers corresponding to the multiple failures to determine a corresponding target reasoning strategy based on the target framework and the corresponding query results;

[0126] Alternatively, in response to an instruction that multiple faults are not exclusive, the multiple faults and the corresponding query results are utilized to determine a corresponding target reasoning strategy for parallel processing.

[0127] Specifically, considering the degree of correlation between multiple faults, if the correlation is low or even 0, it is possible that multiple faults are mutually exclusive, and it is necessary to order the underlying framework of the server, such as hardware, software and applications. Therefore, based on the multiple faults, determine in which underlying framework the faults occurred. If fault a occurs in the hardware and fault b occurs in the software, the hardware needs to be replaced.

[0128] For example, if server S1 has both a CPU and memory fault, the reasoning strategy for the CPU fault is to replace the hardware, while the reasoning strategy for the memory fault is to restart the service. However, a CPU fault may also cause a memory fault. Therefore, it is recommended to first replace the hardware to resolve the CPU fault, which may also alleviate the memory fault. If the memory fault persists, restart the service. Based on the target framework and the corresponding query results, the corresponding target reasoning strategy is determined.

[0129] In the case where multiple failures are not exclusive, multiple failures are processed in parallel.

[0130] The multiple fault reasoning strategy provided in this embodiment determines the target reasoning strategy corresponding to the respective query results in both exclusion and non-exclusion situations, thereby avoiding secondary faults caused by exclusion factors and improving the efficiency of fault handling.

[0131] In some embodiments, when the target failure mode is an occasional failure, matching the corresponding target reasoning strategy according to the mapping relationship and the degree of correlation between multiple failures includes:

[0132] Obtain historical case data containing multiple faults; and construct a fault correlation map based on the historical case data;

[0133] Combining multiple faults to generate a first embedding vector;

[0134] determining a similarity between the first embedding vector and the historical fault embedding vector;

[0135] Filter out target historical case data similar to the vector information based on the relationship between similarity and similarity threshold;

[0136] Extract features from the target historical case data to obtain the historical correlation degree of multiple faults;

[0137] Determine the matching degree of the degree of association based on the historical degree of association and the degree of association;

[0138] Determine the corresponding root cause hypothesis based on the degree of match from historical case data;

[0139] The root cause hypothesis is verified to determine the cause corresponding to the root cause hypothesis as the target reasoning strategy.

[0140] Specifically, historical case data containing multiple faults is obtained to construct a fault association graph. Nodes in the graph represent faults, edges represent the associations between faults, and edge weights can indicate the strength of the association (such as the certainty of the causal relationship, the frequency of concurrent relationships, etc.). For example, if fault A is often the direct cause of fault B, and this relationship occurs repeatedly in historical cases, then the edge from A to B in the graph will have a higher weight.

[0141] The combination of multiple faults generates a first embedding vector. The similarity between the first embedding vector and the historical fault embedding vector is determined to screen out target historical case data with similar vector information. Feature extraction is performed on this target historical case data to determine the historical correlation degree of multiple faults, such as fault A leading to fault B, which in turn leads to fault C, thus extracting the causal chain. The matching degree of the correlation degree is calculated based on the historical correlation degree and the correlation degree. This calculation process can use a graph matching algorithm to match the current fault correlation graph with the historical fault correlation graph, and calculate the node matching degree (similarity of fault type and manifestation) and edge matching degree (similarity of correlation relationship).

[0142] The corresponding root cause hypothesis is inferred from historical case data based on the degree of correlation. If the fault combination in the historical case highly matches the current fault combination in terms of correlation, and the root cause of the historical case is known, the root cause of the historical case can be used as the hypothesized root cause of the current problem. For example, if fault A in the historical case was caused by equipment aging, and fault B was caused by fault A, and the current fault combination has a high degree of correlation, then the root cause of the current fault A is assumed to be equipment aging.

[0143] Root cause hypotheses are validated to identify the corresponding cause as the target reasoning strategy. Based on the generated root cause hypothesis, preliminary verification is performed in conjunction with the actual situation of the current system. Equipment usage and maintenance records, as well as the chronological order of failures, are examined to determine if the hypothesis supports it. For example, if the hypothesis is that equipment aging caused Fault A, and Fault B was caused by Fault A, the equipment's age and maintenance records can be checked for signs of aging.

[0144] Verify the reasonableness of the correlation between faults. If fault A is assumed to cause fault B, check whether fault A occurs before fault B and whether the behavior of fault A is consistent with the logic of causing fault B. For example, if fault A is an electrical short circuit and fault B is a motor burnout, check whether the electrical short circuit causes the motor burnout and whether the time sequence of the two occurrences is reasonable.

[0145] This embodiment provides a method that matches the correlation between multiple faults, utilizes a large model for semantic retrieval and analogical reasoning, and generates reasonable root cause hypotheses. This significantly improves the accuracy, efficiency, and comprehensiveness of fault diagnosis, reduces operation and maintenance costs, enhances the system's adaptability, and improves the user experience. It helps operation and maintenance personnel quickly locate problems and implement effective solutions.

[0146] In some embodiments, when the target fault mode is a potential fault, matching the corresponding target reasoning strategy according to the mapping relationship and the correlation degree of multiple faults includes:

[0147] Construct a fault correlation diagram based on the correlation degree of multiple faults;

[0148] Determine fault nodes and edges based on multiple faults; a node represents a fault; an edge represents the association between faults, and the edge weight indicates the degree of association;

[0149] Modeling the fault association graph based on fault nodes and edges to determine the graph neural network model;

[0150] Calculate the abnormal probability of the faulty node through forward propagation;

[0151] Rank the root causes of abnormal probabilities of faulty nodes to identify potential faults;

[0152] The processing strategy corresponding to the potential fault is used as the target reasoning strategy.

[0153] Specifically, a fault correlation graph is constructed based on the correlation degree of multiple faults. Fault nodes and edges with correlation degrees are determined based on the multiple faults. Each node represents a fault or fault type. Edges represent the correlation between faults, and edge weights can represent the strength or probability of the correlation. Fault data and correlation relationships are organized into a graph structure. For example, an adjacency matrix or adjacency list can be used to represent the fault correlation graph.

[0154] During the joint probabilistic encryption process, a suitable model can be selected, such as a Bayesian network, a Markov random field, or a graph neural network. The graph neural network used in this embodiment is combined with a deep learning method, which can automatically learn the complex relationships between faults and perform probabilistic modeling.

[0155] Each fault node is represented as a feature vector, including fault type, frequency, impact range, etc. The edge weight is represented as the correlation strength between faults, which can be the frequency obtained from historical data statistics or the weight annotated by experts.

[0156] Use a graph convolutional network or a graph attention network to model fault correlation. The modeling method here is the same as the conventional method, or it can be different and is not limited here.

[0157] The abnormality probability of each node is calculated through forward propagation, and the root causes are ranked according to the abnormality probability. For example, the fault with the highest probability is considered to be the root cause with the highest confidence, thus determining the potential fault. The handling strategy of the potential fault is used as the target reasoning strategy.

[0158] The prediction process for potential faults corresponding to multiple faults provided in this embodiment facilitates fault analysis from multiple perspectives, improving the comprehensiveness and accuracy of diagnosis. Joint probability modeling allows for consideration of correlations between faults, such as causal relationships and concurrency. This helps identify the common causes of multiple faults and improves the accuracy of root cause analysis. Outputting a ranked list of root causes with the highest confidence levels helps operations and maintenance personnel quickly locate problems and reduce fault response time. This reduces underreporting caused by single indicators or log analysis.

[0159] In some embodiments, when the response information output in response to the instruction does not conform to the instruction during the training process of the pre-trained model, the method further includes:

[0160] Obtain actual response information for historical fault analysis questions;

[0161] The pre-training model is reinforced according to the degree of deviation between the response information and the actual response information to obtain a new pre-training model.

[0162] Specifically, for instructions whose response information does not conform to the requirements, it is necessary to perform enhanced learning processing on the pre-trained model based on the degree of deviation between the response information and the actual response information. The enhanced learning processing here is to fine-tune the model parameters of the pre-trained model to dynamically adjust the model according to actual feedback.

[0163] The deviation can be regarded as numerical deviation, classification deviation or reward deviation. When the response information is close to the actual response information, a positive reward is given; when it deviates from the actual response information, a negative penalty is given to adjust the model parameters to obtain a new pre-trained model for subsequent application to other fault analysis problems.

[0164] The feedback library of the design database (MySQL) stores user error correction data. When the system determines that the result is inconsistent with the actual situation, the incremental learning process is triggered through the comparison module: the LLM is fine-tuned using low-rank adaptation (LoRA) technology, and the knowledge graph edge weights are updated.

[0165] It should be noted that the model training and updating process and the model application process in this embodiment can be in a double-buffered queue to ensure that the business is not aware of the hot update of the knowledge base.

[0166] This embodiment provides that when the answer result is inconsistent with the actual result, the correct information of the question can be fed back through text, allowing the intelligent fault analysis and reasoning system to learn so that the next answer will be more correct, and the answer accuracy rate can be gradually improved through learning.

[0167] Figure 2 A flow chart of a server fault diagnosis method provided by an embodiment of the present invention is as follows: Figure 2 Shown, including:

[0168] S21: User login interface platform;

[0169] S22: Determine whether the operation type is a diagnostic operation. If so, proceed to step S23; if not, proceed to step S24;

[0170] S24: Screening query;

[0171] S25: View historical diagnostic records;

[0172] S26: Determine whether re-fault analysis is required. If yes, proceed to step S23; if no, end.

[0173] S23: Upload fault analysis problem;

[0174] S27: Input the fault analysis problem into the pre-trained model for reasoning analysis;

[0175] S28: In reasoning analysis, the multimodal features of the fault analysis problem are integrated;

[0176] S29: Determine whether it is a known problem based on the fused multimodal features; if so, proceed to step S30; if not, proceed to step S31;

[0177] S30: Automatically match the corresponding target reasoning strategy;

[0178] S31: Generate a diagnostic report;

[0179] S32: Generate visual report;

[0180] S33: The user views the diagnostic report;

[0181] S34: Determine whether the fault analysis problem is solved, if so, proceed to step S35, if not, proceed to step S36;

[0182] S35: Mark the fault analysis issue and close it;

[0183] S36: Submit feedback information to correct the actual response information;

[0184] S37: Perform reinforcement learning processing on the degree of deviation between the response information and the actual response information to update the knowledge graph and optimize the vector model to obtain a new pre-trained model.

[0185] The above describes in detail various embodiments corresponding to the server fault processing method. On this basis, the present invention also discloses a server fault processing device corresponding to the above method. Figure 3 This is a structural diagram of a server fault handling device provided by an embodiment of the present invention. Figure 3 As shown, the server fault handling device includes:

[0186] A receiving module 11 is configured to receive a fault analysis question, wherein the fault analysis question includes a potential fault analysis question;

[0187] Input module 12, used to convert fault analysis problems into vector information; and input the vector information into the pre-trained model for reasoning analysis;

[0188] The analysis module 13 is used to perform fault classification on the vector information in the reasoning analysis of the pre-trained model to obtain the corresponding target fault mode and the correlation degree of multiple faults under the target fault mode; and match the target reasoning strategy corresponding to the pre-trained model according to the priority of the corresponding target fault mode and the correlation degree of multiple faults;

[0189] The output module 14 is used to output the target reasoning strategy as the response information of the pre-training model after completing the reasoning analysis of the pre-training model.

[0190] Since the embodiments of the device part correspond to the above embodiments, the embodiments of the device part please refer to the description of the embodiments of the method part, and will not be repeated here.

[0191] For an introduction to a server fault handling device provided by the present invention, please refer to the above method embodiment, and the present invention will not be repeated here. It has the same beneficial effects as the above server fault handling method.

[0192] Figure 4 A structural diagram of a server fault handling device provided by an embodiment of the present invention, such as Figure 4As shown, the device includes:

[0193] Memory 21, for storing computer programs;

[0194] The processor 22 is configured to implement the steps of the server failure handling method when executing a computer program.

[0195] The server fault handling device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer, or a desktop computer.

[0196] The processor 22 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 22 may be implemented in at least one hardware form: a digital signal processor (DSP), a field-programmable gate array (FPGA), or a programmable logic array. The processor 22 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU; the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 22 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content required to be displayed on the display screen. In some embodiments, the processor 22 may also include an AI processor for processing computing operations related to machine learning.

[0197] The memory 21 may include one or more computer-readable storage media, which may be non-transitory. The memory 21 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 21 is at least used to store the following computer program 211, wherein, after the computer program is loaded and executed by the processor 22, it can implement the relevant steps of the server fault handling method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 21 may also include an operating system 212 and data 213, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 212 may include Windows, Unix, Linux, etc. The data 213 may include but is not limited to data involved in the server fault handling method, etc.

[0198] In some embodiments, the server fault handling device may further include a display screen 23 , an input / output interface 24 , a communication interface 25 , a power supply 26 , and a communication bus 27 .

[0199] Those skilled in the art will understand that Figure 4 The structure shown in the figure does not constitute a limitation to the server fault processing device, and may include more or fewer components than shown in the figure.

[0200] The processor 22 implements the server failure handling method provided by any of the above embodiments by calling the instructions stored in the memory 21.

[0201] For an introduction to a server fault handling device provided by the present invention, please refer to the above method embodiment, and the present invention will not be repeated here. It has the same beneficial effects as the above server fault handling method.

[0202] Furthermore, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor 22, the steps of the above-mentioned server failure handling method are implemented.

[0203] It is understood that if the methods in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0204] For an introduction to a computer-readable storage medium provided by the present invention, please refer to the above method embodiment, which will not be described in detail herein. It has the same beneficial effects as the above server fault handling method.

[0205] Furthermore, the present invention also provides a computer program product, which implements the steps of the server failure handling method when the computer program / instructions are executed by a processor.

[0206] For an introduction to a computer program product provided by the present invention, please refer to the above method embodiment, which will not be described in detail herein. It has the same beneficial effects as the above server fault handling method.

[0207] The above is a detailed introduction to a server fault handling method, device and medium provided by the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the present invention.

[0208] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A server failure handling method, characterized in that: include: Receiving a fault analysis question; wherein the fault analysis question includes a potential fault analysis question; Converting the fault analysis problem into vector information; and inputting the vector information into a pre-trained model for reasoning analysis; In the reasoning analysis of the pre-trained model, the vector information is divided into faults to obtain a corresponding target fault mode and the degree of correlation of multiple faults under the target fault mode; the target reasoning strategy corresponding to the pre-trained model is matched according to the priority of the corresponding target fault mode and the degree of correlation of the multiple faults; After completing the reasoning analysis of the pre-training model, the target reasoning strategy is output as the response information of the pre-training model.

2. The server failure handling method according to claim 1, characterized in that: Failure modes include at least explicit failures, accidental failures and latent failures; Performing fault classification on the vector information to obtain a corresponding target fault mode and a correlation degree of multiple faults under the target fault mode includes: Pre-acquire the vector knowledge database and the server's current vulnerability information; Matching the vector information, the current vulnerability information, and the vector knowledge database to determine a target failure mode of the vector information; determining the number of faults corresponding to the vector information under the same target fault mode; In response to the instruction that the number of faults is greater than 1, the degree of correlation of the multiple faults is determined based on multiple relationships of the corresponding faults under the same target fault mode.

3. The server failure handling method according to claim 2, characterized in that: Determine the degree of correlation of multiple faults based on the multiple relationships of the corresponding faults under the same target fault mode, including: Obtaining the fault phenomena, time series, causal relationships, and dependency relationships of the multiple faults; The correlation degree of the multiple faults is determined based on the fault phenomenon, the time sequence, the causal relationship and the dependency relationship.

4. The server failure handling method according to claim 2, wherein: Matching the target reasoning strategy corresponding to the pre-trained model according to the priority of the corresponding target fault mode and the correlation degree of the multiple faults includes: Establish a mapping relationship between failure modes and corresponding reasoning strategies in advance; In response to an instruction that the target fault mode is one, matching a corresponding target reasoning strategy according to the mapping relationship and the degree of association of the multiple faults; In response to the instruction that the target fault mode is multiple, the corresponding mapping relationship and the corresponding correlation degree of the multiple faults are determined one by one according to the priority of the target fault mode to match the corresponding target reasoning strategy.

5. The server failure handling method according to claim 4, characterized in that: When the target fault mode is an explicit fault, matching a corresponding target reasoning strategy according to the mapping relationship and the correlation degree of the multiple faults includes: In response to the instruction that multiple faults occur on the same server, a knowledge graph corresponding to server hardware components, middleware, fault types, fault symptoms, and handling strategies is pre-built, and the relationships between entities in the knowledge graph are represented in the form of triples of a resource description framework; the vector information of the multiple faults is matched using regular expressions to determine a target problem template; Convert the target question template into a corresponding query statement; and determine a corresponding query result based on the query statement and the knowledge graph database; Generate a corresponding target reasoning strategy based on the entity relationship of the knowledge graph, the corresponding query results and the correlation degree of the multiple faults; Alternatively, in response to an instruction that multiple faults occur on multiple servers, a knowledge graph corresponding to server hardware components, middleware, fault types, fault symptoms, and processing strategies is pre-built in parallel, so as to represent the relationships between entities of the knowledge graph in the form of triples of a resource description framework; Performing matching processing on the vector information of the multiple faults by using regular expressions to determine a target problem template; Convert the target question template into a corresponding query statement; and determine a corresponding query result based on the query statement and the knowledge graph database; Generate a corresponding target reasoning strategy based on the entity relationships of the knowledge graph and the corresponding query results.

6. The server failure handling method according to claim 4, characterized in that: When the target failure mode is an occasional failure, matching a corresponding target reasoning strategy according to the mapping relationship and the degree of association of the multiple failures includes: Acquire historical case data containing multiple faults; and construct a fault correlation map based on the historical case data; Combining the multiple faults to generate a first embedding vector; Determining a similarity between the first embedding vector and a historical fault embedding vector; Filtering target historical case data similar to the vector information based on the relationship between the similarity and the similarity threshold; Extracting features from the target historical case data to obtain historical correlation degrees of multiple faults; determining a matching degree of the association degree based on the historical association degree and the association degree; Determining a corresponding root cause hypothesis from the historical case data according to the matching degree; The root cause hypothesis is verified to determine the cause corresponding to the root cause hypothesis as the target reasoning strategy.

7. The server failure handling method according to claim 4, characterized in that: When the target fault mode is a potential fault, matching a corresponding target reasoning strategy according to the mapping relationship and the correlation degree of the multiple faults includes: Constructing a fault correlation graph based on the correlation degree of the multiple faults; Determine fault nodes and edges based on the multiple faults; wherein the node represents a fault; an edge represents an association relationship between faults, and the weight of the edge represents the degree of association; Modeling the fault association graph according to the fault nodes and edges to determine a graph neural network model; Calculate the abnormal probability of the faulty node through forward propagation; Rank the root causes of the abnormal probabilities of the faulty nodes to determine potential faults; The processing strategy corresponding to the potential fault is used as the target reasoning strategy.

8. The server failure handling method according to claim 1, wherein: When the response information output in response to the training process of the pre-trained model does not conform to the instruction, the method further includes: Obtain actual response information for historical fault analysis questions; The pre-training model is subjected to reinforcement learning according to the degree of deviation between the response information and the actual response information to obtain a new pre-training model.

9. A server fault handling device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the server failure handling method according to any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the server failure handling method according to any one of claims 1 to 8 are implemented.

Images